Cdna06693enc 001

Advanced
Structural Dynamics
Edited by
J. DONA
APPLIED SCIENCE PUBLISHERS
This book presents a review of

modern numerical procedures for structural dynamics which are illustrated by
their application to significant practical
problems. The first part of the text
covers techniques for dynamic response
in the frequency domain as well as
methods for transient response by time
integration. The variational methods of
structural dynamics are presented and
their finite element implementation
discussed in detail. The use of modal
methods for calculating a transient
response is described with a particular
emphasis on the problem of reducing a
multi-degree of freedom system to a
smaller set of degrees of freedom. A
review is then made of implicit, explicit
and combined explicit-implicit operators for transient response by time
integration. This is followed by a
presentation of finite element formulations adapted to implicit and explicit
time integrators.
The second part of the text is devoted
to a detailed presentation of practical
applications. These cover a wide range
of situations including dynamic fracture
mechanics, anti-missile design, pipe
whip accidents and transient fluidstructure interaction. The latter topic is
given particular attention in view of its
growing importance in nuclear reactor
safety studies. Applications are described based on finite difference,
finite element and boundary integral
methods.
(for List of Contents see inside back flap)
ADVANCED STRUCTURAL DYNAMICS
APPLIED SCIENCE PUBLISHERS LTD

RIPPLE ROAD, BARKING, ESSEX, ENGLAND
British Library Cataloguing in Publication Data

Advanced structural dynamics.
1. Structural dynamics
I. Dona, J
624'.) 71
TA654
ISBN 0-85334-859-6
WITH 26 TABLES A N D 205 ILLUSTRATIONS
C ECSC, EEC. EAEC, Brussels and Luxembourg, 1980
Publication arrangements by. Commission of the European Communities, DirectorateGeneral for Scientific and Technical Information and Information Management,
Luxembourg
EUR 6693 EN
LEGAL NOTICE
Neither the Commission of the European Communities nor any person acting on behalf of
the Commission is responsible for the use which might be made of the following information.
All rights reserved. No part of this publication may be reproduced, stored in
a retrieval system, or transmitted in any form or by any means, electronic,
mechanical, photocopying, recording, or otherwise, without the prior
written permission of the publishers. Applied Science Publishers Ltd.
Ripple Road. Barking. Essex. England
Primed in Gre.il Britain b> Gailurd (Primers) Lid. ureal Yarmouth
Preface
This book, is based on the lecture notes prepared for the advanced course on
structural dynamics held at the Joint Research Centre of the Commission of
the European Communities, lspra Establishment, in October 1978.
Although the subject of structural dynamics has a long history, the last
two decades have seen a remarkable development of numerical methods for
dynamic analysis of engineering structures. The intention in organising the
lspra Course was to provide a review of modern numerical procedures for
structural dynamics and to illustrate their use by application to significant
practical problems.
The first part of the book covers techniques for dynamic response in the
frequency domain as well as methods for transient response by time
integration. The variational methods of structural dynamics are presented
and their finite element implementation discussed in detail. The use of
modal methods for calculating a transient response is described with a
particular emphasis on the problem of reducing a multi-degree of freedom
system to a smaller set of degrees of freedom. A review is then made of
implicit, explicit and combined explicit-implicit operators for transient
response by time integration. This is followed by a presentation of finite
element formulations adapted to implicit and explicit time integrators.
The second part of the book is devoted to a detailed presentation of
practical applications. These cover a wide range of situations, including
dynamic fracture mechanics, anti-missile design, pipe whip accidents and
transient fluid-structure interaction. The latter topic is given particular
attention in view of its growing importance in nuclear reactor safety studies.
Applications are described based on finite difference, finite element and
boundary integral methods.
The present book will prove to be useful to researchers and engineers
engaged in the development and use of numerical methods for dynamic
analysis of structures.
JEAN DONA
Contents
Preface .
Lisi oj Contributors
.
.
.
.
.
.
.
.
.
.
ix
Introduction und General Overview oj Structural Dynamics

Problems
.
.
.
.
.
.
.
.
.
xi
J. R E Y N E N
1 Variational Methods of Structural Dynamics and their Finite

Element Implementation
.
.
.
.
.
.
M.
GERA DIN
2 The Modal Method for Transient Response and its A pplication

to Seismic A nalysis
.
.
.
.
.
.
.
43
K . FULLRD
3 Transient Response by Time Integration: Review of Implicit

and Explicit Operators .
.
.
.
.
.
.
S. W. KEY
71
4 Explicit Time Integration of StructureMechanical Systems
97
T. BELYTSCHKO
5 Implicit Finite Element Methods for the Dynamic Transient

Analysis of Solids with Particular Reference to Nonlinear
Situations
.
.
.
.
.
. 1 2 3
D. R. J.
OWEN
Vill
CONTENTS
6 Application of Transient D y n a m i c Numerical

Problems in F r a c t u r e Mechanics
D . R. J.
Methods
to
153
OWEN
7 Survey of C o m p u t a t i o n a l M e t h o d s for Transient Fluid-Structure

Problems in Reactor Safety
.
.
.
. 1 9 1
P. F A S O L I - S T E L L A and
A.
V.
JONES
8 Finite Element Analysis of Transient D y n a m i c F l u i d - S t r u c t u r e

Interaction
.
. - .
.
255
J. D O N E A
9 Coupled F l u i d - S t r u c t u r a l D y n a m i c s in Blowdown Suppression

Systems: Numerical Schemes and Applications
.
.
R.
291
KRIEG
10 C o m p u t a t i o n of Flow-induced Vibrations in Piping Systems

R. J. Gibert
315
11 Analysis of Aircraft Impact Problems

N . J . K.RUTZ1K
337
12 Pipe W h i p Analysis
387
L. L A Z Z E R I
13 Material Behaviour and

Situations
Modelling in Transient
C. A L B E R T I N I . J. P. H A L L E U X and
Index
Dynamic
427
M.
MONTAGNANI
465
List of Contributors
C. ALBERTINI
Commission of the European Communities. Joint Research

Applied Mechanics Division, J 21020 lspra. Italy.
Centre,
T. BELYTSCHKO
Department of Civil and Nuclear Engineering, The Technological

Institute, Northwestern University. Evanslon, Illinois 60201, USA.
J. DONA

Applied Mechanics Division. 1 21020 lspra, Italy.
Centre,
P. FASOLI-STELLA
Commission oj the European Communities. Joint Research

Information Analysis Division, I 21020 lspra. Italy.
K.
Centre,
FULLRD
Central Electricity Generating Board, Research Division,

Nuclear Laboratories, Gloucestershire GLI3 9PB, UK.
Berkeley
M. GERADIN
Laboratoire de Techniques Aronautique s et Spatiales, Universit de

Lige, Rue du Val-Benoit 75, B-4000 Lige, Belgium.
R. J. GIBERT
Commissariat l'Energie Atomique, Division des Etudes Mcaniques

et Thermiques, Centre de Saciar. F-91190 Gif-sur-Yvette, France.
ix
LIST CONTRIBUTORS
J. P. HA LLEUX
Commission of the European Communities, Joint Research

Applied Mechanics Division, 1 21020 lspra, Italy.
Centre,
A. V. JONES
Commission of the European Communities, Joint Research

Information A nalysis Division, I 21020 lspra, Italy.
S. W. KEY
Division 5521, Sandia Laboratories, A lbuquerque, New
USA.
Centre,
Mexico87185,
R. KRIEG
Kernforschungszentrum Karlsruhe GmbH, Institut fr Reaktorent

wicklung, Postjach 3640, D7500 Karlsruhe 1, West Germany.
N. J. KRUTZIK.
Kraftwerk Union A .G.. Berliner Str. 295299, D6050 OJjenbach

(Main), West Germany.
L. LA ZZERI
AMN,
Via P. Pesce, 1 16151 Genova Sampierdarena,
Italy.
M . MONTA GNA Nl

Applied Mechanics Division, 1 21020 lspra, Italy.
Centre.
D. R. J. OWEN
Department oj Civil Engineering, University of Wales, Singleton Park,

Swansea SA 2 8PP, UK
J. REYNEN

Systems A nalysis Division. I 21020 lspra. Italy.
Centre,
Introduction and General Overview of Structural

Dynamics Problems
J. REYNEN
Commission of the European Communities, Joint Research Centre, lspra,

Italy
1.
INTRODUCTION
The advent of digital computers some two decades agoand the development
of numerical methods such as the finite element method (FEM) exploiting
fully the typical features of digital computers (speed and memory) have
caused a revolution in the approach to technical problems, i.e. a shift from
testing towards analysis.
Indeed, until recently the structural engineer has depended on testing to
serve (1) as a proof of design and (2) as a proof of manufacture.
Constant improvement in instrumentation and test equipment increased
the dependence upon testing. However, as the engineer became more
proficient at analysis, he found out that the cost of using analysis for design
certification is far less than that of testing and that the time lapse between
completion of a design and the performance results is far shorter.
Moreover, the engineer is now able to inquire into the behaviour of
situations and locations too remote to be accessible by testing and
instrumentation, of particular importance in the field of postulated
accidents with nuclear reactors.
This triple impact of lower cost, less time and more complete information
has given modern analysis its proper role: analysis is for proof of design;
testing is for proof of manufacture.
In the last decade this revolution has caused a social impact in the sense
that fewer testing personnel are employed and a great demand exists for
digital programmers to make the analysis process more efficient by
developing pre- and post-processors, computer graphics, remote computing techniques, etc. Testing laboratories have been recycled towards the
more fundamental experiments to define physical properties (constitutive
XI
xii
j . REYNEN
relations) with occasional small and clean experiments on semi-finished

products to validate the solutions of mathematical models produced by
computers, although numerical experiments with different meshes and
algorithms seem more attractive for this purpose. The engineer who has to
produce an answer to a structural problemreal, anticipated or more or
less postulated arbitrarilyhas a wide variety of methods at his disposal.
At present, the choice between them depends mostly on 'cost" in the widest
sense of the word. Indeed, not infrequently a directly available in-house
computer program is used which might be less adapted to a specific problem
and at the expense of hours of CPU, i.e. at apparently high cost. However.
in the industrial, administrative and even academic environment where the
answer is needed, 'cost' is not only represented by CPU time but also
includes a great many other aspects related to the lapse of time if a special
algorithm more adapted to the problem has to be developed and
implemented.
Even if on the software marketcommercial or academicsuch an
implementation exists already (and it has been retrieved!), (1) the delay in
raising the eventual budget in order to get the piece of software in house,
including a detailed 'how-to-use', (2) the time needed to get acquainted with
the product and with the difficulties related to the eventual non-portability
of the product, and (3) the psychological aversion to working with a'black
box' are all factors contributing to the continuing use of computer
programs which are not always the most suited for the problem.
Answers might come quickly, but are they reliable?
The process of remedying this situation may be a long one. but is
nevertheless necessary. It is the aim of international gatherings such as the
present lspra Course to contribute to it, by diffusing the present state-of-art
and identifying trends. The proper understanding of the merits of various
numerical approaches and underlying algorithms in relation to the physics
of the problem for which an answer is needed will contribute widely to the
effective portability of software and the effort spent on it. being the final
scope of informatics. The argument against working 'black boxes' is
overcome if the software package becomes a clear and transparent tool of
which the merits and drawbacks and the limits of application are known by
the user, including the reasons for it.
In this course the field of structural dynamics in particular will be
considered, including fluids and their interactions with structures, in view of
applications to design problems and safety issues in the nuclear field, for
both light water reactors (LWR) and liquid metal fast breeder reactors
(LMFBR). The lectures not only deal with the fundamental algorithms and
INTRODUCTION AND GENERAL OVERVIEW
Xlll
their computer implementations, but also include many examples of case

studies.
In the present introductory chapter an attempt is made to classify various
dynamic problems and their solutions and to marry the various detailed
individual contributions together.
2.
CLASSIFICATION OF SOLUTIONS FOR DYNAMIC

PROBLEMS
Until fairly recently, standard university courses in structural mechanics

addressed mainly three classes of topics:
(1) Theory of elasticity. This topic dealt with more or less academic
problems, but nevertheless was of interest for the student as an introduction
to the principles of equilibrium and compatibility. Applications for the
engineer were limited and usually of a qualitative rather than a quantitative
nature.
(2) Strength of materials. This topic represented the bulk of their
theoretical luggage for many generations of engineers. Depending on the
specialisation (civil, mechanical, aeronautical), the courses consisted of
case studies of various mono- and two-dimensional structures, with some
displacement constraint in the 'thickness' direction (frames, plates, vessels,
rotating machinery, etc.).
(3) Vibrations. Applications were limited to idealisations of discrete
masses, springs and dashpots, and to modal analyses of strength-ofmaterial type structures.
Analytical difficulties in dealing with plasticity and creep, even in twodimensional situations, excluded these topics from the standard courses.
In fluid mechanics a similar situation existed: academic solutions for
continuum problems, and the main accent on duct flow.
Nevertheless.it has been withthis limited knowledgethat many of today's
technical achievements have been designed, with which mankind has
learned to live with confidence: civil works, ships, cars, aircraft, nuclear
reactors.
The engineer had to combine physical intuition, limited analytical means
and test results on prototypes to defend his project, sometimes backed up
by design codes (ASME. TU EV, Lloyd, etc.), which, as a matter of fact,
were based on the same limited knowledge but including the experience of
many generations of engineering profession.
XIV
J. REYNEN
At present we are in a completely different situation. For most of the

technical achievements mentioned above, and for the load specifications
and accepted idealisations at the time of their design, there exist now more
or less general-purpose computer programs capable of reproducing
automatically the original stress reports, including a confrontation with
established design codes.
However, as numerical methods performed better, the specifications for
both the load and the idealisations became more stringent. Nowadays the
structural engineer is confronted with problems of which an experimental
verification is not even possible, as, for instance, in the field of postulated
hypothetical accidental situations in the nuclear reactor business, or the reentry problem of a space vehicle, etc.
The advice of the structural analyst is sought in fields other than the
classical technical onesfor instance, bioengineering (prostheses, simulation of the human body in accidental situations, flow problems in veins).
The material with which he has to deal is no longer limited to wood, metal or
concrete but includes plastics, composites, multiphase fluids, bones, human
tissues, etc., at elevated temperatures and/or far beyond the classical limit
of proportionality : Hooke's law is replaced by a set of constitutive relations
relating stress (pressure) to strain (volume change), strain rate and strain
history. His analyses should also include deteriorated structures and the
effects and behaviour of defects (fracture mechanics).
The numerical treatments of these various problems differ widely and not
infrequently the engineer has to use his physical intuition to arrive a priori,
with a heuristic argument, at a conclusion as to which phenomena in a
particular situation are of importance and which can be neglected.
Otherwise, he might end up with an impractical tool which could not be
handled by even the fastest and biggest computer.
Besides qualitative physical intuition, the analyst has at his disposal
published results of numerical experiments by alternative algorithms and
implementations. This source of information, even if not yet confirmed with
an a posteriori theory of convergence, is not always fully recognised by the
software developer, who is inclined to dismiss any empiricism in his work.
However, this course will teach that the experience gained from numerical
experiments is of paramount importance, and often at the base of the latest
and best performing algorithms and of their implementation: the casestudy literature has an important role in diffusing the art outside its
academic place of birth.
The classification of the solution methods for dynamic problems, or
basically the integration of partial differential equations representing the
XV
conservation of mass, momentum and energy, interrelated by constitutive

equations, is focused around integration with respect to space and
integration with respect to time. Analytical integration, in both space and
time, is limited to rather simple geometries and linear problems.
The classification linear versus non-linear, from the point of view of both
material (constitutive equations) and geometry, is the most decisive in
structural problems, and in the past, before the advent of digital computers,
has also been the threshold. This is related to the possibility of application
of the principle of superposition for linear problems. In particular, for
dynamic problems it renders possible the separation of the space and time
variables, the so-called modal approach. In this approach time integration
is carried out analytically for the individual eigenfrequencies and, by proper
superposition of eigenmodes based on the decomposition of the load,
transients can also be dealt with. The spatial integration in the modal
approach can be analytical or numerical and is focused around the
eigenvalue problem of finding the natural frequencies and corresponding
spatial eigenvectors.
Historically, we have to mention here:
Rayleigh-Ritz methods for simple strength-of-material type structures:
the transfer matrix method for chain-type structures;
spatial discretisation by means of FDM (finite difference method) and
FEM.
In this course the spatial discretisation by means of FEM in particular will
be considered, with chapters centred around various aspects such as:
discretisation, stiffness matrix, mass matrix;
automatic condensation of a large system to one with the so-called
master and slave degrees of freedom, substructuring;
extraction of eigenvalues from large matrices (typically up to 200
masters, being the condensed order of some 5000-10 000 degrees of
freedom);
response of damped structures;
applications (anti-earthquake, pipe vibrations, etc.).
The modal analysis is limited to so-called inertia-type transients; for the
wave propagation type of responsein particular, with a steep frontthe
method breaks down and the modal approach has to be replaced by a time
integration.
Linear wave-propagation-type problems can be solved by the method of
XVI
J. REYNEN
characteristics in vogue up to some years agoin particular for fluid

dynamics. This method is not dealt with in the present course.
More recently, for linear continuum problems, a method similar to the
transfer matrix method forchain-type structures has gained popularity: the
boundary integral equation (BIE) method. In this course a case study is
presented using BIE for a fluid-structure interaction problem.
Owing to the prerequisite of linearity in the discussion up to now. the
non-linear convective terms in the Eulerian formulation were excluded and
the above methods, for both structures and fluids, all use the Lagrangian
formulation.
The bulk of the remainder of the present course is centred on the methods
of numerical time integration of ordinary differential equations obtained
by some spatial discretisation, their impact on the spatial formulation
( Lagrangian. Eulerian or a mixture) and discretisation (FDM. FEM) on the
one hand, and on the other hand the impact of constitutive equations
(linear, plasticity, viscoplasticity. compressible, incompressible) and
geometric factors (small versus large displacements) on both time
integration and spatial discretisation.
The argument as to the relative merits of FDM and FEM for spatial
discretisation is not a fundamental one. since linear FEM discretisations on
a regular FDM mesh sometimes prove to give exactly the same equations:
FDM is a special case of FEM. However, the greater versatility of FEM for
dealing with more complicated geometric configurations due to the fact
that the logic of t he computations is embedded in the mesh topology, as well
as the possibility of mixing different kinds of elements, including higherorder ones, has made FEM more popular than FDM. Nevertheless, a great
many computer programs dealing with fluid dynamics still use F D M . As
soon as structures are involved. FDM gives difficulties, and this course will
teach that the more recent developments employ FEM for fluid-structure
interactions.
The choice between mesh formulations is:
Eulerian
Eulerian Lagrangian
pure Lagrangian.
updated Lagrangian.
convective coordinates
The choice is mostly dictated by the physics of the problem. Even if

Lagrangian formulations are preferredbecause of the numerical
difficulties in dealing with the non-linearconvective terms and with material
XVII
interfaces in a Eulerian meshfor problems in fluid mechanics with large

relative movements the Eulerian formulation is necessary, since a
Lagrangian mesh would distort too much. Continuous updating is possible,
as will be shown in the chapter dealing with mixed Eulerian-Lagrangian
meshes, particularly elegant in dealing with slip-flow of fluids around
submerged structures.
For structures with history-dependent constitutive equations, rather
than an equation of state, the Lagrangian formulationpure (for small
displacements), updated or convective (for large displacements)is a must.
Ample case studies are presented in the course to show the relative
advantages and drawbacks in implementation of the various Lagrangian
formulations for various element types, including the more recently
developed isoparametric shell element SEMI LOOF.
The classification of time integrators involves implicit time integrators,
explicit time integrators and mixed implicit-explicit meshes. This topic will
be discussed in detail, including the numerous aspects which define the
choice:
inertia type of loading and response versus wave propagation type;
stability and accuracy of the solutions by means of Fourier-type methods
and energy methods;
linear versus non-linear material behaviour, including the corresponding
formulations of plasticity and viscoplasticity;
linear versus non-linear geometric effects, with discussions on the
relative advantages of the various Lagrangian formulations;
complexity of the finite element and corresponding mass lumping
techniques (simple triangle, isoparametric elements, beam, shell elements,
including the SEMILOOF element);
computer implementation (ease of programming, CPU consumption,
necessary memory requirements, etc.).
In particular, the newer ideas will be discussed concerning combined
implicit-explicit meshes: logics; efficiency of implementation by means of a
frontal-type skyline triangulation solution routine; time-step matching,
etc.
3.
CONCLUSION
An attempt has been made to highlight the role of modern computerised

analysis in the process of design and of the safety assessment of today's
technical achievements.
XVIII
J. REYNEN
Various types of approach to dynamic problems as they will be dealt with

in this lspra course are classified:
for phenomena of linear, vibrational type: the modal approach as
implemented in standard software;
for non-linear and for wave-propagation-type phenomena : research and
trends in time integration algorithms for ordinary differential equations
resulting from a spatial discretisation by means of FDM or FEM.
1
Variational Methods of Structural Dynamics and
their Finite Element Implementation
M. GERADIN
Universit de Lige, Belgium
1.
INTRODUCTION
Certain numerical methods, such as finite differences, apply most naturally

to physical problems presented in differential or local form. By contrast, the
finite element method requires a statement of the problems in integrated or
global form. Finding such global equations is an essential step in applying
the finite element method to a new domain, and to this purpose general
procedures such as weighted residuals or Galerkin methods can be
applied. 2 ^
In the case of structural dynamics, virtual work expressions yield, in
general, adequate formulations. If. as is the case here, we restrict ourselves
to conservative systems, variational principles exist which furnish such
global representation of the problems with additional advantages related to
the existence of extremum principles. Their systematic use is motivated by
the insight into the numerical techniques that they provide. They also allow
for a firmly grounded and clear presentation of the finite element method.
A variational principle is a mathematical expression which states that one
given physical property is approximated in some best sense: for steady state
problems of elasticity, this can be static equilibrium orcompatibility of the
strain field. More essential conditions, such as the symmetry of the stress
tensor and the constitutive equations, are usually assumed to be a priori
verified, but could also be relaxed in the most general context.
When the time dimension is introduced, the dependence in time between
displacements and velocities can be further relaxed, as in Hamilton's theory
of mechanics, this leads to a still wider class of variational principles
applicable to elastodynamics. The present chapter is largely denoted to
their exposition.
Although it provides the basis for the complementary formulation in
1
M. GERADIN
elastodynamics, Toupin's dual formulation for describing the dynamics of

a conservative system of particles is still largely unknown. Section 2 will
thus be denoted to its presentation in the case of a discrete mechanical
system: it will be shown how, starting from Hamilton's principle of least
action, and using Friedrichs transformations, a reciprocal form can be
obtained in which the only variables are the restoring forces in the elastic
restraints of the system.
Taking the Euler derivatives of both variational principles yields the
general form of the discretised equations of structural dynamics: in the
kinematic approach they express dynamic equilibrium in d'Alembert's
sense, while in the complementary approach they express the geometric
compatibility of the system. In the absence of external forces, both
formulations yield to alternative forms of theeigenfrequency problem. The
variational properties of its solutions will be briefly recalled at the end of the
section.
The same path will be followed in Section 3 for obtaining the variational
principles that can be used in elastodynamics: however, more attention will
be focused on the mixed formulations, since some of them are of
considerable practical importance in a finite element context.
Section 4 is an introduction to the finite element implementation of some
of the variational formulations obtained in Section 3. It will be limited to
kinematical elements, equilibrium elements and mixed models based on
simultaneous approximations and velocities.
Section 5 presents some applications of displacement, equilibrium and
mixed models to plate bending and shell problems. In particular, the
concept of dual analysis will be illustrated for the case of a skew cantilever
plate.
An interesting conclusion is that equilibrium may (but does not
necessarily) yield to lower bounds to the eigenspectrum, in which case a
bracketing of the exact eigenfrequencies results from a dual analysis.
The numerical examples presented will also enhance the respective
advantages of the various models that can be imagined on the basis of
distinct variational expressions.
2.
VARIATIONAL THEOREMS FOR DISCRETE ELASTIC

SYSTEMS
2.1. Hamilton's Principle

In order to facilitate the interpretation of the variational formulations of
elastodynamics. let us consider first the case of a conservative, discrete
VARIATIONAL METHODS OF STRUCTURA L DYNA MICS
FIG.
1.
system of particles (Fig. 1). The mass particles m7 are supposed free of
kinematic restraints, and the resulting 3 degrees of freedom system is
assumed to have a potential energy V(r'a) depending only upon the
3N(N l)/2 coordinate differences:
; > y.
/ = 1.2.3: =
(1)
For conciseness, let us adopt a matrix formalism. Equation (1) can be

rewritten
r = Lx
(2)
where L isa Boolean matrix of dimension 3N( l)/2 3V. If theelastic

restrains of Fig. 1 have spring stiffnesses k^,. alternative forms of the
potential energy of the system are
V(r) = hJSr
V(x) = i v ' Kx
and
(3)
with the alternative definitions of the stillness matrix

S = diag(A ;;()
and
K = L'SL
(4)
The kinetic energy of the system is that of the individual particles

= i v ' Mx
(5)
with the mass matrix of the system M = diagfm!,).

For such a conservative system, Hamilton's principle states that for the
actual motion the Lagrangian action is stationary:
(
\') = 0
(6)
M. GERA DIN
over a time interval subject to the boundary conditions

x = 0
at
= ,
and
(7)
Performing the variation of eqn. (6) and integrating by parts in time

yields
c . '
~
.Y
I
"
d cT\
\)
tV
= 0
cx
from which result the 3/V Lagrange equations of motion

d (cT\
CV
(8)
- + = 0
\(/
cx
By referring to the explicit forms of 7~and Fin terms of kinematic variables,
we obtain the dynamic equilibrium equations
Mx + Kx = 0
(9)
describing free vibration motion.

2.2. Principle Operating on Displacements and Velocities
11 is classical to transform the 3 ' second-order equations (9) into two sets
of first-order equations. This can be performed most conveniently by
relaxing into eqn. (6) the time dependence between displacements and
velocities:
[T + p\x
- ) - F]dr = 0
7~is now a function of t he independent velocities v, and is a set of Lagrange

multipliers.
Let us perform the variational derivatives to obtain the Euler equations:
op: restores the constraint =
. shows that the multipliers are kinetic momenta
(10)
: expresses dynamic equilibrium in terms of momenta

CV
ex
(11)
The next step consists of inverting relation (10) in a form

T*
i =
(12)
c
where 7*. the complementary kinetic energy, isa function of the momenta.
This can be achieved through a Legendre transformation
T*(p)=p\
T(v)
(13)
It is obvious that T* has the same numerical value as the kinetic energy T,
but is considered as a function of p:
T* = \j)x M'
With the definition of the extended functional

H(x,p) =/;'.v T* V
(14)
Hamilton's principle can be transformed into
=0
(15)
with the subsidiary conditions ox = 0 for = , and 2 .

Its Euler equations are
=
cT*
c
compatibility in time
(16)
and
cV
= ^
ex
dynamic equilibrium
(17)
They can be made explicit in the form

x = M~1p
and
+ Kx = 0
(18)
2.3. Toupin's Principle5 21

The fully reciprocal form of Hamilton's principle is obtained by relaxing
also dependence (2) between absolute coordinates and the relative
positions r:
' [pTx - V - T* + sT(Lx - r)]dx = 0
(19)
M. GERA DIN
The variational derivatives of (19) are

s: restores eqn. (2)
dp: restores eqn. (16)
V
dr: s =
c/'
(20)
shows that the multipliers s are the restoring forces s = Sr in the

elastic links of the system
ox:pr=Lls
(21)
expresses dynamic equilibrium in terms of momenta and spring

forces
In Toupin's formulation impulses /; are introduced
(22)
// = /!_ + i SOT
in order to integrate eqn. (21) in the form
p = Lfh
(23)
It is next assumed that eqn. (20) can be solved for the kinematic variables /.
which we write
cV*
(h
with the Legendre transformation
V*(h) = rlh V(r)
(25)
where V*. the complementary potential energy, is a function of the impulses

/;. As linearity has been assumed, it has the same numerical value as V(r):
V* =\S~lh=\l?Fh
(26)
The matrix F S is the flexibility matrix of the system.

Let us integrate by parts eqn. (19) and make use of eqn. (25):
[hJx}\] +
[y* T* + xT(LT/i -)]
= 0
The last step of the transformation consists of assuming a priori dynamic
equilibrium, and calculating the complementary kinetic energy in terms of

impulses
77* = \hLM ' O h = N h
(27)
T
with the definition of the mobility matrix ' = LM " 'L .

Toupin's principle implies the stationarity of the complementary
expression to Hamilton's principle
(V* T*)dT = 0
(28)
with the subsidiary conditions oh = 0 for = , and ,.

Its Euler equations are
d/\
7*
at \ ch J
oh
or, in explicit form.

Fli + Nh = 0
(29)
They can easily be seen to be a disguised form of the kinematic relations (2).
Their number is equal to 3/V(/V l)/2: it exceeds the number of degrees
of freedom 37 of the system by the number of redundancies.
2.4. Free Vibration Analysis of the Discrete Elastic System
2.4.1. Kinematic A pproach
As is well known, the general solution of eqn. (9) can be obtained by
superposition of harmonic solutions of the form x(t) = xcos(t + ). The
eigenvalues and eigenvectors of the resulting eigenvalue problem
Kx = orMx
(30)
are denoted by
0 < \ < ; < < .
(31)
We can define the corresponding Rayleigh quotient
R{x)
xlKx
= -
.v Mx
(32)
M. GERA DIN
and by taking variations with respect to ,v, it follows immediately that eqn.
(32) takes the stationary value OJ2 = ? for the corresponding mode .v(/|.
We recall the wellknown orthogonality properties:
x K
and
x Mx
l) \j) = vA j
l) \j) = Vj
The generalised stiffness and mass so introduced. y and ,, are not
independent of each other, since eqn. (32) gives 2 = }'/p. A convenient
choice of norm consists thus of imposing
xJitMx{jj = d,j
so that
xl)K.\tJ) = 2}
(33)
Note that the possible zero frequency modes of eqn. (30). solutions of
Kx = 0, have to be interpreted as either the rigid body modes or the internal
mechanisms of the system, since eqn. (33) shows that they involve no
potential energy.
2.4.2. Equilibrium Approach
Just as in the kinematic analysis, when we set /.(/) = h cos (cot + ), the
Euler equations (29) of Toupin's principle take the form
h = orFh
(34)
Their number, as it is equal to that of the coordinate differences r. exceeds

that of the dynamic equilibrium equations by the number of redundancies
of the system. Thus, as in the kinematic approach, 2 = 0 belongs to the
eigenspectrum of the eigenvalue problem (35).
From eqn. (27) it is obvious that the zero frequency modes, obtained
from Nh = 0, are the non-trivial solutions of
h = 0
(35)
They are selfstressing modesthat is. internal force distributions in

equilibrium without external forces, capable of existing without inertia
forces.
The selfstressing modes will be denoted by sU), and will be supposed to
form an orthogonal basis
sJt)NstJ) = oj
(36)
The other eigenmodes will be denoted by /;(/) and ranged in increasing

order of their eigenvalues
<1> .
2
/;
.2
K)
0 < < \ < < 2
and, in addition to eqn. (36), we have the following orthonormality

properties:
ALto, r , = 0
hlNh{j) = fou
1
V
hJ.Fh^ = J
and
(371
The eigenmodes of eqn. (34) can be obtained as the stationary points of the
Rayleigh quotient in terms of impulses:
= RU,)
I? N h
bJfh
(38)
Note that for a discrete system of particles, solving the free vibration
eigenvalue problem in the form of either eqn. (30) or eqn. (34) provides the
same nonzero eigensolutions: the choice of either set of degrees of freedom
is essentially a matter of convenience. Once the numerical solution has been
obtained, the transformation from stress parameters to kinematic variables
and its reciprocal are
1
, ,
v, = " 1 ! 1 /;,,,
and
/;, =
SLxU)
(39)
CO:
OJ:
2.5. Properties of the Rayleigh Quotient

2.5.1 Independent or MaximumMinimum
Characterisation of
Eigenvalues
All minimum properties of the Rayleigh quotient can be deduced from
Courant's principle, often called the 'minimax principle", which we give here
without demonstration.
To this purpose, let us consider the variational problem of minimising the
Rayleigh quotient under arbitrary constraints:
v ' Kx
min R(x) = -=
x Mx
] = 0
i = 1,..., r - 1
(40)
Courant's principle states that the rth eigenvalue of problem (30) is the
maximum value that can be given to the minimum of the Rayleigh quotient
(40) by varying the r 1 constraints. This maximum is reached for
v, = Mx,
')
min max { R(x) = ~. : ' = 0 , = co]

Mx
j< r
(41 )
10
M. GERAD IN
2.5.2. Recursive Characterisation oj Eigenvalues

As a direct consequence of the minimax principle, we consider the
restricted class of solutions which consists of displacement modes
orthogonal to the first / - 1 eigenvectors. Then
minJ(.v) = '
: xJMx{j)
1 Mx
= 0} = OJ;
j <r
(42)
and this minimum is reached for u = ulr). This wellknown result defines an
adequate procedure for obtaining eigenvalues recursively by minimisation
techniques.
2.5.3. Minimum oj the Rayleigh Quotient Under Constraints
The maximumminimum property of eigenvalues allows also the
prediction of how the eigenspectrum (31) is modified by imposing a set of
independent constraints
Sjv = 0
7=1,..., j
(43)
Rayleigh's theorem on constraints, which we also give without demon

stration, states that if .v arbitrary constraints (43) are imposed on a
vibrating system of which eigenvalues are given by (31). then the new
eigenvalues 2 separate the old ones in the sense that
; < 2 < o j ; + s
r<- s
(44)
Result (44) provides the basis for predicting the convergence behaviour of
Rayleigh-Ritz and finite element approximations in the eigenvalue analysis
of elastic continuous structures.
3.
VARIATIONAL PRINCIPLES FOR SMALL AND LARGE

DISPLACEMENTS IN ELASTODYNAMICS
The variational principles that have been presented in Section 2 for discrete
systems can easily be extended to the elastodynamics of conservative
systems.
The progression from Hamilton's principle to Toupin's principle is
essentially the same: continuity in space and time are relaxed to provide a
canonical principle operating simultaneously on displacements, velocities
and impulses. Restoring a priori requirements between the three sets of
variables provides either mixed or one-field variational equations that will
be made explicit below.
11
One fundamental difference with the discrete case is that significant

benefit can be obtained from the use of twofield variational expressions.
They can lead to various finite element approximations which can exhibit
interesting properties either with regard to convergence behaviour or by
their relative simplicity in comparison with the purely kinematic and
equilibrium models.
An exhaustive list of the formulations that can be used will thus be given,
together with a tentative judgement as to their respective advantages and
drawbacks.
For geometrically nonlinear analysis a total Lagrangian formulation is
adopted : all quantities are referred to the initial configuration of the body,
of volume V0 and surface S0.
The measure of strain adopted is thus Green's strain tensor:
E; .
= i(Z), M; + DjU + D^D^j)
(45)
and hyperelasticity of the material is assumed, so that the state of stress, if

described by the symmetric KirchhoflTrefftz stress tensor, derives from a
strain energy density W(e:) by the constitutive relations
cW
H =
(46)
which are equivalent, in the case of linear elasticity, to the classical

expression of Hooke's law
= Cjut:u
By a Legendre transformation one introduces the complementary stress

energy density function
() = , / W{i})
(47)
such that
e,j = ^
'
(48)
The complementary nature of and W is illustrated in the onedimensional

casein Fig. 2. For linear elastic materials and IF'are quadratic Junctionals
and are thus numerically equal.
3.1. Hamilton's Principle
Hamilton's principle, or displacement variational principle, states, as in
12
M. GERADIN
,">J
. '::'
. d A ( f f ) ; ;'*.
dw() ,
" vd/'
w{,)
V
Fie. 2.
the discrete case, that for fixedend values of the displacement field, the
Lagrangian action of a conservative system
iff] =
( 7)dr
(49)
takes a stationary value on the trajectory of the motion. is the kinetic

energy of the continuum
T(u)
(50)
Pouiu,dvo
and t/, its potential energy, can be split into distinct parts: the strain energy
< =
W(El})VQ
W(DtuJ)dVa
(51
results from the integration of the strain energy W(Djj) per unit of volume
of reference over the domain V0. The notation W(Du) indicates that the
strain energy density isa function of the displacement derivatives only. The
region I 0 is that occupied by the body in its initial state. If large
displacements are considered, the strain energy density of the hyperelastic
material can be expressed as a function of the Green strain tensor (45) with
which the KirchhoffTrefftz tensor is associated by the constitutive
equations (46).
The second contribution to the potential energy, U2. results from
conservative body loads and surface tractions on the part Sn of the
boundary
I
Ujtj(T)dS
UjXj(T)dV
(52)
13
On the remaining part. S, of the boundary the displacements are

prescribed functions of time:
iij = J(T)
on S at any time
(53)
The variational derivatives of Hamilton's principle express equilibrium in

the body and on the surface S i n a dynamic sense: the dynamic equilibrium
equations are obtained in the form given by Signorini:
A K + aimDmUj) + Xj pit, = 0
"M,j + <rimDmUj) + lj = 0
3.2. The Canonical Principle
One notices that Hamilton's
requirements.
(1)
principle
contains
two
priori
Compatibility in space: the strain field deduced from u by eqn. (45)

has to be integratable. and on the part of the boundary where
displacements are imposed,
u, = (7,
(2)
in V0
on S
on Su
(55)
Compatibility in time: the time dependence between displacements

and velocities
t, = u
in V
(56)
is automatically assumed.
The canonical principle is one in which conditions (45). (55) and (56) are
incorporated. To construct it from Hamilton's principle by the method of
Friedrichs transformations, one proceeds in three steps:
(1)
(2)
(3)
The continuity requirements (45), (55) and (56) are relaxed by

adding to eqn. (49) the appropriate dislocation potentials with a
Lagrangian multiplier.
The meaning of the Lagrangian multipliers utilised in (1) is restored
by using the appropriate variational derivatives of the functional.
To eliminate the strain variables in profit of stresses, the
complementary energy density (,,) is introduced by the contact
transformation (47), the derivation of which with respect to stresses
yields the constitutive equations in the inverse form (48).
The resulting variational principle operates on displacements, stresses and

velocities:
[u. , ] = \' F[u, , ] = 0
14
M. G E R A D I N
with the functional

[() \au(Diuj
F[u, , ]
+ Dju, +
DtumDjUm)
+ Xu \pvv + pvu] d V0
tj(.tijj)dS
i,u. dS
(57)
where the t are the Lagrangian surface tractions on the undeformed

boundary:
tj = ni(aij + o,lDmuj)
(58)
Its Euler equations provide the full set of equations of structural dynamics:
(i)
the variation du yields the dynamic equilibrium equations in terms

of displacements and velocities:
A K + <rJ>mu) + Xi Pij =
"Mij + trimDmUj) + tj = 0
(ii)
o
on Sa
(59)
Varying the stresses j and the surface tractions tj restores the

compatibility in space:
c
(iii)
in
\(DliiJ + Dju, + D,umDjum) = 0
in V0
u. = .
on S
(60)
By the variation 7 one restores the dependence (56) between

displacements and velocities.
The functional (57) is particularised to geometrically linear problems

simply by noticing that, for infinitesimal displacements.
\aij(Diuj
and
+ Djui + DwmDjU) =s auD,Uj
tj = n(aj + <fmDmUj) naj
The only a priori conditions that remain in the canonical principle concern:
(i)
(ii)
The constitutive equations (46).

The symmetry of the KirchhoffTrefftz tensor atj = . which
implies that rotational equilibrium is satisfied everywhere. A still
more general principle could be derived by relaxing this condition,
as will be briefly mentioned in the next section.
V A R I A T I O N A L METHODS OF S T R U C T U R A L
15
DYNA MICS
The continuity requirements on the three fields are the following'.

displacements u
velocities v
stresses ,,
C0 continuity
no continuity
C 0 continuity
3.3. Twofield Variational Principles

Twofield variational principles are obtained by specialising the
canonical principle to the cases where one set of Euler's equations are taken
as essential conditions. The remaining two fields are left independent.
3.3.1. Principle Operating on Displacements and Stresses
The principle operating on displacements and stresses is obtained by
restoring the dependence (56) between displacements and velocities. It can
be written
[,]
F[u. ] dr = 0
with the twofield functional

[(/(,,) , / " , + Dju + DumDjuJ+
F[u. ] =
Xu +
y,u]dV
Vu
t;U;dS
/;(, H;)dS
(61)
JSU
It was formulated independently by Fraeijs de Veubeke 6 and Reissner 3 for

static analysis of geometrically linear problems.
3.3.2. Principle Operating on Displacements and Velocities (Hughes)
This corresponds to the case where compatibility in space is rendered
essential. It is obtained most rapidly from Hamilton's principle by relaxing
the dependence between displacements and velocities. The functional of
this second twofield variational theorem of elastodynamics 14
&[u,v] =
F[u, ] dx = 0
takes the form

[ W(D,Uj) + Xu \pvV + -,
F[u, v]
J ,,
t:U: dS
(62)
16
M. GERA DIN
As it allows for independent assumptions on displacements and velocities, it

can be used for reducing in a consisten t way the number of parameters in the
representation of the kinetic energy.
3.3.3. Principle Operating on Stresses and Velocities (Reissner)
The third twofield variational principle corresponds to the case where
dynamic equilibrium (59) is an essential condition. A s shown by Fraejis de
Veubeke.' ' the existence of an explicit form of the functional F[v. ) is,
however, limited to the geometrically linear case.
The functional of Reissner's principle, 1,2 valid only, for geometrically
linear structures, is
F[r,<xj=
[^^-^,-
tdS
(63)
The formal independence of the stress and velocity fields in the associated
variational principle is, however, seriously restricted in practice by the
requirement of satisfying a priori dynamic equilibrium. As one has to
consider infinitesimal displacements, the coupling between the two fields is
reintroduced by
Dau + Xj pij = 0
in F
(64)
One way of generalising the functional (63) to finite elastic displacements,

based on the polar decomposition of the Jacobian matrix of the
transformation, was suggested by Fraeijs de Veubeke.' ' The correspond
ing variational principle operates then on three fields: the Piola stress
tensor, the velocities Vj and the matrix of finite rotations associated with the
deformation. The resulting functional will not be made explicit here.
3.4. Complementary Energy Principle (Toupin)
Like Reissner's principle, the complementary energy principle is limited
to linear elastodynamics.
The requirement of satisfying the dynamic equilibrium equations (64)
suggests the introduction of an impulse field 0,75 such that
Vj^Dfiu
Hence, in the absence of body forces,

eu = ['>,,] +
and
(65)
u =
(66)
17
The resulting functional is
F() =
0)
YpD,PJm~\ dV
tjjdS
(67)
s.
with the modified expression of the surface tractions.
Euler's equations of this principle are obtained by variation of the
impulses. They appear to be a disguised form of the time derivatives of the
compatibility equations of linear elasticity:
dr\cj;)
Tp (D'DJ-
+ D D
' "'0""] =
in v
dii
(68)
3.5. Special-purpose Two-field Variational Principles

Special forms of the twofield variational principles are useful in the finite
element context to derive the socalled hybrid elements. The motivation for
using these principles is always the desire to make independent assumptions
inside the domain constituted by a finite element and along its boundary for
the choice of connection modes.
Such principles have been introduced by Pian 7 and Pian and Tong, 8 and
play an important role in the justification of elements based on engineering
intuition but violating the rules of the general principles.
The first of these principles allows for assumptions on the displacements
inside the boundary and along the boundary S that are independent of the
assumptions on the Lagrangian surface tractions along Su, as recalled by
the notation
^lu,/(SJ]
F[u, t(SJ] = 0
The corresponding functional is

F[uJ(SJ)
%pu,u
W(D,uJ) + Xtu,]dV
t.ii: dS +
1,(11, 7,)dS
(69)
The complementary functional obviously exists in the geometrically

linear case. One then assumes that dynamic equilibrium is a priori satisfied
18
M. GERA DIN
t'T
Su
7(/,)=0
87(</,=0
83"(i-,<r)=0
prion
= /
i/ = 7(5/)
prion
prion
-
Scr
1r
7+->
- /~(5)
"
'Sif
8?[/,/(,)] = 0
o
8?[,i/(5,)]=0
prion
-
c - Du
'8
s,
'
prion
-
fc-iX-p'-O
-"
?(/) = 0
?() = 0
prion
prion
Du
t - Du
compatibility equations
D\A- X-py-0
equilibrium
FIG. 3. Relationship between the variational principles.

only in the volume. The resulting principle operates on the impulses 0, in
the volume and on the displacements on Sa
Jf[0,u(Sa)] =
F[e,u(Sa))dr
=0
with the functional

F0,u(a) =
(,)--,,^ d F +
Uj(tj n&jjdS
(70)
The filiation of the various principles is schematised in the flow chart of

Fig. 3.
4. FINITE ELEMENT IMPLEMENTA TION OF THE
VARIATIONAL PRINCIPLES OF ELA STODYNA MICS
We shall limit ourselves here to a rather brief description of the finite
element approximations that can be based on the variational principles
established above.
First, it is worth while indicating the respective properties and limitations
VARIATIONAL METHODS OF STRUCTURAL DYNAMICS
19
of the various formulations that have been given, according to the use that
can be made of them. These are summarised in Table I.
The variational principle of widest use, on which the kinematic approach
is based, is of course Hamilton's principle. No limitations are encountered
in its practical implementation. For eigenvalue analysis in the linear range,
a direct consequence of Courant's minimax principle is that the kinematic
approach using purely conforming models guarantees obtaining upper
bounds to the eigenfrequency spectrum, but the price paid for this
behaviour is an excessive cost in the representation of kinetic inertia.
Up to now, the canonical and F[u, ] principles have always been
considered of no practical interest from the point of view of finite element
discretisation. It seems, however, that this assertion ought to be
reconsidered for non-linear analysis. The observation that the com
plementary energy remains a quadratic form of the stresses for
geometrically non-linear problems indicates that a mixed discretisation
could perhaps provide a more adequate approach than a purely kinematic
modelling.
The F[u, v] functional has been regarded independently by Hughes 1 4 and
Geradin, Sanderand Nyssen 16 as a consistent way of lumping masses. The
results obtained with it in eigenvalue analysis are quite promising, as will be
shown in numerical examples. However, the [u, v] approach does not give a
systematic convergence to the eigenspectrum by either upper or lower
bounds. Its main drawback seems to be the practical impossibility of
implementing it together with explicit time integration schemes in non
linear problems.
A few observations can be made in common on the variational principles
based on the complementary energy (Reissner, Toupin and Pian): they
cannot be implemented for geometrically non-linear problems; they do not
allow for an easy handling of other body loads than inertia loads; according
to the Euler equations (68), the actual impulse field contains two sets of
impulse (or stress) modes, just as in the discrete case: an infinite set of selfstressing modes, solutions of >,CT;J = 0, and particular solutions of the nonhomogeneous equation (64).
The first two observations indicate that the practical application of
complementary energy principles in structural dynamics is limited to
eigenvalue analysis. The last has two consequences:
(1)
For two- and three-dimensional elastic structures, the number of

self-stressing modes included in a finite element model is generally
much larger than the number of stress modes that equilibrate the
body loads. This gives a significant reduction of the number of
SJ
o
TABLE I
APPLICABILITY OF VARIATIONAL PRINCIPLES IN ELASTODYNAMICS
Variational
principle
Functional
Eider equations
Non-linear
elasticity
Reduction
of inertia
forces
Practical
interest
Other
limitations
F[u. , ]
compatibility in space
compatibility in time
equilibrium
yes
yes
Eraeijs
de Veubeke
F[u, a)
equilibrium
yes
no
Reissner
F{v,a\
no
yes
'1
body loads
rr.
7=
Hughes
F[u,v]
equilibrium
yes
yes
yes
non linear
explicit
schemes
Canonical
c.
Hamilton
F[u]
equilibrium
yes
no
yes
Toupin
Flo]
no
yes
yes
body loads
'And-Pian'
/[. i(SJ]
equilibrium in V
compatibility on S
yes
no
Pian
f[ff.w(S.)]
compatibility in V
equilibrium on ,
no
yes
yes
body loads
>
(2)
21
parameters in the representation of the kinetic energy, just as with

the F[u, v] functional.
The convergence properties of equilibrium models are strongly
influenced by the balance between self-stressing modes and other
stress modes, as can be shown from Courant's minimax principle.
Increasing the number of self-stressing modes raises the calculated
eigenfrequencies. and injecting more stress modes equilibrating the
body loads has the converse effect. The smaller the ratio between
the number of particular stress modes and that of self-stressing
modes, the larger the chance of getting convergence behaviour to
the eigenvalues by lower bounds. 1 2
In what follows a short description is given of the implementation of the

finite element method for the functionals F[u], Flu, v] and F(a), limited to
linear elastodynamics.
For the sake of simplicity let us represent displacements, velocities,
surface tractions, strains, stresses and impulses as row vectors:
wT = (u1u2ui);
T
vT = (,, 2 , 3 ); tT =
:
ffl
K = (,, 2 2 332 3 23);
(tlt2ti)
= ( 2 2 3 3,
3^23)
( ) = ( 0 , ,0^0330,20,3^23)
Then, with the help of a matrix differential operator

~Dl
/,
D2
D2
D3
>,
.0
Z>3 D2
Dl
and a corresponding matrix of direction cosines for the outward normal

, 0
n3
2'
= 0
n2
/I3
n2
/;,
the basic equations of linear elasticity of Section 3 take the form

= Du
DV + X
strain-displacement relations
dynamic equilibrium
p = 0
/=
surface equilibrium equations
constitutive relations
22
M. G E R A D I N
Following these notations, thefunctionals(49). (57) and (67) can be written

in matrix form:
FM =
liP W(Du) + X\t]dV
'TitdSG
F[u,v] I [ W(Du) + XJu ipuTi) + pvy]dV
TudSa
(71)
(72)
Sa
FIO]
[(0) {(Z) T 0) T (Z) T 0)]dF4
t'tdS.,
(73)
4.1. The Kinematic A pproach

The displacement field within an element e of volume Vc can, in general
be discretised using an approximation of the form
(74)
, = Q(x)qt
where Q(x) is a matrix collecting the polynomial shape functions of the

element. The generalised displacements of the element, qc. are related to the
structural generalised displacements q through a localisation operation
(75)
'L = c<?
Operation (75) involves the continuity of the displacement field throughout

the structure, provided that the local displacements q of the elements have
been chosen so as to preserve the continuity of shape functions between
elements. From eqn. (74) one can derive the strains within the element:
Bqc
with
B = DQ(x)
(76)
B is thus a matrix of strain functions, and the rigid body modes of the
element are the solutions of Du = 0.
Substituting approximations (74) and (76) into Hamilton's functional
yields the discrete variational equation
\\Mqt \SjKqc
+ Wg e
=0
(77)
with the definitions of the mass and stiffness matrices of the elements
Mc== I pQ ' d l -
and
K =
BlHBdV
(78)
23
Both matrices are symmetric, and the mass matrix is positive definite. The
positive semidefinite character of the stiffness matrix results from the
existence of rigid body, modes. The generalised loads of the elements,
defined by
QJtdS
. s\
include the external loads applied to the structure and the reactions on the
boundary of the element.
The assembling of the finite element model results from the substitution
of eqn. (75) in eqn. (77): one obtains
lifMq
\qJKq + qTg]dz = 0
(79)
with the structural matrices

' = YzJ# e L e ; M = YLJMCLC
c
(80)
and the structural loads to which only external loads contribute
l.
(81)
since interelement reactions cancel each other after assembling.

The Euler equations of (79) with Hamilton's rule bq = 0 for = , and
= 2 take a form similar to that obtained for the discrete system:
Kq + M q = g(z)
(82)
The integration in time of the fundamental equation (82) of structural

dynamics will be discussed in subsequent chapters, and will not be
considered here.
Under the assumption of no external load, one is led back to the problem
of solving the generalised eigenvalue problem
Kx =
'Mx
(83)
K and M are both symmetric and positive definite if the structure is at least
isostatically supported and free of mechanisms. Otherwise, the rigid body
and kinematic modes are the solutions of Kx 0.
24
M. GERADIN
According to eqn. (78), the mass matrix constructed on the basis of the
shape functions of the elements is consistent with the stiffness matrix
obtained from the same approximation. In contrast with the discrete system
considered in Section 1, the resulting mass matrixwhich is generally
referred to as a consistent mass matrixis a full matrix. In comparison with
the discrete case, a significant amount of computational effort is thus
required to evaluate the inertia forces in eqn. (82) and to solve the
eigenvalue problem (83).
This observation is the reason for a fairly common practice, which
consists of making the approximation that the mass of the structure is
concentrated at the nodes of the finite element model. Such an
approximation, which is then inconsistent with the repartition of stiffness in
the structure, leads, as in the discrete case, to a mass matrix of diagonal
form. It is then referred to as a lumped mass matrix.
When a lumped mass matrix is used, the guarantee of obtaining upper
bounds to the eigenfrequencies of the structure is lost: in some particular
cases, it is possible to show that the eigenfrequencies will be underestimated, 22 but no proof of it can be given in general.
Obtaining cheaper, but still consistent, representation of the kinetic
energy is one of the goals that have been aimed at in the development of
finite element approximations based on other variational equations.
4.2. Simultaneous Discretisation of Displacements and Velocities
If the mixed functional (72) is used, simultaneous finite element
approximations can be made on displacements and velocities.
The requirements on the displacement field are the same as in kinematic
models: the admissible approximations to the displacement field are
continuous on Fand have to satisfy a priori the kinematic conditions u = 7
on S.
The requirements on the velocity field are less restrictive, since continuity
is not required from a variational point of view. In a finite element
approximation based on eqn. (72), the discretisation of velocities can be
limited to piecewise continuous functions, the associated degrees of
freedom remaining unconnected at the structural level. Let us discretise the
displacement field by eqn. (74), and use a similar approximation for the
velocities
t, = F(.v) Pc
(84)
the content of F(.v) being generally limited to the rigid body modes of the
element. As the velocity parameters remain independent at the structural
25
level, the elementary parameters are simply juxtaposed to give the

structural set p.
Substituting approximations (74) and (84) into eqn. (72) yields the
discrete variational equation
WKq + i2jTtMtP< qTYA jpt q1 gdr
(85)
= 0
with the definitions of the stiffness, mass and coupling matrices of the
element
BTHBdV;
K,=
pVrVdV: A e=
Mc=
Jt c
Ji\
pV1 M dV
(86)
Jv,
and the structural stiffness matrix

K
=Yw.
The Euler equations of eqn. (85) with Hamilton's rule q = Ofor = , and
= , are
q: Kq +
: Mj)c A^q^ = 0
(87a)
g(t)
for each element
(87b)
The former equation is the discretised form of the equilibrium, eqn. (59),
and is expressed at the structural level. Equation (87b) relates the velocity
parameters of the element to its generalised displacements. The system of
equations (87a, b) can be further simplified by noting that the element mass
matrix is nonsingular and, thus, decomposable in triangular form. The
successive operations
Mc=
c/TC/,;
Bt = U;JA C
(88)
together with the definition of a new set of parameters

>\ = Uj)e
(89)
yield the simpler set of equations

Kq + B'r=g(T)
(90a)
r Bq = 0
(90b)
26
M. GERA DIN
with the structural matrices rT = [r],..., rj] and BJ = [/{B,,

([&[}'
which, in practice, will never be assembled. The substitution of eqn. (90b)
into eqn. (90a) would reduce the firstorder system above to the classical
equilibrium equation (82) with the structural mass matrix expressed in
terms of generalised displacements:
M = \
(BlBJ,
(91)
In this case we would be led back to the classical methods for eigenvalue
analysis and direct time integration in terms of generalised displacements.
However, greater computational efficiency can be obtained if the
parameters r are used explicitly in the computational scheme to express the
inertia forces on the structure.
If free harmonic motion is assumed by setting g(z) = 0.
q = qs\r\ojz
and
r = rcoscoz
Equations (90) reduce then to the linear eigenvalue problem

Kq coBTr = 0]
r coBq = 0 I
(92)
As the form (92) of the eigenvalue problem is unclassical, its properties and
the way of solving it efficiently will be briefly discussed.
Eigenso/utions. Let us denote by , an eigenfrequency of system (92).
The corresponding eigenmode can be expressed in terms of either
displacement or velocity parameters. We note thus it is (q0). rU)). It is easy to
verify that between two eigenmodes of distinct eigenfrequencies , and Wj
the following orthogonality relations hold:
qlMu)
i'l'J = ,2<5,;
rJ^Bq^ = ,<5,;
and also
ql)B*Bqm
= lj;
r^BK^B1,^
= <o*y
Rayleigh quotient. The Rayleigh quotient can be written in terms of either

displacements or impulse parameters:
qrKq
q Bq
rJBK'lBr
r r
The comparison of both definitions (93) shows that solving the eigenvalue
27
problem (92) in terms of velocity parameters is more rewarding for three

reasons:
(1)
(2)
(3)
The eigenvalue extraction is reduced to that of the symmetric

matrix
'.
The number of velocity parameters r can be kept smaller than the
number of generalised displacements, without alteration of the
stiffness properties of the model.
If < nq, the mass matrix = used in the displacement
method of solution is singular.
Vector iteration scheme. If the velocity parameters are used as variables,

the various available methods of eigenvalue extraction based on power
iterations (ordinary power method, inverse iteration, simultaneous
iteration, Lanczos method, etc.) can be implemented using the following
iteration scheme:
initialisation r0 arbitrary starting vector
solve Kqi+ , = BTr
iteration
[compute r i + 1 = Bqi+X
The applications of mixed models to plate vibration problems presented in
Section 5 have been achieved using the Lanczos method of minimal
iterations. The algorithm is particularly economical and easy to implement.
Its main drawback is a rapid loss of orthogonality due to progressive
cancellation of the starting vector, which can be prevented by an
appropriate reorthogonalisation procedure. Details of the implementation
of the algorithm can be found in references 23 and 24.
4.3. The Equilibrium Approach
The discretisation of the impulse field in one fini te element domain can be
separated into two parts:
0(x, ) = R(x)c(z) + S(x)s(z)
(94)
In S(x), each column is a set of stresses in equilibrium without body loads:

DJS(x) = 0
In R(x), each column is a distribution of stresses that requires a nonzero
distribution of body loads:
DJR(x) = v(x)
In general, R(x) and S(x) together constitute a complete representation of
28
M. GERA DIN
(, ) in terms of polynomials up to a given degree, and R(x) contains at

least the inertia loading due to rigid body motions of the element. The
intensities c(z) and s(z) of the stress distributions are the unknown time
functions, to be determined so that deformation compatibility be satisfied
in some best sense, using the functional (73).
Let us assume that all the boundary of the element e pertains to Su. Then
assumption (94) generates on the boundary a set of independent boundary
traction modes T(x):
te(x) = NTR(x)c
+ NrS(x)xc
= T(x)gc(z)
(95)
The associated parameters g(z) are the generalised loads of the element: in
contrast to the kinematic approach, they give considerable information
about the surface traction distribution. Each time a choice is made for the
measure of the intensity of a surface traction mode T(x), this measure
becomes related to the values of c and s. Consequently, we can write
gAz) = Gce + Hs< = [GH\K
(96)
with the load connection matrices of the element G and H and the set of
parameters b] = [CJJJ].
The potential energy associated with prescribed boundary displacements
is subject to a discretisation coherent with eqn. (95):
J.t,.dS =
jT(x)ge(x)S
= qjgc
(97)
using the definition of the boundary displacements
qc= i r(x)t.dS
(98)
which provides only little knowledge of the displacement field on the

boundary. It will be observed that both with considerable and little
knowledge the role of displacements and forces is reversed as compared
with the kinematic models.
We are now in a position to discretise Toupin's functional
\' iY^FîYcjN^
e
+ qJg, dr = 0
with the positive definite flexibility matrix of the element

Fe =
~F
F.,
F~
(99)
29
defined by
Fec=
RJ(x)HR(x)dV
Fcs=
RT(x)HS(x)dV
Fss=
ST(x)HS(x)dV
(100)
and the 'mobility' or inverse mass matrix

A'
1
vcP
(D'R)'(D'R)dV
(;
At the element level, the variational derivatives of eqn. (99) together with
bc = 0 at time limits yields the discretised form of the compatibility
equations:
FJ + FJ + Nc - Gq = 0
Fscc + Fsss - Hq = 0
(102)
It is fundamental to notice that the use of unknown generalised

displacements for assembling finite elements at the structural level is a
convenient procedure that is also applicable to equilibrium elements. The
only difference is that the generalised displacements used are 'weak'
unknowns, and are essentially of interface type.
To transform the system of equations (102) into a stiffness equation, one
notes that its terms have the meaning of velocities and that the change of
variables
-N.C
(103)
therefore introduces displacements conjugated to the impulse parameters c.

Equations (102) are then brought into the form
= C7
;i04)
with the extended load connection matrix
c-
G H
'
(105)
30
M. G E R A D I N
Equation (104) is then solved for c and s and put into the form
CF-'C1
(106)
Using eqn. (96), the left-hand side of eqn. (106) is also

G
1
11
0
and, consequently, by partitioning the stiffness matrix into the form

K =
K,
we obtain the set of dynamic equations

K
+ KqPP = g
Kpqq+Kqpp
= -'1
w
(107)
The final stiffness equation results from the definition of the full sets of
displacements and loads of the element
OJc =
'q
1\
j>_
i'.. =
c
g~
_0_
and that of the extended mass matrix

M.. =
"0
_o N: '
whereby one obtains at the element level

Kw, + M,w, = r,
(108)
Expressing equilibrium of the boundary tractions along element interfaces

leads to structural equations of the same form.
The structural eigenvalue problem can be written in the standard form
(K - co2M)w = 0
(109)
The main difference with the kinematic approach lies in the fact that the
mass matrix has a degree of singularity equal to the number of boundary
displacements q, since only the internal parameters c contribute to the
kinetic energy. Hence, the boundary displacements q can be eliminated
statically from the eigenvalue problem,
q=
-KMlKqpp
V A R I A T I O N A L M E T H O D S OF S T R U C T U R A L
DYNA MICS
31
and the condensed eigenvalue problem keeps the standard form

(K%, 2)
= 0
(110)
l
with the condensed stiffness matrix K*p = Kpp KpqK~q .

It can thus be solved with the same methods of extracting eigenvalues as
in the kinematic approach.
The interesting property of the equilibrium approach is the reduction of
the number of dynamic degrees of freedom (that is, the degrees of freedom
kept in the eigenvalue problem) to which it leads naturally.
5.
NUMERICA L A PPLICA TIONS
5.1. Dual Eigenvalue Analysis of a Skew Cantilever Plate

The first test example adopted is a skew cantilever plate. Its analysis has
been performed for various skew angles0, 15, 30, 45with three
different finite element models corresponding to the displacement,
equilibrium and mixed approaches (Fig. 4).
N o n dimensional
Circular f r e q u e n c y
Cuff2
J -
m
: mass/unit of area
D -- bending rigid ty
- skew ancle
Skew cantilever plate.

The plate bending conforming model (CQ). The element used in the
displacement approach is the purely conforming Fraeijs de Veubeke's
quadrangle represented in Fig. 5. 9 This wellknown element is obtained
from an assemblage of four triangular regions with cubic deflection. The
generalised displacements of the element are the deflection w and the two
slopes w/x and cw/cy at each vertex, and the slopes cw/dn normal to the
external interfaces. The resulting model has 16 degrees of freedom. It is
denoted by C Q in the tables and diagrams.
The plate bending mixed (u, v) model (MXQ). The mixed model, based on
separate approximations of displacements and velocities, is derived from
the purely conforming quadrangle CQ described above. It is denoted by
32
M. GERADIN
FlG. 5.
MXQ. A linear variation of velocities over the whole element has been
adopted to discretise the kinetic energy: account is thus taken of the only
inertia forces associated to rigid body motions. The resulting element
possesses 19 degrees of freedom: 16 generalised displacements and 3
generalised velocities.
As shown in Section 4.2, all the generalised displacements can be
eliminated before solving the eigenvalue problem, from which results a
considerable economy in eigenvalue computation.
The plate bending equilibrium triangle (EQT). This element, represented
in Fig. 6, results from the discretisation of Toupin's principle. The field of
self-stressing modes is obtained by assuming linearly varying moments. The
resulting variables are three concentrated loads at vertices, two bending
moments and one Kirchhoff shear load on each interface. The particular
solution of the equations of dynamic equilibrium is introduced next by
superimposing a cubic moment field that provides a linear variation of
velocities. 1013 The resulting model has 15 degrees of freedom: 12
generalised connecting forces, and 3 integral parameters in terms of which
the inertia forces of the element are represented. A considerable economy in
eigenvalue computation can thus be expected with this EQT model, just as
with MXQ elements, since both are based on a similar approximation of the
kinetic energy.
The numerical results. In the displacement analysis using CQ elements,
the finest grid consists of 6 6 elements and 231 degrees of freedom, from
which 37 are fixed. The results collected in Table II confirm the upper
boundedness guaranteed by the kinematical approach. A significant
deterioration of the finite element solution is observed when the skewness of
the plate is increased.
33
PARTICULAR
SOLUTION
CONJUGATED GENERALISED DISPLACEMENTS

W, - DEFLECTION AT VERTICES
Wlk -- MEAN DEFLECTION
W; and Wnlk- MEAN SLOPES
W - MEAN DEFLECTION
W- and W - MEAN SLOPES
ON EDGE / -
HE AREA
\ ON TH
FIG. 6. Equilibrium plate element.

The equilibrium approach (Tables III and IV) underlines the same
phenomenon. Distinct mesh patterns have been used in the square and skew
cases. In the latter case the finest mesh uses 50 elements and 420 degrees of
freedom. Only 150 of them contribute to the eigenvalue problem.
Note that in most cases the computed frequencies appear to be lower
approximations, except when the discretisation yields to a too rough
representation of inertia forces.
Another important factor, when using triangular elements, is the
skewness of the plate. The best subdivision is obtained as expected, using
two elements per quadrangle, with their common edge along the shorter
diagonal.
Mixed models exhibit not so systematic behaviour as conforming and
equilibrium elements. In most cases they give a convergence by lower
bounds, but such behaviour is not guaranteed.
The analysis with MXQ elements has been pushed up to a 8 8 mesh,
with a total of 352 generalised displacements that are eliminated statically
and 192 velocity parameters in terms of which the eigenvalue problem is
solved. The corresponding results are given in Table V.
All the remarks that can be made about the comparison of the respective
models arise from Fig. 7, which represents, formode 1, the convergence as a
function of three factors: (1) the number of degrees of freedom in terms of
34
M.
GERADIN
TABLE 11
CONFORMING APPROACH (CQ)
1 1
2x2
3.x 3
4x4
5x5
6x6
3-489
9-222
26-96
35-29
43-75
3-486
8-606
21-50
27-63
31-96
3-479
8-535
21-38
27-39
31-27
3-476
8-518
21-33
27-28
31-08
3-474
8-513
21-31
27-24
31-02
3-473
8-511
21-30
27-22
30-99
15
3-652
9-376
27-88
35-21
47-42
3-602
8-872
22-66
26-78
35-14
3-591
8-765
22-42
26-52
34-38
3-587
8-731
22-32
26-41
34-10
3-586
8-717
22-28
26-37
33-99
3-585
8-710
22-26
26-36
33-94
30
4-159
10-12
28-72
4009
57-33
3-988
9-683
26-22
27-06
4303
3-954
9-532
25-75
26-36
4205
3-943
9-480
25-56
26-12
41-67
3-938
9-456
25-47
2603
41-52
3-935
9-443
25-41
25-99
41-45
45
5-070
1219
31-42
48-87
72-04
4-722
11-47
28-43
34-15
56-91
4-624
11-37
27-65
32-62
52-35
4-582
11-32
27-37
32-18
51-38
4-560
11-30
27-25
31-98
51-04
4-547
11-29
27-18
31-86
50-90
which the eigenvalue problem is solved; (2) the skewness of the platethe
curves associated with different skew angles have been shifted horizontally;
(3) the type of model used.
The equilibrium models would normally give three distinct convergence
curves according to the mesh subdivision adopted. Only the most
favourable case (subdivision ZV) has been represented in Fig. 7.
TABLE III
EQUILIBRIUM APPROACH (EQT): 0 =
1 * 1
2 subdivisions
2*2
44
3x3
MODE
no.
El
3.447
3.453
3 462
3466
3.467
3.469
7850
8.246
8.410
8.470
8.475
8494
8.492
20.74
20.76
21 16
21.21
21 2 4
21.26
21.26
25.07
2673
2 6 89
27.07
27.10
27.16
27.16
2886
2892
30.14
30.72
30.75
30.89
30.87
35
VARIATIONAL METHODS OF STRUCTURA L DYNAMICS

TABLE IV
EQUILIBRIUM A PPROA CH (EQT):
3 types of
subdivision
15;30;45 t
2x2
3x3
2x2
1 1
^ s
2x2
5x5
4 x 4
15
3.378
7 374
11.82
22.35
39.37
3.400
7733
16.02
2018
47.97
3.502
8096
21.46
24 03
32.45
3.537
8.377
21.23
25.60
31.17
3.549
8478
21.64
2558
32 54
3.567
8563
21.97
2598
33.03
3.374
7917
10.95
23.11
37.92
3562
8.681
19.17
2108
4515
3.709 3 7 5 8
8.884 8.929
22.89 22.07
25 58 25.10
36.79 3 4 8 8
3.348
9 169
2405
25.33
40.02
3.830 3.857 3.892 3 9 0 6

9.257 9 166 9.297 9 3 3 8
2 4 5 6 23 91 24.76 2 4 9 7
2 5 3 3 2 5 3 2 25.68 2 5 9 8
40.33 39.14 40.82 41.03
3907
9 266
24.57
2560
40.34
3907
9.338
24.97
2578
41.03
3.914
9365
2509
25.85
41.18
3.344
8302
11.66
22.80
40.01
3 9 4 2 4.084
1025 1043
23.35 26.27
2 6 8 8 31.56
4 3 0 1 40.71
4.002 4.321 ! 4.370 4 2 3 5 4.406 4 432

11 16
10.34 11.00 i 11.06 10.82 11.14
2 2 6 9 25.57 26.01 24.67 26.35 26.53
25B3 3 0 3 7 3 0 4 0 2 8 5 4 30.84 30.98
35.32 48.09 48.79 4 6 3 7 49.97 50.27
4441
11.00
2658
2987
4816
4.441
11.18
25.60
31.06
5041
4.450
11.20
26.72
31.17
5056
45
3 5 6 7 3.565
8.571 8.607
21.84 2 2 . 0 0
2 5 9 8 26.12
3289 3332
^
3.570
8651
22.13
2624
33.59
a I ^
3.576 3.574 3 5 7 9
8632 8 6 4 9 8 667
2 2 0 5 22.11 22.16
26.18 26.25 2 6 2 9
33 41 33.60 33.71
5 0
convergence to mode no 1
4 5
- -
nondimenstona
frequency
.
tuO
/fm
QUADRANGLE C Q
TRIANGLE EQT
QUADRANGLE MXQ
40
frequency versus
sxew angle
35
|20Q
FIG. 7.
Skew cantilever plates.
log
36
M. GERADIN
TABLE V
MIXED APPROACH (MXQ)
2x2
4x4
6x6
8x8
3-332
8031
19-33
24-98
30-59
3-439
8-362
20-91
26-71
30-63
3-457
8-442
2114
26-99
30-84
3-462
8-471
21-21
27-08
30-90
15
3-455
8-304
20-20
25-15
32-57
3-552
8-579
21-82
26-05
33-42
3-569
8-641
22-07
26-22
33-68
3-575
8-665
2214
26-27
33-76
30
3-845
9-153
22-97
26-78
38-90
3-908
9-331
24-74
2601
40-72
3-920
9-375
25-07
25-94
41 05
3-923
9-391
25-17
25-93
41-17
45
4-594
10-96
28-40
31-41
53-82
4-552
11-15
27-35
30-89
51-32
4-533
11-25
27-14
31-27
50-79
4-524
11-24
27-08
31-40
50-71
The results reported in Table VI allow a comparison with analyses

achieved by other authors. Among others, note that the experimental
eigenfrequencies given by Barton, 4 even after the correction compensating
the virtual inertia due to aerodynamic forces, are not bracketed by the
bounds obtained from the dual analysis. The advantage offered by several
simultaneous analyses appears clearly in this case, as it allows for a better
understanding of the discrepancies between experimental and numerical
results. The differences are probably due to the difficulty of achieving a
perfectly clamped support.
A comparison of the various numerical results confirms the impression
that the 'consistent way of lumping masses' provided by mixed models can
yield to a significant saving of computational effort in free vibration
analysis.
5.2. Free Vibration Analysis of a Cylindrical Shell
An interesting application of plate equilibrium models is found in the
37
TABLE VI
SKEW CANTILEVER PLATES: COMPARISON OF RESULTS
Theoretical
(1)
3-60
8-87
15
3-96
1019
30
4-82
13-75
45
CQ
Test Conectei :/ Dawe Zienk iewic:
6 x 6 (4)
test (I)
(2)
(3)
(1)
3-38
8-63
21-49
2604
3301
3-44
8-68
3-82
9-23
24-51
25-54
40-64
3-88
9-33
4-26
11-07
26-52
30-13
50-19
4-33
11-21
MXQ
EQT
5 5 (4) 6 x 6 ( 5 )
3-59
8-71
21-59
3-57
8-60
21-75
3-58
8-71
22-26
26-36
33-94
3-58
8-67
22-16
26-29
33-71
3-58
8-67
22-14
26-27
33-76
3-95
9-42
25-56
3-98
919
24-56
3-93
9-44
25-41
25-99
41-45
3-91
9-37
25 09
25-85
41-18
3-92
9-39
25-17
25-93
41-17
4-59
11-14
27-48
4-67
1101
27-56
4-55
11-29
27-18
31-85
50-90
4-46
11-20
26-72
31-17
50-55
4-52
11-24
27-08
31-40
50-71
(1) Theory (Rayleigh-Ritz) and experiment: J. Appi. Mech. (1951).

(2) Parallelogram element (4 4, 75 degrees of freedom): Dawe. J. Strain Anal.
(1966).
(3) Triangular element (8 elements, 18 degrees of freedom): Zienkiewicz, Int. J.
Solids Struct. (1968).
(4) Displacement and equilibrium approaches with CQ and EQT elements.
(5) Mixed approach with MXQ elements.
analysis of shallow shells, for which quite good results can be obtained using
flat shell elements: within the limits of linear analysis, this approximation is
known to lead to errors on the displacement field and on strain energy
which are of the same order of magnitude as when using shallow shell
elements. 17
Flat shell elements, which are much simpler to derive, exhibit interesting
properties, such as an easier connection to other types of elements such as
beams, and allow for an easier discretisation of shells with strong curvature
discontinuities.
Within a flat shell element, the extensional and bending displacement
fields are fully uncoupled and can thus be analysed separately.
An interesting solution consists of combining an equilibrium plated
bending element, constructed on the basis of a linear approximation of
bending moments, with a quadratic membrane element as shown in Fig. 8.
38
M. GERADIN
conforming
membrane
triangle
equilibrium
bending
triangle
FIG.
FIG. 9.
Flat shell element: mixed assumptions.
/?=24in
L - 12 in
/ = 0 120 in
E- 6 0 3 5 7 lb/in 2 (steel)
P- 0.2817 lb/in 3
=0.3
FIG.
10.
39
The membrane elements are defined in terms of the local displacements

(u, v) at vertices, and average displacements (, ) on the edges. The plate
bending element, when expressed in terms of generalised displacements, has
a stillness matrix expressed in terms of the local deflection at vertices, and
two local rotations and the average deflection on the edges. Internal
parameters have to be added to represent the kinetic energy of the rigid
body modes of the bending plate.
The resulting element, corresponding thus to mixed assumptions is
particularly easy to use owing to the system of degrees of freedom in terms
of which its stiffness matrix is obtained : the local displacements (u, v, w) at
each vertex, the average displacements (7, v, w) and two local values of the
normal slope along the edges, and three more internal parameters to
represent the kinetic energy of the plate. This element eliminates the
difficulty associated with the use of rotations at vertices. The internal
degrees of freedom can be condensed statically at the element level, in which
case the final element (Fig. 9) has 24 degrees of freedom.
The element described above has been used to analyse a constant
thickness cylindrical compressor blade (Fig. 10), for which experimental
and numerical results were available. 18
TABLE VII
COMPRESSOR BLA DE EIGENFREQUENCIES
Modes
1A
2S
3S
4A
5S
6A
7S
8A
9S
10A
I IS
12S
13A
14S
15A
Olson and Lindberg

cylindrical (04)
Mb :ed
Test
2x2
4x4
2x2
4x4
957
1449
2648
4134
4612
5838
866
1392
2513
3486
3934
5334
7464
7521
7901
8138
1 0089
12318
1 2662
12456
1 2862
670
1107
2450
2624
3182
845
1377
2471
3314
3696
519
6941
7015
7646
81317
9861
1 1658
1 2340
1 1680
1 2640
856
1345
259
351
395
531
751
743
790
809
997
1216
1241
1252
1281
40
M. GERADIN
|\^
\ \^
; y
\/
; \-
M0DE3
Frequency 2 4 7 1 Hz
/
\.
\ ^
:^\
y'
Frequency 331 4 Hz
\
M0DE6
MODE 5
MODE 4
\
y
w
Frequency
Frequency^ 3 7 0 Hz
; \
;\~
;/
\ \,
519 Hz
y'
: \.
./,
'; /
MODE 7
MODE 8
MODE 9
Frequency : 701 5 Hz
Frequency : 6 9 4 - 1 Hz
Frequency 7 6 4 5 Hz
MODE i o
Frequency : 8 0 1 Hz
FIG. 11. Compressor blade: eigenmodes.

Table VII compares the numerical results obtained with the element
described above with those of Olson and Lindberg, 18 and with their
experimental investigation. Olson and Lind berg's element is a shallow shell
element, obtained using a fifth-degree interpolation of displacements. It
uses not only displacements and rotations, but also curvatures as
connecting variables at vertices.
One finds, as in plate bending problems, a tendency for the element to
give lower bounds to the exact frequencies of the structure. This is to be
expected because the eigenmodes, which are represented in Fig. 11, involve
essentially bending deformations.
REFERENCES
1. Reissner, E. (1948). 'Note on the method of complementary energy', J. Math.
Phys., 27(2), 159-60.
2. Reissner, E. ( 1949). 'Complementary energy procedure for flutter calculations".
J. Aero. Sci., 16(2), 316-17.
3. Reissner, E. (1950). 'On a variational theorem in elasticity', /. Math. Phys., 29,
90.
4. Barton, M. V. (1951). 'Vibration of rectangular and skew cantilever plates', / .
Appi. Mech., 18, 129-34.
5. Toupin, R. A. (1952). 'A variational principle for the mechanical type analysis
of mechanical systems', J. Appi. Medi., 19, 151-3.
VARIATIONAL METHOD S OF STRUCTURAL D YNAMICS
41
6. Fraeijsde Veubeke, B. M. (1965). 'Displacements and equilibrium models in the

finite element method', in Stress Analysis, eds. O. C. Zienkiewicz and G. S.
Holister, Wiley, New York.
7. Pian, T. H. H. (1964). 'D erivation of element stiffness matrices by assumed
stress distributions', AI AA J., 2, 1333-6.
8. Pian. T. H. H. and Tong, P. (1968). 'Rationalization in deriving element
stiffness matrices by assumed stress approach', in Proc. 2nd ConJ. Matrix
Methods in Structural Mech., AFB, Ohio. USAF Report AFFD L-TR-68-150,
441-69.
9. Fraeijs de Veubeke, B. M. (1968). 'A conforming element for plate bending'.
Int. J. Solids Structures, 4, 95/108.
10. Geradin, M. (1971). 'Computational efficiency of equilibrium models in
eigenvalue analysis', in I UTAM Symposium on High Speed Computing of
Elastic Structures, Vol. 2, pp. 589-623. Congrs et Colloques de l'Universit de
Lige.
11. Fraeijs de Veubeke, B. M. (1972). 'A new variational principle for finite elastic
displacements', I nt. J. Sci., 10, 745-63.
12. Idelsohn, S. (1972). 'Influence du nombre des modes d'autotension dans
l'analyse dynamique des plaques au moyen d'une famille d'lments
statiquement admissibles", LTAS, Universit de Lige, Rapport VF-21.
13. Geradin. M. (1973). 'Analyse dynamique duale des structures par la mthode
des lments finis', Coll. Pubi. Fac. Sci. Appi., No. 36.
14. Hughes, T. J. R. (1976). 'A reduction scheme for problems of structural
dynamics', I nt. J. So/ids Structures, 12, 749-67.
15. Geradin, M. and Sander. G. (1977). 'Variational principles in elastodynamics',
LTAS, Universit;de Lige, Rapport VA-18.
16. Geradin, M., Sander. G. and Nyssen, C. (1977). 'Mixed variational
formulations in linear and non-linear structural dynamics', Symposium on
Appi, of Computer Methods in Engineering, Los Angeles. 23-26 August.
17. Idelsohn, S. ( 1974). 'Analyse statique et dynamique des coques par la mthode
des lments finis". Thse de Doctorat, Laboratoire d"Aronautique, Universit
de Lige, Rapport SF-25.
18. Oison. M. D . and Lindberg, G. M. ( 1971). 'Dynamic analysis of shallow shells
with a doubly curved finite element', J. Sound Vibration, 19(3), 299-318.
19. Fraeijs de Veubeke. B. M., Geradin, M. and Huck, A. (1972). Structural
Dynamics, CISM Courses and Lectures No. 126, Springer Verlag, Berlin.
20. Courant, R. and Hilbert, D . (1937). Methods of Mathematical Physics, Vol. 1,
Interscience, New York (English edition. 1953).
21. Elias, . M. (1973). 'On the reciprocal form of Hamilton's principle', J. Appi.
Mech., 40, 93-100.
22. Key, S. W. and Krieg. R. D . ( 1972).'Transient shell response by numerical time
integration', 2nd US-JAPAN Seminar on Matrix Methods in Structural
Analysis. Berkeley, California.
23. Geradin, M. (1977). 'Application of the Lanczos Biorthogonal Algorithm'.
LTAS Universit de Lige, Rapport VA-17.
24. Geradin, M. (1978). 'On the Lanczos Method for solving large structural
eigenvalue problems", ZAMM, 59(3/4), 127/129.
25. Zienkiewicz, O. C. ( 1971 ). The Finite Element Method in Engineering Science.
McGraw-Hill, New York.
The Modal Method for Transient Response and its

Application to Seismic Analysis
K . FULLRD
Central Electricity Generating Board, Berkeley Nuclear Laboratories,
1.
UK
INTRODUCTION
The method of calculating the response of vibrating structures known as the

modal method is widely used where elastic response is expected. Although it
can be used for a wide variety of forcings, with and without damping, it is
most useful for problems where the loading varies with time. This is simply
because the response is treated as being a combination of those deflected
shapes which define its normal modes and this is equivalent to using a
restricted number of degrees of freedom (the modes themselves), compared
with the actual degrees of freedom of the structure. The transient analysis
by modal methods is, consequently, more economical than a full analysis in
terms of all degrees of freedom.
This paper deals first with the problem of reducing a multi-degree-offreedom system to a smaller set of degrees of freedom. This problem seems
to have come into its own with the increased use of finite element methods.
In static analysis there is basically no difficulty with large numbers of
degrees of freedom, but when modes of a system have to be extracted, the
cost may become astronomical. Eigenvalue methods are touched on before
modal methods for calculating transient response are discussed. The paper
ends with a discussion of earthquake excitation treated as a transient
phenomenon and by response spectrum methods.
2.
REDUCTION OF MULTI-DEGREE-OF-FREEDOM
MATRICES FOR EIGENVALUE COMPUTATION
The finite element method is now in widespread use for both dynamic and
static analyses. It is a powerful technique but, for dynamics problems, has
43
44
K. FULLRD
the drawback that a large number of degrees of freedom are generated in

defining the structure. The drawback is associated with the methods
commonly in use for extracting natural frequencies and mode shapes of a
system. Typically, the full set of degrees of freedom in a finite element model
of a structure will be numbered in the thousands, while the natural
frequencies are obtained by operating on matrices of the order of one or two
hundreds. For this reason, economisation procedures were outlined by
Guyan 5 and Irons 9 which would reduce stiffness and mass matrices from a
large to a small size. In effect, a transformation is made from the original set
of variables to a smaller set, by ignoring certain of the forces which would be
required to displace the structure.
Consider the static set of equations
]\\
= >p\
or
K12
K2I
(1)
K27
where , , are stiffness, displacement and force, respectively. Now let

{P, } = JO}. It follows that
{. = -[[12){2}
(2)
Using eqn. (2), we can rewrite the displacement vector:

^11
^12
!<52} =
(3)
[]{2\
From eqn. (3), substituting into eqn. (1), we have a new set of
relationships:
l-K21K-,xKl2
+ 22]{2}= \P2\ = [*]{2]
(4)
This could also have been derived from strain energy considerations:
=2-{1}\\[]]{1}
= 2-{1}*]{2}
(5)
If a similar transformation is applied to the mass matrix, keeping the

kinetic energy constant, we obtain a reduced mass matrix:
|{} T [M]{} = 4{ 2 }T[n T [M][ri{ 2 } = i{2}T[M*]{<52}
where {} is a velocity vector and
IM]* =
IK2IKU\I]
My,
M,
M,
= M22 + K2K;llMiK;llK2
"
l l
12
,,'/,,
K2iK7{Mi
(6)
THE MODAL METHOD FOR TRANSIENT RESPONSE
45
Guyan described reduction in the above manner, while Irons gave details
of how the method could be programmed in a similar manner to the front
solution technique used by many finite element programs.
The reduced stiffness matrix is, in fact, simply the inverse of a flexibility
matrix in the same degrees of freedom. Consider an encastre-free beam with
a transverse and rotational degree of freedom at its end. The
displacement-force relationship is given by
EI
12
-61
71
-61'
Al2
(7)
using simple bending theory.

If the rotational term is eliminated, using eqn. (4) we obtain
EJ7-36/2
or
4/3EI
12 F v = F
= F = *
(8)
Now, solving eqn. (7) we obtain a flexibility matrix:

.Vi
9)
/ ~2l2
6EI _ 3 /
3 F
\
6_ \M
(9)
and if M is zero,
I3
=
A = A*- ]F
(10)
which is simply the flexibility version of eqn. (8).

Selection of the limited set of degrees of freedom (the masters) requires
common-sense and some reasonable knowledge of the mode shapes of the
structure. Discussing a cantilever-like structure, Irons identified 'slave'
degrees of freedom which should be eliminated as : ( 1 ) lengthwise movements
if these are associated with high frequencies; (2) movements near the
encastred root, because these contribute little kinetic energy in the lower
modes; and (3) the slopes, because although these are vital variables, they
are unlikely to be used in plotting the modal shapes from the eigenvectors.
Similar arguments could apply in setting up a finite element mesh, large
elements being used where the kinetic energy is likely to be low. However, as
a result of the availability of the reduction procedure, the sizing of the mesh
can be considered a secondary problem in many cases.
46
K. FULLRD
Byway of illustration of the results of economisation, consider a simple,

uniform section cantilever beam. Its first three natural frequencies are
3516, 22033a and 6F101ccrad/s, where a yJ(E\/pAl4),
assuming
classical theory. If an analysis is carried out using Timoshenko beams, quite
a small number of degrees of freedom should give good results. One
transverse master degree of freedom at the end of the beam gives the first
frequency as 3567; two equally spaced transverse degrees of freedom give
3522; three transverse degrees of freedom give 3517a; and if the latter
have three rotational degrees of freedom as well, the fundamental
frequency is 3516a. The last result confirms Irons" recommendation about
treating slopes as slaves. If we have three equallength elements to form the
beam but only use the outer two transverse degrees of freedom to calculate
the first mode, we get the same result as if two elements had been used of
twothirds and onethird length from the encastre end. This is, again,
3517a, which supports Irons" view about neglecting movements near the
encastre end and also lends weight to the relative sizing of elements as far as
the fundamental mode is concerned.
The third mode requires a minimum of three degrees of freedom for its
computation. Figure 1 shows, in comparison with the theoretical value,
how choice of these degrees of freedom and their location influence the
calculated value. Biggs1 indicates nodes (points of zero transverse
displacement) at 05/and 087/from the builtin end, where /is the length of
the cantilever. It follows that locating a transverse degree of freedom at
either of these positions is wasteful. Very approximately, rotational degrees
of freedom at 03/and 07/are similarly wasted. Referring to Fig. 1, it can be
seen that appropriately placed transverse degrees of freedom need not be
supplemented by rotational degrees of freedom and that case II gives a
value little better than that given by case III.
The cantilever beam is a trivial problem compared with other structures.
Even a typical piping run, with its changes of direction and multisupport
configuration, will give rise to difficulties in terms of choosing the master
degrees of freedom. Henshell and Ong 6 offered a method by which the
choice of master degrees of freedom to give the lowest frequencies might be
made automatically. Irons had shown how the necessary changes to the
stiffness and mass matrices could be made using a frontal approach. In
other words, slaves were to be eliminated as soon as all contributions to
their mass and stiffness had been made, and in the process of elimination, all
remaining variable values were modified to reflect the slaves' contribution.
The final mass and stiffness matrices would, of course, be in terms of the
selected degrees of freedom. Henshell and Ong proposed that in the absence
47
IY
Degrees of freedom
FIG. 1. Third-mode frequencies for a cantilever.
of identified slaves, after assembly of elements to the point at which a
predefined number of degrees of freedom were exceeded, completed mass
and stiffness terms could be removed to bring the total degrees of freedom
back to that number predefined. This process would then be repeated each
time the addition of further degrees of freedom caused excedance of the
required number. The removed degrees of freedom are, of course, the
system slaves, and the modification to the mass and stiffness matrices is as
48
K. F U L L R D
Irons described. The selection of thosedegrees of freedom to be removed is

done as follows. The ratio of stiffness to mass is equivalent to frequency
squared. The desired set of frequencies is the lowest, so degrees of freedom
which give high frequencies should be eliminated. Henshell and
Ong proposed that all that was necessary was to scan the leading diagonal
stiffness divided by mass values of completed degrees of freedom and
remove the highest. As the matrices change with every elimination, the scan
has to be done prior to each elimination.
selected master
""
FIG. 2.
Pipework exampleautomatically selected degrees of freedom.
It is usual to combine selection of some master degrees of freedom with

automatic choice of the rest. Forcing of a structure is carried out in degree
of freedom directions, and it would cause difficulties if the automatic
method eliminated those forcing degrees of freedom. Figure 2 shows the
master degrees of freedom for a simple piping run, with one master selected
for forcing reasons and twenty automatically chosen masters. Figure 3
shows a projected view of the fundamental mode shape for the system.
Although there are no rotational degrees of freedom for the view shown,
and only two degrees of freedom in the Jfdirection. the Zdisplacements
give sufficient information to draw a sensible mode shape. In fact, for this
mode, eight of the selected masters displace less than 01% of the
maximum, and three of these are torsional degrees of freedom which might
not have been selected in advance at all. However, no displacement is less
than 01% of the maximum for the third mode, one torsional is for the
7777
49
*"Z
FIG. 3. Fundamental mode of pipework.

fourth mode and none is less than 0-1 % for the fifth mode. The degrees of
freedom automatically chosen for the branch are almost certainly those
which would have been selected, since they are needed for the fundamental
mode of the branch. Bearing in mind the F-direction restraints, again there
is probably little difference between the automatically chosen Y degrees of
freedom and those which might have been selected 'by eye'.
50
K. FULLRD
Generally speaking, Henshell and Ong found that for a large degree of
freedom problem, automatically chosen masters gave more accurate
frequencies than the same number of selected masters. They also said that if
the number of masters was too low, the method was not very satisfactory in
association with the frontal method, because there would be too few
potential choices at any stage of the frontal process and important
freedoms could be eliminated. This would not happen if the whole of the
stiffness and mass matrices were derived before any reduction commenced.
They suggested that the method should not be used if less than about twenty
masters were needed.
The slave displacements should not, strictly speaking, be recovered from
eqn. (2), since they would be static and not dynamic displacements.
Referring to the equations of motion, we can establish a relationship
between slave and master displacements which includes inertia terms:
Ml2
M22
Mn
M21
^11
-^12
-^2 1
-^22
Then
<, = (
+1(-22
+ 12){2}
(12)
Provided that we are considering values of 2 which are lower than any
eigenvalue 2, of ( 2 , , + ){1},
then eqn. (12) is approximately
(<5, = ( A n ' + 2 1 1 1 / / 1 1 1 1 ) ( 2 1 2 + A 1 2 ){ 2 }
(13)
It follows that, for each mode of interest, a full deflected shape in terms of
both slaves and masters may be obtained by first solving ( 2*
+ A '*){2 and then substituting for {d2 in eqn. (12) or eqn. (13). A n
alternative derivation of the reduced mass and stiffness matrices may be
made incorporating the consideration of eqn. (11). This gives the same
results as eqns. (4) and (6), and was discussed by Kidder 1 0 1 ' and Flax. 3 The
important points are that the lowest frequencies are those sought and the
elimination of degrees of freedom is based on both mass and stillness
considerations (i.e. highfrequency elimination), which is similar to the
Henshell and Ong approach.
It may well be that the differences between the slave displacements
recovered from eqn. (2) and those obtained from eqn. (12) or eqn. (13) are
very small for lowfrequency modes. However, the effect on stresses in
individual elements may not be acceptable, because the tendency with the
Guyan slaves will be to smear out load variations between master degree of
freedom locations. This is obvious when it is remembered that the
THE MODA L METHOD FOR TRA NSIENT RESPONSE
5]
assumption of zero load in the slave displacement direction still applies in

this case.
3.
DERIVA TION OF EIGENVA LUES A ND EIGENVECTORS
It is not intended in this section that a comprehensive review of eigenvalue

extraction methods be given. Many excellent reviews have appeared in the
technical press over the years, and Wilkinson 12 provided a valuable
reference work for the interested researcher. A few of the important
properties of eigensolutions will be mentioned in so far as they are relevant
to modal methods for calculating the response of structures.
In dynamics a solution is sought to the unforced, undamped multi
degreeoffreedom problem which can be written thus:
[M]{y] + [K]{y] = {0}
(14)
A solution to this problem is

{>'(')} = {>'(0)}cosji + ' { j ( 0 ) } s i n w .
(15)
where {y(0)\, {>'(0)} are initial displacements and velocities, and is a

natural frequency of the system.
Making the substitutions in eqn. (14) for \y) and \y] using eqn. (15)
yields
(2[]
+ ]){}={0]
(16)
which is the eigenvalue problem. There are as many solutions as degrees of

freedom, although all eigenvalues, 2 , may not be distinct. Associated with
each eigenvalue there is an eigenvector, {j, j, which is arbitrary to the extent
that A : {y j would also be a solution corresponding to the same eigenvalue, k
being a constant multiplier. It is usual to choose k so that the components in
the eigenvector have a preferred numerical property. The eigenvector is
then 'normalised'. Typical normalisations are those for which (a) the sum of
the squares of the moduli of the components is equal to unity, (b) a
component of the vector which has the largest modulus is equal to unity, (c)
the sum of the moduli of the components is equal to unity. Type (b) is widely
used and is sometimes referred to as 'engineer's normalisation'.
Equation (16) can be rewritten in the form
([]'[]2[/]){>}={0!
(17)
Provided that there are no rigid body modes in the system, the stiffness
52
K. FULLRD
matrix, A , is symmetric and positive definite. M is also symmetric and

positive definite ifall massless degrees of freedom have been removed from
the system. M''Kis
not symmetric and, for some eigenvalue extraction
techniques, has to be transformed to make it symmetric. Other methods
work directly on matrix equations in the form of eqn. (16) or, more
generally,
[][] = [][][2]
(18)
2
where the columns in [] are the lowest eigenvectors and [ ] is a diagonal

matrix with the corresponding eigenvalues. Yet another technique relies on
the fact that for a nontrivial solution to eqn. (16) or eqn. (17) the matrix
( 2M + A ) must be singularthat is, the determinant det (2M + A )
is zero.
Whatever method is used to obtain the eigenvalues and vectors, a most
important property of the eigenvectors is their orthogonality. That is, for
any two eigenvectors {} and {}, where m , their scalar product
\,[]{) is zero, where IB] is a matrix of the system. For a dynamics
problem the proof of this may be shown as follows.
Let
2[]{} = []{)
(19)
be a solution to eqn. (16) and

2]{)
= []{\
(20)
be another solution. If eqn. (19) is premultiplied by {}, we have

2,{}]{} = {]\\
(21)
and writing this in the transposed form gives

2 {, [] {} = {, }[{}
Premultiplying eqn. (20) by {,\

2
(22)
and subtracting from eqn. (22):
( ,- ){,} ]\}
=0
(23)
and for , , we have the orthogonality conditions

{[]{} = {]{)
=0
(24)
There are circumstances when the stiffness and mass matrices are not
symmetric. Such a case arises in a method of treating rotationally periodic
structuresthat is, structures whose geometry defined for any radial and
axial position at angle 0 is identical at + 0, where 0 is 2/ and and
53
TV are integers. N is structure dependent and n is any integer less than TV. A
cooling tower on legs is a typical example of a rotationally periodic
structure. The dynamic characteristics of the complete structure can be
obtained by considering only a substructure, but the matrices are
Hermitian.
[Hermitian matrices have the form (A + iB), where A is real and
symmetric and is real and skew symmetric (BT = B). If C = A 4 iB,
C T = C, where C is the matrix whose elements are the complex con
jugates of those of C. C T or C H is referred to as the Hermitian transposed
matrix. It is a property of Hermitian matrices that their eigenvalues are real
and their eigenvectors are complex.]
4.
THE USE OF M O D A L METHODS FOR CA LCULA TING

TRANSIENT RESPONSE
The expression 'modal method' here refers to the technique of solving

elastic structural response problems in terms of variables which are
multipliers of the free, undamped mode shapes of the structure
(eigenvectors). There are circumstances in which the mode shapes may be in
terms of complex numbers, but this is not discussed in the following.
The structural response problem may be written in the following matrix
form :
lM]\x) + [C]{xi + [K]{x] = [Fit)}
(25)
where \x\ is the response vector in terms of master degrees of freedom;
[M], [A ] are (reduced) mass and stiffness matrices; [C] is a damping matrix;
and \F(t)', is the timedependent loading vector.
When \F(t)\ is simply a set of sinusoidally varying forces, {P}eia", the
steady state solution of eqn. (25) is straightforward.
Let the response, {x] be written as
\XR\e,u" + i{Xx}em'
(26)
where {A "R} and {",} are real and imaginary components of response.
Substituting eqn. (26) into eqn. (25) gives
( A 2){]
wC{Xt) + i(K orM)\XK\
+ iC{X^\ = {)
(27)
This can be arranged in matrix form:

(KorM)
cuC
toC
A( o j 2 M )
XA
P[
0
(28)
54
K. FULLRD
The solution to this gives

{XR} = O J " 2 C " ' ( A orM)C\I
+ [_1(
OJ2M)C"
1 2
Aj = c u _ 1 C I ( / + [ w " ' ( A W 2 M ) C ' ] 2 ) ^ P
] ) " ' (29)

(30)
It can be readily seen from this form of expressing the response that if OJ
corresponds to a natural frequency of the system, then (A urM) is zero.
In this case {A 'R | is zero and
{Xl)=a>l[C]l{P)
(31)
If there is no damping, clearly the only valid solution to eqn. (25) is

[XRl = (KorM)'{P\
(32)
The above discussion is given only to show that with harmonic excitation
there should be no difficulty in obtaining a solution to the structural
response problem. If complex solution techniques are available, then
solving eqn. (27) is simply equivalent to solving a complex stiffness matrix
or impedance matrix:
[(A 2) + itoC]{XR + iX, ! = {P}
(27a)
When the forcing is not harmonic, the above impedance matrix approach
can still be used, with little extra difficulty if the forcing is periodic (via
Fourier components) and by periodic approximation if the forcing is non
periodic. However, as an alternative to this frequency response technique
(so called because the response is related to harmonic excitation) we have
modal methods of solution.
Returning to eqn. (25), the unknown response vector {x} is expressed as
the sum of proportions of the normal mode vectors:
W = > {,t= []!?}
(33)
so that eqn. (25) becomes
]]\) + [][]{4) + [K\m\q)
= {F{t)}
(34)
If eqn. (34) is premultiplied by []1, it becomes

mTlMW]{q}
+ [0]T[C][<D]t) + [][][]{<7} = mJ{F(,)} (35)
55
We have already seen that there is an orthogonality condition of the

eieenvectors:
, )[]{] = {, \[]{ I = 0
and
or
for m
{,! []{,1 = {</>, ; []!,
(36)
(37)
(o'm"2mmm = k
m
Thus [][][] and [][][] are diagonal matrices which we can

write as [M*\ and [A *]. In addition, the triple product []'[0[] will be
diagonal if the damping matrix, [C], is proportional to mass or stiffness.
For con venience, [C] is often taken to be represented by a[ A] + 2[M],and
this is Rayleigh damping, is a form of inverse frequencydependent
hysteri tic damping, while pMs frequencydependent viscous damping. When
this form of damping is assumed, eqn. (35) becomes
[M*]{q] + (2[*] + x[K*]){ql + [K*]\<ll = {^(0*)
(38)
Thus, for each mode there is an equation

m + 2miqi + ak ej, + kq =Jt(t)
(39)
Ol
ii, + (2 + ?)?, + (ohi = / ( 0 M
(39a)
Now the form of damping which is probably most commonly used is

viscous, and in this case eqn. (39a) is replaced by
q, + ,,, + (of q, = f(l)m,
(40)
where , is a constant equal to the ratio of the actual (modal) damping

divided by the critical damping in the mode (CcrU = 2mw). It is possible to
select values of and which will give the same coefficient of q as 2 for
just two modes. A ll other modes would then be over or underdamped
depending on a comparison of (I + oteo2) with 2. For this.reason,
when the damping of a structure is specified as a proportion of the critical
viscous damping (possibly varying from mode to mode), eqn. (40)
is a more appropriate form of the modal equation than eqn. (39a).
It should be pointed out that damping matrix [C] does not have to be
known if viscous damping is assumed. The form of damping used in eqn.
(40) merely implies the form of the triple product. That is,
[] [] = 2[*][][]
(41)
56
K. FULLRD
where [] is the diagonal matrix of natural frequencies and [] is the

diagonal matrix of critical damping ratios.
The result being diagonal means that matrix [C] must be an orthogonal
damping matrix. This is not the same matrix as would be formed if the
damping consisted, say, of dashpots between masses. Wilson and
Penzien 13 show how an orthogonal damping matrix can be derived with
arbitrary modal damping assigned to selected modes.
In order that the response of the structure may be determined, solutions
to either eqn. (39a) or eqn. (40) are required for each participating mode. In
principle, all modes participate in the response of the structure, although in
practice only a selection of modes need be employed, depending on the
nature of the loading. For example, if a simply supported beam is loaded by
a time-varying, uniformly or symmetrically distributed load, those modes
whose eigenvectors are asymmetric need not be considered. This emerges
from consideration of the right-hand side of eqn. (34). {F(t)} is written in
the form Fmd!i{g(x) }p(f ), where {g(x)} is a normalised spatial distribution of
load, having a maximum value in location and time of F max , and/?(/) is the
time variation of the load. Then, if \) is an eigenvector with symmetry
such that the product \, {g(x)} is zero, the response in mode m is zero,
provided that initial values of qm and qm are also zero. The product
fm}T{g(x)} is often called the participation factor, Fm.
Highfrequency modes of structures which are essentially uniformly
loaded, while giving rise to nonzero participation factors, may also be
neglected if is very small. If one considers the uniformly loaded simply
supported beam, the nth mode shape is given by sin (nnx/l), I being the span
length and a distance from one of the supports. , is 2// and is 21/
for odd, zero for even. In addition, many forms of transient loading give
rise to modal response which is inversely proportional to the modal
frequency squared. These two factors permit us to neglect highfrequency
modes. It should also be remembered that exponential decay of the
transient part of the response is the more rapid the higher the natural
frequency of the mode.
Modal methods can also be used when a structure is treated by a
substructuring technique. Fricker 4 describes a modal synthesis method in
which the substructure degrees of freedom are transformed into a reduced
set of modal degrees of freedom. These are then used to synthesise the
vibration modes of the complete structure. All degrees of freedom attaching
a substructure to adjacent substructures are constrained to zero in
calculating the substructure modes. These modes are supplemented by a
further set of modes which will release the 'attachment' degrees of freedom
57
('constraint' modes). There are as many constraint modes as attachment

degrees of freedom, thus ensuring a statically complete set of modes and
also giving low-frequency accuracy. The basis for the method was first
proposed by Hurty 7 , 8 and further refined by Craig and Bampton. 2
With constraint modes, the internal degrees of freedom are free, while all
but one of the attachment degrees of freedom are held to zero. The single
free attachment degree of freedom is given unit displacement. It follows that
the constraint mode shapes are readily derived from the stiffness matrix of
the substructure:
âa
âi
A;.,
A;;
(42)
where subscripts a, i refer t o attachment or internal degrees of freedom; [ F J

is the set of forces required to give the appropriate attachment
displacements; and [<J are the constraint mode shapes.
From eqn. (42)
[^]=-Ai71Aia
(43)
[Although there must be as many constraint mode shapes as attachment
degrees of freedom, the number of internal degrees of freedom defining the
mode shape can be limited by the use of economisation procedures.]
The original attachment and (reduced set of) internal degrees of freedom
can be related through the constraint and substructure mode shapes to a
smaller number of degrees of freedom:
I
0"
0c
q.
= [T\{q]
(44)
where [</>J are the substructure mode shapes; {x.d\ = {qa} are attachment
degrees of freedom; {x-A are original internal degrees of freedom; and {qs}
are modal degrees of freedom.
Matrix [77] is used to transform the substructure mass and stiffness
matrices defined in terms of both attachment and internal degrees of
freedom (sinusoidal excitation):
(-2[[][\
or
+ [Tf[K][T)){q}
-co2[m) +
[*] =
= IT]T{F}
[k]){q}={f}
(45)
(45a)
(46)
58
K. FULLRD
where
kCi = :sc = 0
ks =
1.,
Note that kSi is a diagonal matrix

[m] =
(47)
m.
here
Wcc
= M a a - 3 ,7 >M ia - MiBK^Kia
"Jss =
+ A ai Ai7 ',,, ' A ia
and 7??ss is also a diagonal matrix: [ss] = [a>2][mss], where [OJ2] is a diagonal
eigenvalue matrix for the substructure.
From the above it can be seen that Acc and mcc are identical with the
reduced stiffness and mass matrices which would have been derived if
attachment degrees of freedom had been the only masters, [k] and [m] can
be regarded as Guyan reduced matrices supplemented by a truncated set of
fixed constraint substructure modes.
The force transformation of eqn. (45) is
:/":
(/il
UJ
JF^K^K^FJ
* F
(48)
where the forces include all externally applied forces and also forces on the
attachment degrees of freedom imposed by adjacent substructures.
When, for example, two substructures are joined, the following matrices
are formed :
"*+*
0
0 "
(49)
[A*] = 0
'11 0
0
0
k[2>
[M*] =
(50)
\F*\
(51)
since attachment forces at the interface are equal and opposite.
59
The undamped equation of motion for the coupled substructures is now

given by
( 2 * + A *) {q*}= {F*}
(52)
With substructures it is not possible to include modal viscous
damping, but Rayleigh damping, giving a diagonal transformed matrix,
css = akss + 2ms, can be included when the final system matrices are
assembled. A matrix icoC* is then included on the lefthand side of eqn. (52).
The advantages of using substructuring techniques for dynamics
problems may in some respects be outweighed by the extra information
required concerning the substructure boundary degrees of freedom.
However, the size of each substructure in terms of semibandwidth of the
matrices is usually less than for the complete structure, so that the reduction
procedure costs less. If eqn. (52) is solved to give the natural frequencies of
the complete structure, the number of degrees of freedom involved is
generally substantially less than would be the case if substructuring had not
been employed. That is, provided that the number of attachment degrees of
freedom is also relatively small. (If the substructure modes up to the
frequency of interest for the complete structure are retained in the analysis,
then accuracy should be ensured.) It follows that the structures which lend
themselves to the use of substructuring techniques are those which can be
split at sections with few degrees of freedom. A principal advantage of
substructuring is simply that if a substructure is modified, the reanalysis
costs of the assembled structure are much less than would be the case for the
complete structure.
5.
EA RTHQUA KE EXCITA TION OF STRUCTURES
When a structure is excited by an earthquake, its supports have a transient

motion impressed upon them. The complete motion involves three
translations and three rotations at every support, although it is not unusual
to find that analysts have in effect only considered one set of translations
applied simultaneously at all supports.
The support motion is often defined as a set of acceleration versus time
values. This is the way in which the original information about a ground
movement is derived from instrumental recordings. The time points will not
be at fixed intervals, and it is assumed that the acceleration varies linearly
between consecutive time points. The instruments can be of various types,
and skill is required in extracting data from the traces. The accelerograms
need correcting subsequently so that erroneous behaviour is avoided (for
example, the final velocity given by integrating the accelerogram should be
zero; the mean acceleration should be zero; and so on).
60
K.FULLRD
The acceleration time histories at the supports can be integrated to give

velocity and displacement time histories. Now, for the structure let [M], [C]
and [] represent the unrestrained mass, damping and stiffness matrices. If
{x}, {y}, {z}, jo}, {} and {} are vectors representing the structure
translations and rotations and if {}, {}, {}, {}, {} and {} are
vectors of displacements at the support points, then the resulting equations
of motion may be written in the form
(53)
[tit]
where \p\ represents the supports displacement vector:
{{},{},{},{},{},{\}'
and {q} represents the structure displacement vector:
{{},{}}}}}'
{Fp} is the reaction force vector applied by the structure to the points of
motion input. These unknown forces maintain equilibrium, and the
equations in (53) which involve these forces are equilibrium equations and
can be neglected.
There are no externally applied forces on the structure.
Since the unrestrained matrices have been used, a datum displacement,
velocity and acceleration must be chosen to essentially fix a 'ground' point.
Let Xy, Y0,ZQ, O 0 ^ 0 a n d 0 be the datum displacements in each of the six
directions. Then in the direction:
and
*i = {c/} +
{x}= {} +
(54)
where {U} and {u} are support and structure relative displacements.
Similarly for the other five directions.
Thus, the vectors {/?} and {q} can be expressed in terms of the relative
displacements {U} and {} by
iV}^
|1 \Xo
{lô
w<
'MZ0
{>}= { / M + </>2= < {A} > + < {1! 0 >
[B\
{
v i l To
{V\
(55)
61
and
OH
{}
!H
= <
{ ? } = W i } + {<h
{oj
>+
0
!0 >
(56)
{1}
\i
Lir) J
Substituting eqns. (55) and (56) into eqn. (53) gives
',/y
(/>'i} + {p2j
{?! + i?2
c.e;
.
AjAV
A, A
til + {q2}
{Pi} + iPi
i9iHg2)
P J
{0}
(57)
where [M], [C] and [A ] have been partitioned as above.

Neglecting the equilibrium equations involving {Fp} in eqns. (57) leads to
[M]{ql} + [C]{ql} + [K]{ql}=
{R}
(58)
where
\R] = lM][q2) + [C]{q2] + lK]{q2] + [M2]{p{ + [C2]{p) + [A 2 ]{pj
The vector {R} in eqn. (58) is a known quantity involving the relative
support motion {p} and the datum support motion in the vectors |p 2 }and
{q2}. Thus, eqn. (58) can be solved to give the relative structural
displacements, velocities and accelerations; {ql |, {qy} and {q\ } (see later).
Since strain energy and viscous damping energy are proportional to
differential displacements it might bethought that [K]{q2\ + [K2]{p2] and
[C]{g2J + [C2]i7'2J would be zero. However, this is only correct for the
linear directions XYZ, since the relative motion equations (54) do not
describe true differential displacements when rotations are present.
Differential displacements as described above are defined as being equal to
the amount of deformation in the structure. Equations (54) include, in the
rotational directions, deformation and a certain proportion of rigid body
motion.
The definitions in eqns. (54) present no difficulty in obtaining meaningful
answers. The absolute displacements are obtained by adding the datum
ground motion vector {q2} to the displacement vector {q} for each
calculation time. The forces and moments used to calculate the stresses in
62
K. FULLRD
the structure can then be found by use of [A ] {}, where \q\ = {qy\ + \q2\
and \p\ = \Pi\ + \p2\ If the fixing moments and reaction forces are not
needed, the forces and moments may be found by use of [A ] {q J 4 [A 2]{/?j.
As stated previously, true relative motions only occur in the absence of
rotational input motions. In this case the terms [K]{q2\ + [K2]'Kp2\ and
{ 2 } + [C2]{p2} are zero, giving the vector [R] as [M]{q2} +
[M2]{p) + [C 2 ]{,} + [A 2]{/>,}.
If the base mat. on which the structure rests, is stiff enough compared
with its environment to be considered as rigid, then the supports of the
structure all seethe same input and the vectors {pi j , {p1} and {p\ } are zero.
Thus, for linear directions and a rigid base mat {R} = [M]{q2} +
[M2]{p2}. If the elements through which the support motion is trans
mitted have a mass distribution such that their mass matrices are
diagonal, then [/ 2 ], which is not, in general, a square matrix, has all zero
elements.
With the above conditions the familiar seismic equations are formed
from eqn. (58):
[M]{q1} + [C]{ql} + [K]{q1} = W]{q2}
(59)
An alternative definition of the datum displacement can be used so that

the relative displacements are true, taking into account datum rotations as
well as translations. Thus, we have
{x} = {u} + {1}XQ+
{nfiMo {,,'
(60)
where /. mx are direction cosines of the vector O, a is the length of the

vector OI, / i s the location of the structure degree of freedom; and Ois the
location of the datum.This may be compared with eqns. (54). {>'}and {rjare
similarly related to the datum motions, while rotational relationships are
unchanged. Then {q} = { ? ] } + {q2}, where {<j,} are true relative
displacements and {q2} contains rotation terms. However, the need to
determine the rotational relationships is a deterrent to the use of true
relative displacements and the preceding definition of relative displace
ments in eqns. (55) and (56) is typically used.
When one support motion (X2, say) is being considered, eqn. (59) can be
written as
[M]{ql} + [C\{ql} + [K\{q1}=
[M]{I}J2(1)
(61)
where {I}x is a unit vector for displacements and null for the other
directions of support motion.
63
If we have calculated the eigenvalues and vectors of the structure on

earthed supports, then we can substitute for {q) and make the
transformations described in the section on modal methods to eqn. (61) to
give
[M*]{S] + [C*]{S] + [K*]{S] = -[A]J{I\xX2(t)
(62)
where
tl =- []\\
[A]- = [] []
[M*]--= []'[][]
[C*] == 2 [*] [] [c
[A*] = [][][] = [*][ 2 ]
Note that a viscous damping assumption has been made. Although it

would have been valid to use other orthogonal damping matrices, critical
damping ratios are often specified as those which must be used in seismic
analyses of various structures, particularly when these are nuclear power
station structures.
When eqn. (62) is premultiplied throughout by [M*]~ \ we have a series
of single degree of freedom equations:
S\ + 2coiSi + 2St = yx,xX2(t)
(63)
where {y}x = [M*\~ 1[A ]T{I}X is the modal participation vector.

When [M] is a diagonal mass matrix, the modal participation factor
()
>';..v = =
(64)
where subscript . indicates that the summation is restricted to the degrees of

freedom in the support motion direction.
Ignoring the modal participation factor, since it only has the effect of
scaling the calculated response, equations of the type of (63) are typically
solved for successive intervals defined by points on the X2 versus /
accelerogram. Thus, at the start of an interval, X2, time, S, S and S are
known and X2 and time are known at the end of the interval. As stated
previously, the acceleration is assumed to vary linearly between consecutive
time points. Rewriting eqn. (63):
S -f 2OKS
+ o/S
= - (a + b,/)
(65)
64
K. FULLRD
where
=
u
2('-)
X2(tn)X2(tn^)
/ - ,
t = /
and S(tn_),
S(tn_)
and S(tn_) are known from a previous solution at
',,-
The solution to eqn. (65) gives

S(t') = (C, sint' + C.costfc^1'
where
4h
C,
+ A + Bt'
,) + S ( t _ 1 ) + - | - - - ^ 6 I 1
(66)
+ ^
2
" +4-0
-b
"7
and = (1 2 ).
2
These can be simplified for computation purposes:

A = - (,, + 2)/ 2
C2 = ( ; _ , ) -
C, = ( ( / _ , ) + C . ) / / ;
Differentiating eqn. (66) twice yields S(t) and 5 ( 0 , and, specifically, we
obtain S(t), S(tn), S(tn) when t' is set to tn /_,.
Clearly, values of {7, (/)}, etc., are obtained just as easily when the other
support motions are considered, although for the multisupport case the
righthand side as in eqn. (58) does require more computations than when it
is as in eqn. (59). In the simplest case of irrotational, rigid base support
movement we can write the relative displacements as
{?.(')> = ( ( {i)
(67)
where yn is the participation factor for the z'th mode in the nth direction.
65
Sn,(/) is the response of a damped single degree of freedom system of

natural frequency ,, due to excitation in the nXh direction, at time /.
[-,] is the eigenvector for the /'th mode of the earthed structure.
Once the absolute displacements have been established, the structure
stresses can be derived from the system forces and moments. In the
particular case of the rigid base, with irrotational movement, it is possible to
FIG. 4. Acceleration-time history for Parkfield 5N05W, 27.6.1966.

assemble structure stresses from modal stress distributions in a manner
exactly analogous to that shown in eqn. (67).
The acceleration versus time points which define the accelerogram can
run into many hundreds (see, for example, Fig. 4) and the transient analysis
described above can be very costly to perform. For this reason the
technique known as the response spectrum method is widely used.
THE RESPONSE SPECTRUM M E T H O D OF SEISMIC

ANALYSIS
It has been shown that when an acceleration time history is applied to a
structure's support, the derivation of the transient response is simple in
principle. Once the absolute displacements are obtained, the stresses can
also be derived. However, a more economical approach is often used for at
least part of the design process and is the basis of one type of analysis
66
K. FULLRD
recommended by the US Nuclear Regulatory Commission (USNRC) for

nuclear plant. This is the response spectrum method.
If a singledegreeoffreedom system is subjected to a transient excitation
through its support, it is possible to establish the maximum response of the
mass during the transient. This response can be the peak absolute or relative
displacement, absolute or relative velocity or absolute or relative
acceleration. A s we are discussing a singledegreeoffreedom system, not
all of these values are relevant to establishing the forces on the system.
The equation of motion for a single degree of freedom in terms of
absolute displacement is
mx + c(x y) + k(x y) = 0
(68)
where is the displacement of the mass and y is the displacement of the

support.
Making the substitution y = u, the relative displacement, the above
equation becomes
mii + cu + ku = my
or
u + 2 + 2 = y
(69)
(69a)
assuming viscous damping. This is directly comparable with eqns. (59) and
(63) or (65) of the previous section, where the participation factor is unity. It
follows that, given (/), u(t) can be readily found for a range of
frequencies, , and a range of damping ratios, c. Hence, |w(/)max| can be
found. Although the maximum values of u(t) and u(t) + y(t) will occur at
different times from the time at which the maximum relative displacement
occurs, this is ignored for the purpose of plotting a response spectrum.
Clearly = (2 + co2u), but it is usual to define a spectral acceleration.
Sa, equal to CJ2SD, where SD is the maximum value of \u(t)\, and plot this
against frequency. Spectral velocity. Sr, is defined as being equal toc/;.S"Dand
so we have the opportunity to extract relative displacement, relative
velocity and absolute acceleration from one plot. The log velocity versus log
frequency graph is widely used for plotting response spectra.
When the procedure just outlined is followed, we have a set of response
spectra for a given support acceleration time history. This accelerogram
may be one extracted from an instrumental recording or it may be
artificially generated to meet some specific local conditions. It is quite
normal to input a ground-level accelerogram into a mathematical model of
a building and calculate the acceleration response at a number of locations
within the building. These 'floor' acceleration-time histories are in turn
67
1000
20
01
0.2
05
Frequency, cps
Fo. 5. Horizontal design response spectrascaled to 1 g horizontal ground

acceleration (USNRC 1.60).
used to give floor response spectra distinct from the ground response
spectra in that they reflect dynamic characteristics of the intervening
structure. Other artificially generated time histories may have a more
fundamental purpose in that they modify a given free-field time history to
take into account ground conditions different from those where the original
recording was made.
There is another type of artificial accelerogram which is not used to
generate a response spectrum but is created to fit an idealised response
spectrum reasonably closely. The response spectrum is generated from a
large number of distinct single-degree-of-freedom systems, and, strictly
speaking, is a series of points. Joining the points together results in a spiky
plot. The U S N R C have defined, for design purposes, smoothed ground
response spectra which, on a log velocity versus log frequency plot, are a
connected series of straight lines for different levels of damping, based on
the characteristics of a set of response spectra of actual earth tremor
recordings (Fig. 5). In order that floor response spectra may be obtained,
68
K. FULLRD
artifical ground time histories are generated which envelop the design
ground response spectrum.
Whatever the source of the response spectrum, the principle of its use is
simple. For a specific system, we can calculate its natural frequencies. For
each mode, a level of damping is assumed and a spectral value of response is
read off the corresponding response spectrum. This response has to be
scaled by the participation factor for the particular mode and excitation
direction, and the result is a modal response value for the structure,
5/ = i^D.i- This can be compared with the calculation of response in the
previous section. The major difference is that we no longer have a timevarying response, only a peak value. The effect of excitation duration is lost
and the phase differences between the responding modes are also lost.
The U S N R C spectra are defined for two horizontal and the vertical
directions. The three components are independent and the manner in which
the various modal responses are to be combined reflects this. The most
conservative method of combining the responses would be to simply add
them. What this means is that all the modal response values at a given
location in the system are added together. Ifall the modes are assumed in
phase for this summation, some will have to be out of phase to give a
maximum response at another place in the system. Summation is usually
only carried out for those modes whose natural frequencies are close
together (usually within 10% of the lowest frequency). When there are no
closely spaced modes, the responses are squared and added and a square
root of the total taken (square root sum of squares or SRSS method).
Response here is taken to be stresses, forces or displacements, rotation
effects being ignored. A fuller description of acceptable methods of
combining modal responses is given in the U S NRC Regulatory Guide 1.92,
Revision 1, published in February 1976.
Although the response spectrum method is easy to use, it should not be
forgotten that it is based on the response of single-degree-of-freedom
systems. The method should not be applied to multisupported structures
without a realisation that the relative support motions are being
approximated.
ACKNOWLEDGEMENTS
This paper is published by permission of the Central Electricity Generating
Board. The author wishes to acknowledge the advice and assistance of his
colleagues in the preparation of the material for the paper.
69
REFERENCES
1. Biggs, J. M. (1964). Introduction to Structural Dynamics, McGraw-Hill, New
York.
2. Craig, R. R. and Bampton, M. C. C. (1968). 'Coupling of substructures for
dynamic analysis', AI A A J I, 6, 1313-9.
3. Flax, A. H. (1975). 'Comment on "Reduction of structural frequency
equations'", AIAA J I, 13, 701-2.
4. Fricker, A. J. (1978). 'A mathematical analysis of the modal synthesis method
used in DYSAN2', CEGB report RD/L/N107/78.
5. Guyan, R. J. (1965). 'Reduction of stiffness and mass matrices', AIAA
Jl, 3, 380.
6. Henshell, R. D. and Ong, J. H. (1975). 'Automatic masters for eigenvalue
economisation', EESD, 3, 375-83.
7. Hurty, W. C. (1960). 'Vibrations of structural systems by component mode
synthesis'. Proc. ASCE, 86, No. EM4, 51-9.
8. Hurty, W. C.( 1965).'Dynamic analysis of structural systems using component
modes', AIAA Jl, 3, 678-85.
9. Irons, B. M. (1965). 'Structural eigenvalue problem: elimination of unwanted
variables', AIAA Jl, 3, 961-2.
10. Kidder, R. L. (1973). 'Reduction of structural frequency equations', AIAA Jl,
11, 892.
11. Kidder, R. L. (1975). 'Reply by author to A. H. Flax', AIAA Jl, 13, 702-3.
12. Wilkinson, J. H.(1965). The Algebraic Eigenvalue Problem, Oxford University
Press, London.
13. Wilson, E. L. and Penzien, J. (1972). 'Evaluation of orthogonal damping
matrices', Int. J. Num. Meth. Eng., 4, 5-10.
Transient Response by Time Integration: Review

of Implicit and Explicit Operators
S. W.
Sandia Laboratories,
1.
KEY
Albuquerque, New Mexico,
USA
INTRODUCTION
It is now an accepted practice to obtain the response of structural systems to

suddenly and rapidly applied loads by integrating forward in time. The
result is a prediction of acceleration, velocity, displacement, stress and
strain histories at nodes and elements in a finite element or finite difference
discrete model of the structural system. The procedures in use are capable
of predicting the response of linear as well as non-linear systems. The nonlinearities may arise from large deformations, finite strains and/or inelastic
material behaviour.
The purpose here is to review the principal time integration operators in
use and their characteristics in terms of stability, accuracy and damping.
We will start by examining a number of spatially discrete models to which
time integration operators will later be applied. These are idealised linear
problems, but serve well to make the points needed. Once the spatially
discrete systems in which we are interested are defined, the most popular
explicit and implicit time integration operators will be examined in some
detail. An added set of time integration operators will be studied with the
aid of graphs.
The real need for studying time integration operators in such detail
becomes evident when a choice is to be made as to which one is the best to
use with a given spatially discrete model. At this point the type of loading to
be applied becomes a factor. Impact loads require one kind of decision,
while seismic loads dictate another kind of decision.
Finally, some of the latest results in combined explicit-implicit operators
suitable for fluid structure interaction will be presented.
71
72
S. W. KEY
2.
SPA TIA LLY DISCRETE MODELS
The majority of today's structural response calculations are carried out

using finite element modelling. It is this form of spatially discrete modelling
which is integrated forward in time to generate a transient dynamic
response. It is appropriate to review these models in this elementary form in
order to understand later decisions on the choice of a time integration
operator.
FIG. 1. Onedimensional finite element mesh of piecewise linear axial

displacements.
The wave equation in one dimension characterises dynamic axial loads in
bars and beams, plane waves in two and threedimensional solids and
membrane waves in plates. It is given by
4"
ult=f(x,t)
(1)
Here u is the displacement in the direction of motion and and t denote

partial differentiation in space and time. The sound speed is cm, where c 2
equals Ejp for the uniaxial stress rod and beam. Ej[p(\ v2)] for the
membrane wave in a plate, and (1 v)/[p(l f v)(l 2v)] for the plane
wave in a solid. E is Young's modulus, v is Poisson's ratio, and is the
density.
As shown in Fig. 1, a piecewise linear finite element spatial descretisation
is introduced:
U(JC, 0 = uM)(\ )/2 + ii//)(] + f)/2
= ( Xj)/i + (x x,)/l,
I = Xj .v,
(2)
The local coordinate runs from 1 to + 1. For convenience, the length of

the element is written as /. The resulting element stiffness and mass matrices
are given in eqn. (3):
k =
1
1
m = I
md = I
o
I
(3)
TRANSIENT RESPONSE BY TIME INTEGRA TION
73
The stiffness matrix k and the consistent mass matrix w c are wellknown
results and are easily obtained by standard methods. The diagonal mass
matrix is the result of lumping at the nodal points.
For uniform meshing, the equation at any interior node/ is given by eqn.
(4):
(1 +rD)j-l-fj
c^2
Duj=jj
(4)
where the operator notation Du means u_ , 2uj + uj+ ,. The consistent

mass results are obtained with r = \ and the diagonal mass results are
obtained with r = 0.
There is, in fact, a continuous variation in mass representation as r varies
between and 0. The product solution in eqn. (5) is assumed for the
homogeneous equation:
uj(t) = SjT(t)
(5)
Separation of variables leads to the pair of expressions in eqns. (6):
DS. + ff U +*,-<>
(6)
+27"=0.
The first expression in eqns. (6) is satisfied by the discrete harmonic solution
in eqn. (7):
S j - C . s i n ^ + Cjcos^
(7)
For a free-free or fixed-fixed boundary condition, = nn, while for the freefixed case, = (2A? 1)/2, the same as in the continuous case. The positive
integer is less than or equal to N, the total number of elements making up
the mesh. Since the mode shapes are the same in both the discrete and
continuous cases, a frequency-by-frequency comparison can be made. The
continuous frequency is given by = kcJIN. The result is given by eqn. (8)
and shown in Fig. 2.
. /
= sin
2/V
- 1 4rsin'
2N\
2/V,
^' 2 "
(8)
This is the basic hyperbolic wave equation spatial discretisation which
74
S. W. KEY
underlies the bulk of the transient dynamic analyses carried out. The
piecewise quadratic displacement assumption is discussed by Belytschko
and Mullen. ' The piecewise cubic displacement assumption is discussed by
Krieg and Key. 5
In two space dimensions much of the same behaviour is obtained.
Goudreau 2 discusses Laplace's equation which governs the vibration of
1, Consistent mass, r - 1/6

2, Diagonal mass, r -
0 5
1 0
Mode = n/N ,(2n-
\)/N
FIG. 2. The ratio of approximate frequency, . to true frequency, . versus nondimensional wave number for a piecewise linear spatial approximation.
membranes. He considered a rectangular domain with a spatial
discretisation based on the classical five-point finite difference 'stencil' and a
bilinear displacement quadrilateral. If the membrane is fixed around the
perimeter and the mesh is elements by M elements, the results can be
displayed for fixed values of m/M. Figure 3 is the result.
The bending equation in one dimension characterises transverse loads in
beams and plates. It is given by
H=/(0
(9)
Here vv is the lateral displacement and and / are again differentiations in

space and time. The constant cb is given by (Eh2/\2pY 2 , where h is the
thickness of the beam or plate.
75
T R A N S I E N T RESPONSE BY TIME I N T E G R A T I O N
12
m/M = 7 / 8
= = = - . - . -
'^*
.___
m/M = 1/2
^f
",/'
m/M = 1 / 8 , . ' " ' '
,~^'
10
~":r=:^2r~^
m/M=\/%~~
/n/Af* 1/8
^.
m//
=3_
0 9==.
x.
^Q\
^ "^^ ^ "
x
\
\ \
\ \
x V/77//VM/2
0 8-
~
m/M
7{^^
.-"--^
07
06
- ^ . \ \
\\
\ \
\
/?// = 7/8
0 4
0
02
0 4
S,
ite difference
Fin
Q4 (consis t en t mass )
Q4 (lumped mass)
06
.
.
01
FIG. 3. The ratio of discrete frequency, , to exact frequency, , versus mode

number, n/N, for selected values of mode number m/M for both a finite difference
and finite element bilinear quadrilateral spatial discretisation of the membrane
equation.2
As shown in Fig. 4, a piecewise cubic displacement spatial discretisation
is introduced:
w(x.t) = M-,/),(,;) + wji2(ii) + /;,- +
wxhx(t])^
/',('/) = i('/ 3 - 3,, + 2).
lH(r) = {(2 2 + 1) >
/;,(,,) = luf
/,4(r,) = {(tf + 2 1)
3 2).
(10)
76
FIG. 4.
S. W. KEY
Piecewise cubic shape functions used in discretising the

displacement. , in the onedimensional bending equation.
lateral
T h e resulting element stiffness and mass matrices are given in eqn. (11):
.
12
V2
+ 2
'l2
'7
~T2
12
12
+ r-
+ /2
~7-
/
';
13/
III2
9/
35
210
I3
105
70
ill2
m,. =
210
9/
70
13/
"420
~42
13/ 2
P_
420
140
13/
420
~35
11/ 2
~2
13/
2
13/ 2 '
/
140
11/ 2
210
105
(H)
TRANSIENT RESPONSE BY TIME INTEGRATION

2-Or
\
\
\
\
\
\
77
1 \
1 5
\
\
\
\
\
\
10
\ \ \\
\ \ \\
0 5
1 , Diagonal mass a = 1
2, Consistent mass
3, Diagonal mass a = 17 5
Mode - n /N pin-pin, ...
FIG. 5. The ratio of discrete frequency, , to exact frequency, , versus mode

number.,/. for both a consistent mass and diagonal mass discretisation of the beam
equation using piecewise cubic displacement functions (pin-pin = n/; for other
boundary conditions see Hurty and Rubinstein,14 p. 203).
The stiffness matrix k and the consistent mass matrix mc are well-known
results and are easily obtained by standard methods. No rotary inertia is
included. A diagonal mass matrix employing a gradient inertia scaling
parameter is given in eqn. (12): 5
7/2
a/ J /420
(12)
1/2
3
a/ /420_
78
s. w.
KEY
Two discrete equations are obtained at each mesh point. The details of this
are covered by Goudreau and Taylor. 4 The discrete solution is given by
eqn. (13):
//)== C, sin ^ + C 2 cos-^ + C 3 s i n h ^ + Q c o s h ^ J T(t)
(13)
The mode shapes are once more the same as those in the continuous case, so
that a frequency-by-frequency comparison can be made. For the relation
between /. and mode number for various boundary conditions, Hurty and
Rubinstein 14 (p. 203), can be consulted. The results are contained in Fig. 5.
In shells the membrane and bending response is coupled, so that at best
one has modes which are predominantly membrane or bending in
behaviour with small components of the other. Nevertheless, the wave
equation and the beam equation represent in the limit of a flat plate the
membrane and bending behaviour of shells.
Figures 2, 3 and 5 tend to favour the consistent mass representation as
being more accurate. These are the results that would be obtained in a mode
shape and frequency evaluation. If only the lower modes are being sought,
there is no clear-cut preference except on the basis of computational effort,
and then a diagonal mass is better.
The important point to be made here is that these are the semidiscrete
equations which are to be discretised further when applying a time
integration operator to integrate forward in time. However, it is still the
original partial differential equations to which solutions are sought.
3.
TIME INTEGRATION OPERATORS
For the discussion here time integration methods are going to be divided
into explicit procedures and implicit procedures. For either the semidiscrete
wave equation, the semidiscrete bending equation or, for that matter, any
discrete model of structural behaviour, we can adopt the notation in eqn.
(14):
Ma + Kd = F(t)
(14)
Here d is the vector of nodal degrees of freedom; a is the vector of nodal

accelerations; M is a mass matrix which may or may not be diagonal; is
the linear stiffness matrix, although the restriction to linearity can be
relaxed under appropriate conditions; and F(t) is a vector of nodal 'loads'.
79
We will assume for the moment that we have the solution at time equal to
nAt so that eqn. (15) holds:
Ma" + Kd" = F"
(15)
+l
If the time integration procedure being used requires (Kd)"

to obtain
i/" + ', then a set of algebraic equations will have to be solved and the scheme
will be called implicit. If, on the other hand, (Kd)"+ ' is not required to
obtain 7" +1 , the scheme will be called explicit. This classification is
consistent with the finite difference literature, where only diagonal mass
matrices are considered.
Explicit Procedures
The most widely used explicit time integration procedure is the central
difference method. As is discussed below, it is the optimal explicit procedure
from a very wide class. First a central difference expression is introduced for
the acceleration a in terms of the velocity and then central difference
expressions are introduced for the velocity in terms of the nodal degree of
freedom. Thus.
yn + ti 2,
d"^
t.i,n
= d" +
2,
/"
\,tM\Kd"
+ 1/2)
F")~)
In practice, eqn. (16) is the form used where one old configuration d" and
one old velocity t"" (1 2 | are used to obtain a new velocity i?""*(1;2) and a new
configuration d"+ '. The velocity is carried at the halflime point. It is also
seen that were the stiffness calculation nonlinear owing to either geometric
changes which had occurred or material nonlinearities, there would be no
trouble in computing K(d")d". To take advantage of the speed with which
each time step can be computed M must be diagonal, so that M~ ' in eqn.
(16) is trivial.
The difference method is only conditionally stable, which means that
there is a critical time step A icnt below which A t must be kept. When A t is
above the critical time step, highfrequency modes in the numerical solution
become exponential in time rather than harmonic solutions of the original
semidiscrete system (14) A fter a number of time steps the exponential
solutions dominate.
A stability analysis for the central difference method is straightforward.
First the velocity is eliminated from eqn. ( 16) to give the homogeneous eqn.
(17):
M _ 2d" + dn' ' + A t2M~xKdn
=0
(17)
80
S. W. KEY
A homogeneous solution of the form given by eqn. (18) is sought.

d" = de1'"*'
(18)
Here is undetermined, is an arbitrary displacement shape and Ar is the

time step. Substituting the trial solution (18) into the central difference
method (17) provides the eigenvalue problem (19):
(e"Al - 2 + ->" + ArM~^K)d
=0
(19)
For each eigenvalue-eigenvector pair there is a separate . Denoting any

given eigenvalue of l by /., the appropriate polynomial for is given
by eqn. (20):
e/M' - 2 + - , ' + At2?. = 0
(20)
Since the motion in eqn. ( 14) is bounded for all time. |/'| must be less than
1. This leads to the requirement that / 2 / must be less than or equal to 4.
Since the maximum eigenvalue of M ~ ' provides the strictist condition,
the limits on At become
/2<^
(21)
'-max
This approach to the stability conditions for the central difference method
is discussed by Leech, Hsu and Mack.'
For the linear wave equation with a piecewise linear displacement finite
element mesh of uniform dimension /. the maximum eigenvalue is given by
c2
= 4-.
lumped mass
= 6-r?.
consistent mass
For the linear bending equation with a piecewise cubic displacement finite
element mesh of uniform dimension /, the maximum eigenvalue is given by
2
max = 2 5 2 0 ^ .
lumped mass 0 < a < 525

(23)
= 2520 .
4'
consistent mass
In a search for an unconditionally stable explicit time integration

operator Krieg 6 examined all schemes of the form of eqn. (24).
d"+ ' = a0d" + a1d"1 + a2d"~2 + b0 Kd" + b, K" " 1 + b2 K" ~2
(24)
81
The schemes of this form include all explicit two-root and three-root
methods. (The three-root methods contain an extraneous solution about
which little is known in general.) The central difference method is a two-root
method.
It was found that there did not exist a scheme which was unconditionally
stable and of order (?2) when compared with the differential equation (14).
In addition, there exists no time integration operator of the form of eqn.
(24) which is of order (/ 2 ) when compared with the differential equation
(14) which is stable for a time step longer than that of the central difference
method. These are very strong results, and point to the optimal nature of
the central difference method in this class.
Implicit Procedures
There are a great variety of implicit schemes; however, the Newmark
method with the explicit central difference method as a special case is the
most popular method in use today and is examined first. While in practice
the full set of semidiscrete equations (14) is integrated together, the
integration may be uncoupled into modes. The model uncoupling leads to
the second expression of eqns. (6) being integrated mode-by-mode. The
Newmark method applied to eqns. (6), where 2 is an eigenvalue of
M ~ ' of eqn. ( 14) gives
DT" + c2A t2l\<\ b f 1 ) + y ( l E~1) + D]T" = 0
(25)
The superscript is used to denote the value of T(t) at time step (nA t) and
the operator i s defined as usual where Epuk = up + k. The central difference
operator in time is then defined as D = (E 1 )( 1 E'1). The method has
two parameters: and y. A s these two parameters vary, all possible two
root methods are obtained and the parameter is tied closely to the
damping in the scheme. The scheme contains no damping when y = \. It is
for these two reasons that the method has gained such popularity.
Equation (25) has a damped sinusoidal solution, but it is useful to
transform with a onetoone and onto transformation to the complex z
plane as follows:
T" = C 5 exp [(a + ico)nA t] = C 5 ( ^ J
(26)
The resulting transformed stability polynomial is then found by

substituting eqn. (26) into eqn. (25) and rearranging in decreasing powers of
:
r 2 [4 4 322(4 - 2v)] + 2 2 (27 - 1) + 2At2 = 0
(27)
82
S. W. KEY
The solution is said to be stable if it is bounded. Since a power solution was

assumed in eqn. (26), this means that if the magnitude of (1 + z)/(I z) is
no greater than 1, then its powers are no greater than 1 and the solution is
boundedthat is, it is stable. In terms ofr, this means that it cannot have a
positive real part. The quadratic has no roots in the right half of the
complex plane if all the coefficients are of the same signthat is, the
solution is stable if
4 + 22(4 - 2y) > Oj
(28)
2y- 1 >0|
Equations (28) are the usual stability results as obtained by Nickell 3 and
others. The damping parameter a and frequency may be obtained from
the roots of eqn. (27). The roots of eqn. (27) are
= + iq = cTjAt
f jo>A/(2y - 1 ) /[4 4- 02(4 - 2y) - jco2At2(2y - 1 ) 2 ] '

4 + 22(4 - 2)
= arc tan + arc tan
1
1 +P
= I log
"d + +
(l-pf+q2
(29)
Equation (29) may be used to obtain / as a function of coAt. The

frequency response for y = j and the damping for = \ are plotted in Fig.
6.
Of particular interest is the unconditionally stable Newmark method
with 2/J > > j . (One generally sees > j cited as the unconditionally
stable case. Implicit in this remark is = y.) Since a set of algebraic
equations must be solved at each time step, it is desirable to have At as large
as possible. This can only occur in the unconditionally stable case.
However, . must still be selected based on accuracy considerations. As can
be seen in Fig. 6, if the loading function can be characterised by frequencies
below ccul.on, then 10% accuracy is achieved with A t < I2/tcul.0lT.
The application of the Newmark method as well as any implicit method
to nonlinear analyses requires some care. Belytschko and Schoeberle. 7
Hughes, 8 and Hughes, Caughey and Liu 11 go into this problem in some
detail. Hereonly the major computational requirement pointed out in these
references is addressed. A s a result of nonlinear material behaviour and/or
large deformations eqn. ( 14) can no longer be written as a constant stiffness
83

13r
1-2
Central
/
difference /
\1313
--0
^ =1/12
= l/2
I io
/3=1/6
09
/3=1/4\
08
to5
S 10
y = 0 501
= 1/4
- 10;
~-0 51
%" I 102
02
05
10
20
50
FIG. 6. The frequency response for y = 05 and damping for = 025 for the
Newmark time integrator.
matrix times the configuration d, but rather as a vector L(<7) representing
the divergence of the stress at the present time in the present geometry. With
this change in notation eqn. (14) becomes eqn. (30):
Ma + L(d) = F(t)
(30)
For the nonlinear problem Newmark's method takes the form in eqns.
(31):
d"+l
t
,+l
= d" + A lv" + A t2[(\

=
vn
lA!{an
fl(1-rl)
)" + + ' ]
(31)
The manner in which a"+ l enters these equations for 0 means that an
'estimate' of <7" + ' to use in eqn. (30) for a"4 ' must be obtained. More
84
S. W . KEY
^ j ^ r ^ r ,
XX.
09
08
=05,y3=-y2=0t
xNv.Farhoomand
\ \
NNwHoubolt
Wilson/ \ \
\ \
07-
0 6
\ \
\\
FIG. 7. Frequency response and damping for the Houbok, Wilson and
WilsonFarhoomand time integrators along with a generalised threeroot method
of Krieg.
precisely, the first expression in eqns. (31) contains d"+ l on both the right
and left sides with the solution of a nonlinear set of equations required to
find d"+ '. A t every step along the way, because of the nonlinearities, the
solution to eqn. (30) is found only approximately. The resulting error e is
computed in eqn. (32):
e" = F" L(d")
Ma"
(32)
To discuss the unconditional stability of the Newmark method, energy

considerations need to be introduced. A discrete internal energy is given in
eqn. (33):
t/'=0;
U"+i = U" + jAd1(L" + L" + 1 )
(33)
A discrete external work is given in eqn. (34):

W1 = 0;
W"+ ' = W + Mi/ T (F" + F,+ ')
(34)
85
007
0 01 -
0 5
010
0 09
/Houbolt
/method
/Wilson
method! =1.4)
008
007
006
:
0 05 004
003
/Nev/
algorithm
(a = - 0 - 3 )
002
0 01 -
Sî
'New algorithm
(a = - 0 0 5 )
1
0 3
0-4
1
0-2
//
D amping ratio versus At for the new Hilber- Hughes-Taylor. Newmark
. Houbolt and Wilson time integrators (/ = (/)).
0
FIG. 8.
/
01
86
S. W. KEY
04
FiG. 9.
Frequency response versus a>At for the new Hilber-Hughes-Taylor.

Houbolt and Wilson time integrators ( = (/)).
The kinetic energy is given in eqn. (35):

E" =
"'Mr"
(35)
Assuming that the material stress-strain relation is stable and that a

bounded amount of external work W is supplied, the principal
computational requirement is to keep the error energy within the bound in
eqn. (36).
\\AdT(e"+l
+e'
< At(K" + U")
(36)
Equation (36) in essence prescribes an iteration tolerance that must be

maintained in solving the non-linear algebraic equations (31 ) for each time
step. Satisfying eqn. (36) is the first step in maintaining unconditional
stability in the Newmark method for 2/J > > \.
Without discussing the computational algorithms the behaviour of the
Wilson, Wilson-Farhoomand and Houbolt methods is presented in Fig. 7.
Also included in Fig. 7 is a generalised three-root method of Krieg and
Key. s All four of these time integration schemes are unconditionally stable,
but with greater frequency distortion than the Newmark method. All four
87
of these schemes possess large amounts of damping which cannot be

controlled independent of time step as in the Newmark method.
As a result of some rather astute observations by Hilber, Hughes and
Taylor, 9 they proposed a new time integration scheme which has vanishing
damping for small values of coAt. The damping characteristics are shown in
Fig. 8 and the frequency distortion in Fig. 9. It is an unconditionally stable
scheme so long as the single parameter is kept between \ and 0. It has
two additional parameters which are the same as the Newmark method
parameters and which are selected according to eqns. (37):
= (1 ) 2 /4]
UU j
(37)
The Hilber. Hughes and Taylor algorithm is a logical replacement for

previous procedures used in transient seismic analyses.
4.
SELECTION OF TIME INTEGRA TORS
A wide variety of time integration schemes have been introduced with very
little comment on their relative merits. It was seen that, depending on the
mass representation, the original frequency was distorted either upward
or downward into 3. Depending on the choice of time integrations, the
semidiscrete frequency c was further distorted upward or downward into
. Since it is not practical to carry out calculations at vanishingly small time
steps to minimise these distortions, they are necessarily present when an
analysis is carried out. The goal here is to examine selection of both a spatial
and temporal descretisation based on their combined effect and the class of
applications under consideration.
The first observation is that, depending on the choice of mass
representation, / is either greater than or less than 1. Depending on the
choice of time integration scheme, / is either greater than or less than 1.
However, our goal in a computation is to have / as close to 1 as
practicable.
The second observation is that, depending on the application, one is
interested in all of the frequency content of the model or only a portion at
the low-frequency end of the spectrum. In an impact analysis where stress
waves emanate from the place of impact and propagate throughout the
structure, it is the entire frequency content which is needed to represent the
rapid changes in stress that occur in time as the stress waves propagate.
However, in seismic analyses the frequency content of the acceleration
88
s. w.
KEY
pulse is quite low compared with the entire spectrum of the spatial
discretisation. The highest modes in the structure are only minimally
excited and then rapidly damp out, not to participate in the overall
structural response. If one is required to carry out a calculation including
them, needless effort is expended.
While all transient dynamic analyses do not fall into either the impact or
seismic class, it is our experience that these areas are sufficiently important
to base criteria upon them. On the basis of the reasons which are listed the
following recommendations are made:
Impact
Spatial discretisationlumped mass
Temporal discretisationexplicit central difference
All frequencies are important, 0-9 < / < 1 1 . (The lumped mass
depresses the frequencies, while the central difference operator distorts
the frequencies upward, thereby offsetting each other.)
Simple and fast computationally.
No numerical dissipation but artificial viscosity added to control
shock wave formation. Controlled independent of time step.
Conditionally stable, Atcm.dX < 2-0. Accuracy is maximised by
operating as close to the stability limit as possible.
Seismic
Spatial discretisationconsistent mass
Temporal discretisationimplicit Hilber-Hughes-Taylor
Unconditionally stable when applied to linear problems.
Possess numerical dissipation which can be controlled by a parameter
other than the time step. In particular, zero dissipation is possible.
Numerical dissipation does not affect the lower modes too strongly.
All frequencies for which it is required that 0-9 < / < 1 ;
Ao)culK)ff < 2.
(The consistent mass distorts the frequencies upward, while the
Hilber-Hughes-Taylor method depresses the frequencies, thereby
offsetting each other.)
If these two procedures are used for all classes of transient dynamic
calculations, it is useful to have a criterion which indicates which method
will require the least effort. A square two-dimensional mesh is assumed. The
work required to advance the solution one time step using a lumped mass
matrix and the central difference involves a matrix times a vector for each
time step; only the non-zero terms are counted.
89
For the implicit scheme the matrix is assumed to be symmetric and

banded. Gaussian elimination is assumed for the solution of equations at
each time step. The matrix is used in triangularised form, so that each time
step only requires the processing of a single vector. In this case the implicit
schemes require four times as much work per time step. In nonlinear
problems the difference is greater. Thus, when comil,1/cucu,.ofris greater than 4,
the implicit procedure will require less work. When jmax/a>cul.ofris less than 4,
the explicit procedure will require less work.
Several bad combinations are worth noting, because they each distort the
frequency spectrum the same way:
Central difference methodconsistent mass
Newmark's method, = \lumped mass
Newmark's method, = consistent mass
In the last example the choices of spatial and temporal discretisations are
individually optimal choices but each raises all of the frequencies. The
combination is certainly not optimal.
5. A
COMBINED EXPLICIT IMPLICIT OPERA TOR
The fluidstructure interaction problem presents a unique difficulty when

the response to transient dynamic inputs is to be calculated. Relative to the
dimensions of the fluid, the structure is thin yet its gross structural motion is
an important part of the interaction with the fluid. The stress wave
propagation through the thickness of the structure is a minor part of the
interaction. In the fluid it is the movement of stress waves that is of interest.
Because of the relative physical dimensions, meshes in the structure can be
To times the size of those in the fluid. If an explicit operator is used to capture
the response of the fluid, the time step will be controlled by the solid and will
be an order of magnitude smaller than ideal for the fluid. If an implicit
operator is used so that the time step can be increased to one suitable for the
fluid, needless equation solving is taking place in the fluid domain.
The best of all worlds would be to have an operator which had the
characteristics of an unconditionally stable implict scheme for the solid
elements where only the lowfrequency modes are of importance and which
had the characteristics of an explicit scheme for the fluid elements where the
high frequencies are important. Many multimaterial problems have these
same characteristics. Hughes and L i u 1 2 1 3 have addressed this problem and
90
S. W . KEY
developed an operator with features just enumerated. Because of the clarity

of their work, it will be used extensively in the following.
The finite elements of the model are assumed to be divided into two
groups: the implicit group, denoted by I, and the explicit group, denoted by
E. The algorithm used is eqn. (38).
(M1 + Mh)a"+ ' + C'v"+ ' + C^v"^
+ 7\V ,+ ' + Klld"+ ' = F'+ ' ^
d"+ ' = d" + A tv" + 4/ 2 (1 2)"

+> = " + A t(l y)a"
d"+1 =d"+x
)> (38)
A
+ t2a"+l
c" + 1 = 0" +1 + A r,a"+1

Here d", " and a" are the displacement, velocity and acceleration,
respectively, at (tiA t). The free parameters and y govern the accuracy and
the stability. M L is assumed to be diagonal. The algorithm is derived from
the Newmark method. When all the elements are members of the implicit
group, the scheme reduces to the Newmark family.
When C = 0, unconditional stability of the implicit element group is
achieved with
7>i
= (7 + i)2/4(
(39)
The maximum stable time step is determined from a 'Courant condition'

3/<((2 + 2 7 ) 1 2 - ) / 7
(40)
which must be satisfied for each frequency and corresponding modal

viscous damping factor , in the explicit element group.
In implementing the method one arrives at the point where, owing to the
implicit part of the problem, eqn. (41) is obtained.
K*d"+i = F * " + 1
(41)
where
K* = (K*)1 + (K*f
...1 +-!
y C' 4- A'1
(K*y = At2 TV/
A t
(K*f =
1
h
At M
2
(42)
Thus, a diagonal system is obtained where the elements are explicit and a

2
10
12
11
91
\
Node number
I - Implicit elements
E - Explicit elements
Finite element m e s h - 2 equations per node
Equation
number
Shaded region
represents
non-zero entries
Symmetric
Profile of K
FIG. 10. Twodimensional, implicitexplicit finite element mesh and profile of

coefficient matrix.
symmetric banded set of equations where the elements are implicit. Figure
10 shows a typical circumstance. With a "profile' or 'skyline' solver
considerable computational effort is eliminated.
Of interest are the computational results from this scheme. Consider the
axial impact of a bar 10 units long with a uniform velocity ;0 equal to 1 0.
The impacting end is held fixed. The bar is composed of 20 elements 05
units long, a modulus equal to 100 and a density of 001. The critical
time is 0005, the element barwave transit time. Figure 11 shows the axial
stress history at the impacting end for various combinations of time step,
92
S. W. KEY
- 1-2
oo
12
=0 0
/ = 0 0 0 4 9
=05
Exact
= 00001
? = 0 25
0 8
0 4 -
00
-04-
-08
\'^
-12
FIG.
00
0 1
02
0-3
04
05
Twenty element, bar impact problem.13 Comparison of predictor

corrector algorithm with exact solution.
93
and y. For 7 = 05 no damping is present. For / = 0005 and = 00001,

the scheme should behave like the central difference method operating at
the critical time step which it does. For A t = 0005 and = 025, the value
of of most interest since implicit elements are unconditionally stable, the
results are still exceptionally good. For / = 00049, just below the critical
time step, and = 00001 and 025, the results pick up a little oscillation;
however, increasing reduces the oscillations.
I 2
Va
* = 00
/ = 0 00353
y=0 5
0 8
= 025
{ A l l elements have the
same material properties)
04
00
-04
Exact
Explicit
Implicit-Explicit
(Elements 1,11 and 21
are implicit )
1 6
01
0 2
O 3
0 4
FIG. 12. Twentyone element, bar impact problem.13 Comparison of

predictorcorrector algorithm, implicitexplicit algorithm and exact solution.
Consider the same bar problem, but now 105 units long and 21 elements
in the mesh. Elements 1,11 and 21 are made implicit, the first, middle and
last elements. ForAr = 0003 53, well below the critical time step,, the results
in Fig. 12 are obtained. As should occur, the results are very similar with or
without the implicit designation.
For the final example, let E = IO7 in elements 1 and 21. The critical time
step for elements 1 and 21 diminishes to about 1 /136th of the critical step
for elements 2 to 20. Figure 13 shows the axial stress history in the middle
of the bar. The time step used is governed by the explicit elements.
These results point to a very effective operator. In addition, the technique
for deriving the operator can be applied to other implicit schemes.
94
S. W. KEY
u
[At
I
1
08-
'
04-
\%
[
1
1
-08
'Stiff' elements
1 and 2 are implicit)
j 1
mphcit Explicit
i
if
11
-04
r~~
0 0045
06
0 3025
11
il
0 0
5 25
00050
0 5
025
1 2-
\
I \
\\
-1 2 --
\
-1 6
00
01
02
1
03
1
04
05
FIG. 13. Twenty-one element, bar impact problem. Comparison of damped and
undamped, implicit-explicit computations.
REFERENCES
This list of references is not intended to be complete. Rather, they have been
selected so that an individual looking to gain a definitive understanding of
transient time integration operators can confine his study to a small crosssection of the literature. This is not to say that there are not other articles
and books which contain this information. The bulk of these articles deal
with the analysis of time integration operators rather than the results which
they produce in applications. In this area it is all to easy to make erroneous
judgements based on applications alone. A study of these references will
reveal a heavy reliance on them in these chapters, particularly the work of
Thomas J. R. Hughes and his colleagues. Our primary goal here is
instructional rather than exhaustive inclusiveness.
1. Leech. J. W.. Hsu. P. T. and Mack, E. W. ( 1965). 'Stability of a finite-difference
method for solving matrix equations'. AI AA J!. 3(11). 2172-3.
95
2. Goudreau. G. L. (1970).'Evaluation of discrete methods for the linear dynamic

response of elastic and viscoelastic solids'. Report 69-15, Department of Civil
Engineering, University of California, Berkeley.
3. Nickell. R. E. (1971). 'On the stability of approximation operators in problems
of structural dynamics", int. J. Solids Structures. 7, 301-19.
4. Goudreau. G. L. and Taylor. R. L. (1973). 'Evaluation of numerical integration
methods in elastodynamics'. Comp. Meth. Appi. Mech. Eng., 2(1), 69-97.
5. Krieg. R. D. and Key, S. W. (1973).'Transient shell response by numerical time
integration'. Int. J. Num. Meth. Eng.. 7, 273-86.
6. Krieg. R. D. (1973). 'Unconditional stability in numerical time integration
methods". J. Appi. Mech.. 40, 417-21.
7. Belytschko.T. and Schoeberle. D. F. (1975).'On the unconditional stability of
an implicit algorithm for nonlinear structural dynamics'. J. Appi. Mech., 42,
865-9.
8. Hughes. T. J. R. (1976). 'A note on the stability of Newmark's algorithm in
nonlinear structural dynamics'. Int. J. Num. Meth. Eng.. 11(2), 383-6.
9. Hilber. H. M.. Hughes, T. J. R. and Taylor. R. M. (1977).'Improved numerical
dissipation for time integration algorithms in structural dynamics', Earthquake
Engr. Struc. Dyns., 5, 283-92.
10. Belytschko.T. and Mullen. R. (1977).'On dispersive properties of finite element
solutions', IUTAM Symposium on Modern Problems in Elastic Wave
Propagation'. Northwestern University, Evanston. Illinois.
11. Hughes. T. J. R.. Caughey. T. K. and Liu. W. K. (1978). 'Finite-element
methods for nonlinear elastodynamics which conserve energy', J. Appi. Mech.,
45, 366-70.
12. Hughes, T. J. R. and Liu. W. K. 'Implicit-explicit finite elements in transient
analysis: stability theory", J. Appi. Mech.. 45, 371-4.
13. Hughes. T. J. R. and Liu. W. K. (1978). 'Implicit-explicit finite elements in
transient analysis: implementation and numerical examples', J. Appi. Mech.,
45, 375-8.
14. Hurty, W. C. and Rubinstein. M. F. (1964). Dynamics of Structures, PrenticeHall, Englewood Cliffs, N.J.
Explicit Time Integration of Structure

Mechanical Systems
T. BELYTSCHKO
The Technological Institute, Northwestern

Illinois, USA
1.
University, Evans t on,
INTRODUCTION
In most numerical solutions of transient structural or continuum problems

the governing partial differential equations are first discretised in space.
This procedure is called a semidiscretisation and yields a set of ordinary
differential equations in time. If one considers this process in the domain of
space-time, the integration then is along lines perpendicular to the space
subdomain, and therefore these methods are often called methods of lines
by mathematicians. This is in contrast to the methods of characteristics,
where the integration is carried out along lines in space-time which coincide
with possible discontinuous surfaces.
There are two basic types of methods for integrating the ordinary
differential equations: explicit and implicit. In explicit methods difference
equations are used that permit the displacements at the next time step to be
found in terms of the accelerations and displacements at the previous time
step, so that the procedure does not involve the solution of any equations.
In implicit methods the difference equations for the displacements at the
next time step involve the accelerations at the next time step, so the
determination of the displacements involves the solution of a system of
equations. For these reasons implicit methods usually require considerably
more computational effort per time step as compared with explicit
methods. On the other hand, in implicit methods the time step is only
restricted in size by accuracy requirements, while in explicit methods the
time step must often be very small to ensure numerical stability. 11 should be
added that explicit methods are far easier to programme and can more
easily account for complex material models and discontinuous phenomena
such as contact-impact and change of phase. Either method is completely
97
98
. BELYTSCHKO
unsuitable for certain types of problems, so the reliance on either method is

unjustified; more will be said on this later.
These notes deal with three aspects of explicit time integration
techniques: implementation, numerical stability and computational
efficiency, particularly in relation to engineering problems such as found in
safety analysis.
2.
DEVELOPMENT OF BA SIC EQUA TIONS
For the purpose of providing a framework for our discussion of the

integration of the equations of motion, we will first define the variables and
basic notation, and present the basic forms of the equations and their
derivations. In these discussions we shall use the formalism of the finite
element method. Many finite difference methods (especially those
applicable to arbitrary meshes) are basically very similar and both can be
viewed as solutions of the weak form of the governing partial differential
equations. In both cases the dependent variables are defined in elements or
zones by a spatial approximation and a weak form of the PDE is used to
obtain ordinary differential equations (ODE) in time for the discrete
dependent variables. A lthough there are substantial differences in the
mathematical accoutrements, in many cases the resulting discrete
equations for the finite element and finite difference methods are identical.
We will use the following notation:
{ j
[ ]
",
"/
It'
[d]
\D}
[L]
iP W
y. {J
,, {"
b{b)
denotes a column matrix

denotes a rectangular or square matrix
lower case subscript denotes components of a tensor
upper case subscripts denote node numbers
lower case superscripts denote a time step
denotes a time derivative
denotes transpose of a matrix
denotes nodal displacements of an element
denotes complete nodal displacement matrix
denotes connectivity matrix, so that {<f\ = IL*]{D\
denotes stress, where the latter designates a column matrix
arrangement of the stresses
denotes strain
denotes displacement
denotes body force
TIME INTEGRA TION OF STRUCTUREMECHA NICA L SYSTEMS
99
We distinguish w,(x), which will generally be the Cartesian components

of the displacements, ux, uy and t/_, from \d\, which may include derivatives
of u with respect to x,y or r.
If we consider a mechanical or structural system, in the absence of
thermomechanical coupling the governing equations are the momentum
equation
Sel(a,j) = puk
(1)
l=fMu><ru)
(2)
the constitutive equation
and the strain displacement equations

,j = J?,>f)
(3)
v
Here <\ and Jz?,^ are partial differential operators and designates an
appropriate frame invariant rate. For the particular case of a Cauchy stress,
velocitystrain continuum description, the above equations may be written
in indiciai notation, respectively, as
O,J,J
+ b, = pu,
=,)
. U:
(5)
CU:\
' = -
,
+ ôx, y
J
(4)
(6)
This system of partial differential equations is hyperbolic in most cases;

however, when internal constraints, such as the Kirchhoff hypothesis in
plate theory or incompressibility assumptions, are introduced, the
equations often become parabolic.
The first assumption in discretising the momentum equations, equations
of motion (1), in finite difference and finite element schemes is that the
functional dependence of the displacements in space and time can be
separated, so that the displacement fields in each element can be described
as products of shape functions () that are independent of time and nodal
displacements \d(t)\ which are independent of and incorporate the time
dependence:
{u(x,t)} = W(x)]{d(t)}
(7)
This is in fact a local separation of variables, and the implications of this

procedure are significant if wave propagation types of problems are to be
solved.
100
T. BELYTSCHKO
The stresses and strains in an element, {} and {}, may be functions of

both space and time. By means of the strain displacement equations, a
matrix IB] may be derived such that
{} = lB]{d}
(8)
We will not at this point restrict our discussions to a particular strain

measure ; the development will be general in the sense that it is applicable to
any linear or nonlinear strain or stress. If the material is linear, we also have
a stressstrain matrix [C] which relates the stresses and strains as follows:
\\ = [C] []
(9)
We will here develop the discrete equations from the principle of the
virtual rate of work, which corresponds to a special case of the Galerkin
method or a weak form of the partial differential equation. The principle of
virtual work states that for any arbitrary kinematically consistent
displacements, the rate of internal work must equal the rate of external
work
Wm = WZM
(10)
The term 'kinematically consistent' is equivalent to what is generally

referred to in strength of materials as compatibility, and requires that the
displacement field be continuous and satisfy all geometric boundary
conditions.
The rate of external work is given by
W" =
\\J\T]dS
<5wT{6'idF
(11)
where 7"j is a surface traction and \b' \ isa body force. For the purpose of
applying the principle of virtual work to dynamics problems, we introduce
inertial body forces through the d'Alembert principle, so that the total body
force is given by
W\ = {b\p{\
(12)
The rate of internal virtual work is given by

WM =
{\{}
(13)
101
By using eqns. (8)(13) we can then go through the following series of

equations:
{)>r = TW] 7 f
Wm =
[]{)
[B]T\a}dV
= {<5>}T{F}in'
(14)
(15)
where
F'"'=y[Lf T {/< e >
[B]J{a}dV
{/n
<T{T\dS +
OW"'
\\T({b}p{u})dV
Yury
{}1
mJ\T\ds
[]}-
(16)
pWYWdV[LM]{D
(17)
I y iti
\D\J{Fext}
where
{'}
V[Z,<e)]T
W1{T}dS
.ye,
[M] = \
(18)
{DYIM]{D}
\)
+
1/(C)
[L(c)]T[m(e)][/Jc)]
(19)
Here lmM] is the mass matrix of the element and [A/] the total mass matrix.
Combining eqns. (10), (15) and (18), we obtain
{D}J{Fr"
= {(]\
>DJ}{F'"})
(20)
102
T. BELYTSCHKO
By invoking the arbitrariness of the virtual velocities of D, we obtain

[M]{D] + {F""} = {Fe"}
(21)
which are the governing equations for the discrete mesh. This is a system of
second-order ordinary differential equations, so the semidiscretisation has
allowed us to replace the exact PDE system by an approximate ODE
system.
One important feature of this development in terms of the finite element
formalism, when compared with, for example, a finite difference format, is
that the effect of each element is isolated in terms of the generalised nodal
forces {}. The introduction of the generalised nodal forces {} imparts a
tremendous versatility to the finite element format. The nodal forces {J} are
simply summed for all elements in a fashion consistent with the topology
that is, the node numbering, regardless of the nature of the elements. Thus,
higher-order elements and structural elements may be mixed arbitrarily
with lower-order elements without changing the character of the equations
of motion. This is particularly important for fluid-structure problems,
where considerable difficulty is often encountered in the logic of finite
difference programs.
We will now specialise the equations for a case of linear materials where
eqn. (9) may be applied. Substituting eqn. (9) into eqn. (16), we obtain the
following set of equations:
{/ , e ) }= f
Ji-io
[BVlC][B]dV{cf^\
(22)
so
{F""} = Y[L , e , ] T {/>}
{Em} = V
or
[Z.,C)]T[A-,C,][/.<C|]{>}
{Fm} = [K]{D]
(23)
where [A] is the total assembled stiffness matrix. Note that the assembly of
[K] and [M] is identical.
Substituting eqn. (23) into eqn. (21), we obtain
IM]{D} + IK]{D} = {F}
a system of linear, second-order ODE.
(24)
103
In developing the equations of motion in a consistent fashion from the

principle of virtual work, the mass matrix, IM], which is obtained is a non
diagonal matrix. Like the stiffness matrix, it is sparse and symmetric, and in
most regular meshes usually banded. However, its inversions still usually
present a formidable task, and although its inverse maintains the same
banded characteristics as the matrix itself, it is no longer sparse. For these
reasons the inertial properties of the mesh are often approximated by a
diagonal matrix, known as a lumped mass matrix. A lthough Fujii 8 has
presented a formal procedure for developing lumped mass matrices, in
most cases lumped masses are obtained by heuristic means. This is a simple
procedure for loworder elements (i.e. the fournode isoparametric, two
dimensional elements, or the twonode onedimensional element), but at
present there are no consistent ways for lumping mass matrices for higher
order isoparametric elements or structural elements. This problem has been
discussed recently by numerous authors (e.g. Krieg and Key 12 ).
3.
CENTRA L DIFFERENCE TIME INTEGRA TION
The central difference formulas are

{D(t + A t)\ = D(t) + A t {D(t + JA /)J
\D(t + \A i\ = {D(t Wil
+ A t>D(t)\
(25)
(26)
Alternatively, if we let superscripts indicate the time step, we can write these
as
{>i+1 } = { ' } + A t{D\]
(27)
\D';l\+A t{D'\'1
(28)
\D\}=
where
\D\} = \D(nA t + \A t)}
(29)
Writing eqn. (27) for time step / and i 1, taking the difference, and
combining with the velocity expression, eqn. (28), we obtain
0']
= ^2(\D'^\2\D'\
+ {)'"'})
(30)
This difference formula is called explicit because this expression does not
involve any drivtes at the last time step.
If we write eqn. (21) at.time i and combine with eqn. (30), we obtain
D , + 1 ] = / 2 [ ] ~ ' ( { ^ " } {FinU}) + 2{D'} I D ' 1 j
(31)
104
T. BELYTSCHKO
It can be seen from the above that time integration by the explicit central
difference method does not require the solution of any equations provided
that the mass matrix is diagonal. Krieg and Key 12 have shown that, in fact,
the spectral errors of central difference integration and mass lumping are
compensatory, so that mass lumping is preferable in explicit integration
both from the viewpoint of computational efficiency, in that no matrix
inversion is involved, and in accuracy.
When a linear viscosity is desired in eqn. (21), a backward difference form
must be used for the velocity term to maintain the explicitness of the
equations. The viscous force is then given by
{ F v i s ' } = ^ [ C ] ( { / ) ' } {'"M)
(32)
In the presence of this form of viscous forces, eqn. (31) becomes

{>,+ 1 ] =2[({"1'}
-At[M]~l[C]({D'\
4.
-{F'nu})
- {Di~1})
+ 2{Di\
{)''"1 j
(33)
STABILITY OF CENTRAL DIFFERENCE M E T H O D
Stability of the central difference method can be examined by either Fourier

methods or energy methods. Fujii 8 has given a proof of stability in energy
for the central difference method, which was extended to material nonlinearities by Oden and Fost, 1 3 and to geometric non-linearities by
Belytschko, Holmes and Mullen. 2
We will here use Fourier methods, which are attributed to von Neuman.
These methods are basically applicable only to linear systems, although,
through the use of perturbation techniques, non-linear systems can also be
examined. We will restrict ourselves here to the analysis of linear systems.
The internal force vector can then be written as a product of the stiffness
matrix, [K], and the nodal displacement vector, so that eqn. (24), when
damping is added, becomes
[M]{D} + [C]{D] + [K]{D} =
(34)
If the damping matrix is either mass or stiffness proportional, this system

may be diagonalised into an uncoupled system of equations
. + 2 + 2 = 0
(35)
where is the fraction of critical damping in each mode and is the natural
TIME INTEGRATION OF STRUCTUREMECHANICAL SYSTEMS
105
frequency. The forcing function has been dropped because examination of

the homogeneous system suffices to determine stability. Attention is now
restricted to the modal equation associated with the highest frequency, for
it governs stability.
The solution to the above difference equations can be written as
x" = '"' = "
(36)
where (note the superscript on is an exponent)

A = '
(37)
and y is a constant to be determined. Substituting eqn. (36) into eqn. (35)

and factoring out A" - 1 , we obtain the characteristic equation
A2 + X(g - 2 + h) + 1 - g = 0
g = 2;
h = 2 / 2
(38)
(39)
Boundedness of the solution (i.e. stability) then requires that the modulus
of /. be less or equal to unity,
;./ < 1
(40)
Solving the quadratic equation (38), we find that this imposes the following
conditions:
A t<Q2
+ \ )
(41)
Thus, for stability, must be positive. When = 0, we get the usual

criterion
>0;
2
At <
(42)
whereas for < 0 an examination of (41) and (42) shows that the stable time
step is reduced by damping. In fact, if the highest frequency is critically
damped,
^ 1 }
(43)
which corresponds to a reduction of about 60% in the stable time step.

The highest frequency of a mesh is bounded from above by the maximum
unconstrained element frequency in a mesh (Hughes 10 ). This can easily be
shown by means of Rayleigh quotients. By unconstrained element
/ <
2 (
106
T. BELYTSCHKO
frequencies we here mean the individual frequencies of the elements

considered individually with all degrees of freedom unconstrained.
We will now consider the resulting stability limits on the time steps for
various types of elements. For simplicity, we will confine ourselves to one
dimensional meshes. The results for onedimensional meshes exhibit the
salient features of the results for two and threedimensional meshes, but
while the latter frequencies can only be established numerically, one
dimensional mesh frequencies can be established analytically.
To begin, we consider the simplest element, the linear displacement field,
constant strain, twonode element. For this element, for linear displace
ment elements.
AE
[k] =
(44)
and the lumped mass matrix is

Al
(45)
where A is the crosssectional area, i s Young's modulus, the density and

/ the element length.
The highest frequency of this element is given by
"
4r 2
A4 E
'PA I2'
I2
(46)
where c is the wave speed. Substituting eqn. (45) into eqn. (50) gives
/
At<(^'2+
1 )
(47)
For the undamped equations this gives

/ <
(48)
which limits the time step to the traversal time of the elastic wave across an
element. This corresponds to the well known CourantFriedrichsLewy
condition (CFL condition). 7 It should be noted that the inequality of eqn.
(47) holds for both uniform and nonuniform meshes; in the latter case the
time step is governed by the element for which l/c is the minimum. This, of
course, makes extreme mesh refinement in one portion of a mesh somewhat
undesirable, for the time step for the entire mesh is governed by the
minimum traversal time.
TIME INTEGRA TION OF STRUCTURE
MECHA NICA L SYSTEMS
107
The consistent mass matrix for the twonode, linear displacement

element is given by
pAl
[m] =
(49)
This yields that the maximum frequency and time step stability limit are
,
12c2
or = p
(50)
/
At < =
(51)
By comparing eqns. (48) and (51) it can be seen that the use of a consistent
mass reduces the stability limit on the time step below the CFL condition,
which, along with the need to invert [M], constitutes a serious drawback in
the use of a consistent mass matrix with explicit time integration.
We will now examine the characteristics of higherorder elements by
considering the threenode, quadratic displacement, linear strain element.
The mass and stiffness matrices for this element are (the middle node is
associated with the second row and column)
cons
lm]
lump
[m]
Al
1"
Al
~6~
4
2
1 0
0 4
0 0
2
16
2
o0
1 _
(52)
(53)
7
-8
(54)
-8
16
[*]
1
-8
where the lumped mass here is the form obtained by scaling according to the
diagonal terms of the consistent mass.
The eigenvector for this element is given by
AE
{rf}T={l,-il}
(55)
and the eigenvalues for the lumped and consistent mass are, respectively,
,
24c2
I2
60c 2
(56)
(57)
108
T. BELYTSCHKO
This yields for the lumped mass that

At <
(58)
V6c
while for the consistent mass

/<
(59)
VI 5c
Both of these stability limits are significantly below the CFL condition.
This, in conjunction with the difficulties of lumping masses for irregular
two or threedimensional elements, makes these elements somewhat
unattractive for explicit time integration use.
Structural elements, such as beams and plates, have another special
property which arises from the constraints imposed by the structural
theory. If we consider a EulerBernoulli beam element with a cubic
displacement field, then the flexural part of its stiffness and mass is given by
156
cons
[m]
Al
=
42
22/
4/ 2
54
13/
156
13/
3/2
22/
4/ 2
(60)
210
[m]
lump
x2
Al
42
(61)
10
2
al
12
[k]
EI
2
6/ '
2/:
61
4/2.
6/ 1 2
4/2 - 6/
12
W ! T = {,.,
(62)
(63)
'2/
where / is the section modulus and is a scale factor for which we usually
use a value of 175.
The highest frequency and associated mode of a beam element is then
given by consistent mass
-3; A . - A r - W l K / l
/
21 Orr,
(64,
109
lumped mass
\92c2r2
/3/2
1
w2=^crA.
Ai = N^LW}T = i / [ 1 6 / _ 1 6 / ]
(65J
/
12crg
/
Here rg is the radius of gyration of the cross-section, which is given by
r\--A
(66)
The lumped mass again yields a larger stable time step.

The important feature of the beam element is that varies with the
square of the element length. This is a property of systems that are
parabolic, such as heat conduction and the Euler-Bernoulli dynamics of
beams. In the standard engineering analysis of frames and beams in which a
linear axial displacement field is combined with a cubic transverse (bending)
displacement field, eqn. (48) governs as long as it is smaller than the bending
stability limit, eqns. (64) or (65). For example, for a lumped mass matrix, if
/ > 4^38
(67)
the time step is governed by eqn. (48).

Below that length, eqn. (65) becomes operative and any mesh refinement
results in a time step reduction proportional to the square of the element
length. It should be added at this point that although the lumped mass
matrix is very successful in large-deformation elastic-plastic problems
where most of the response is in the lower modes, they are not well suited to
flexural wave propagation analysis.
Although these stability analyses apply strictly only to linear systems,
they also provide useful stability estimates for non-linear problems.
Material and geometric non-linearities generally do not provide any serious
destabilising mechanisms. For example, in Belytschko et al.2 it is shown
that the effect of an initial stress on At is given in a bar element by
l2
=
^
2
2, / .
+ ~ 2
c 1+
pi
' \ ' 2,
4
/ 2 <
OJ
(68)
Thus, the effect of the geometrie nonlinearity is proportional to /,

which for most engineering materials is negligible. Therefore, setting the
time step at 80-90% of the formulas given here is usually sufficient to
maintain stability in non-linear systems.
In any case, it is quite important that energy balance be checked
throughout the computation: any significant loss in energy balance ( 5 % or
more) indicates an instability. In many of the computations carried out by
110
T. BELYTSCHKO
the authors, as many as 50 000 time steps on a 200500 element mesh on a

6digit computer have seldom resulted in more than a 12% energy
imbalance. Energy balance checks are particularly important in
elasticplastic problems, for here a temporary instability can often be
arrested by plastic dissipation, yielding results which are completely
erroneous.
The energy balance check is implemented in finite element programs as
follows. The internal energy, WM, and the external work, WM, are
computed by the trapezoidal rule. Thus,
t/im.i + 1 _
pj/int,i _|_ 1
/{'} ({'}+ a i + l J ) d K
(69)
[Ii-1
which by eqns. (8) and (16) gives

WnU+i = Wnti + i \ {A deY({f}
+ {f'+1})
(70)
while
H/CX...+ 1 = WMi + i{A Z>; T ({F exu } + {Fe"'+1})
(71)
The kinetic energy. T, is given by
= \\[][\
(72)
Energy balance then requires that

Wm + - Wa <
(73)
where d is a specified tolerance.

For two- and three-dimensional problems, analytic determination of the
frequencies of various elements becomes quite difficult. However, Holmes 9
has shown numerically for a variety of meshes that for two-dimensional,
constant strain elements, eqn. (54) still holds provided that /is the minimum
distance from any node to an opposite side. Considerable computational
experience also seems to indicate that the flexural criterion can be used for
plate and shell elements with cubic bending displacement fields using for /
the minimum characteristic element dimension.
5.
IMPLEMENTATION
Explicit time integration involves a computer architecture that dillers

significantly from that of other finite element programs. In most cases the
TIME I N T E G R A T I O N OF S T R U C T U R E M E C H A N I C A L SYSTEMS
11 1
TABLE 1
FLOW CHART FOR EXPLICIT INTEGRATION PROCEDURE
(1) Set initial conditions: \D(0)}, {D(0)}; /'= 0.

(2) Find {>, + (1 2 | } , {>, + 1 } by eqns. (25), (26).
(3) Find internal nodal forces by looping through all elements in mesh:
(i)
(ii)
(iii)
(iv)
find strains by eqn. (3) (1 = 1' ' + Ate);

find stresses by material law (see Tables II, III, IV);
find nodal forces/,, by eqn. (16);
add/;, into { F n ' j .
(4) Compute external load nodal forces {F"1}.

(5) Use equation of motion (21) to find [D'+ ' } .
(6) t = t + At, i = i + 1 ; go to 2.
TABLE II
EQUATIONS FOR COROTATIONAL FORMULATION
(1) Computation of deformation displacements (rig = rigid):

{dicf) = \d) - {d"*}
df = dn - iff;
(II.1)
(2) Transformation to corotational coordinates:

( ?t
= a , / # r ; ! ^ r ! = [a]{dd"\
(II.2)
(3) Definition of displacement field:

()=()*<
(II.3)
(4) Strain-displacement equations:

L
njvcX, +cX-,1
77v) =\cX,
ATV^scX
+ JvVn
e = Bm^;
\) = [B][^<)
(II-4)
(II.5)
(5) Stress-strain law:

j=J{iJ,MlJ)
(116)
(6) Nodal force-stress relations:

fn =
h = aL
BMklV;
(/'}= \]{]
L/'i-kl'l/'i
(II.7)
(11-8)
112
T. BELYTSCHKO
assembly or use of a stiffness matrix is unnecessary and. in fact,

counterproductive; only in linear problems with complex elements would it
be desirable to construct a stiffness matrix, since otherwise the direct
computation of nodal forces is computationally less expensive and requires
far less core.
A typical flow chart for an explicittime integration program is shown in
Table I. It can be seen that the number of logical operations are minimal
TABLE III
EQUATIONS FOR CONTINUUM FORMULA TION
(1) Definition of displacement field:

u, = <M*Ki
(2) Velocity straindisplacement equation:
= Bt]Udu
(ULI)
(III.3)
(3) Stressstrain law:

< = < " + audi*) *H<riPa>pj + ajpVpi)
\Xj
(m4)
cx, J
(4) Nodal forcestress relation:

7M =
B^TdV
(III.6)
and that they are very simple in structure. Branching among elements
occurs only in the computation of the element nodal forces, and here the
program can be modularised very effectively, since the operations in various
elements are completely unrelated.
The introduction of geometric or material nonlinearities introduces no
complications, since there is no need for a predictor. Thus, the only
difference between linear or nonlinear material is that the computation of
the stresses, step (3.ii), involves a nonlinear relationship. Similarly,
1 13
geometric non-linearities only affect the form of steps (3.i) and (3.iv). They
add no fundamental complications to the process.
However, for the purpose of streamlining computations as much as
possible, it is desirable to choose strain and stress measures so that the
strain-displacement equations, the constitutive law and the nodal
force-stress relations involve a minimum number of computations. Tables
11 and 111 give formulations recommended for structures and continua,
TABLE IV
COMPUTATION ESTIMATES FOR A TWO-DIMENSIONAL PROBLEM
24 in radius
structure
embedded
in...
Imp! kit
1000 Hz
100 Hz
7 x 1 5 mesh
of 12 in elements
per time step
per ms
105
HP
15 30 mesh
of 6 in elements
per time step
per ms
1-3 106
1-3 106
30 60 mesh
of 3 in elements
per time step
per ms
1-6 10'
1-6 107
Explicit
Exp licit- exp licit or

explicit-implicit
10
(1000 Hz)
104
2-5 105
1-2 104
6 104
1-3 107
(2000 Hz)
5 104
2-5 106
6 104
6 105
(4000 Hz)
1-8 105
1-7 10'
2 105
4 106
1-6 10
respectively. The corotational stretch formulation developed by Belytschko

and Hsieh 3 is recommended for structures, for it permits the use
of simple relations for larger rotation, small to moderate strain structural
problems such as are often encountered in safety analysis. However, they
are not suitable for high-order continuum elements, because the rigid body
rotation is assumed to be constant within each element.
Velocity strain-Cauchy stress formulations such as those described in
Table 111 are very effective for isotropic continua; a formulation identical
with this is used in H O N D O by Key. 1 ' Velocity strain formulations of this
type are not directly applicable to anisotropic materials or structures.
As a basis of comparison, Table IV gives a Green-strain, second
Kirchhoff-Piola stress type of formulation which is often called a
1 14
T. BELYTSCHKO
Lagrangian formulation. It can be seen that the strain displacement and

nodal force-stress relations involve significantly more computations.
The choice of the most suitable formulation depends also on the
constitutive equations to be employed. We cannot consider these additional
factors herein and refer the reader to Belytschko. 1
6.
MESH PARTITIONS
Let us consider a typical engineering problem, such as the cylindrical shell

immersed in a fluid shown in Fig. 1. The relevant dimensions and wave
speeds are:
fluid
cylinder
/ = 4 0 i n (min. length)
c = 56 600 in/s
/ = 3 0 i n (min. length)
c = 202 000 in/s
thickness = 0-12 in
The stable time step for the fluid mesh is 7-06 10~ 3 s, while in the beam
elements used to represent the shell, the membrane limit is active, so the
stable time step for these elements is 1 -5 IO" 5 s. Inan explicit integration
the smaller of the two steps would have to be used, so the time step in the
fluid would be at 20% of the stability limit. Implicit integrations for these
types of meshes are quite uneconomical because of the large bandwidth.
We will discuss here two approaches to integrating meshes of this type:
explicit-explicit partitions, in which both the shell and fluid would be
integrated explicitly, but the shell with a smaller time step; and
FIG. 1. Finite element mesh for a cylindrical shell immersed in a fluid.
115
implicitexplicit partitions, in which part of the mesh, the shell and

surrounding layer of elements, is integrated implicitly, while the remainder
is integrated explicitly. For each of these partitions the stability of the
procedure depends on the coupling between the partitions.
In order to examine the characteristics of such partitions, we will first
consider the linear equations for a partitioned system. For this purpose, we
subdivide the domain into two subdomains, RA and RB. Two types of
'4
<}
<>
'
FiG. 2. Jointed partition.

partitions are commonly used: jointed partitions, where the boundary
between the edge nodes of the two subdomains are connected by elements,
as shown in Fig. 2, and disjointed partitions, where the edge nodes of the
two subdomains lie on the same line and are connected by a kinematic
constraint (see Fig. 3). For a jointed partition the equations of motion can
be partitioned in the form
MA
I l
M AB
M,
K A I)
(74)
(note lower case letters are used for the complete displacement and force
column matrices).
For a disjointed partition it is convenient to subdivide the nodes within
each partition among those interior to the partition and those on the
interface between the partitions, and designate them by upper case and
116
FIG. 3. Disjoint partition.

lowercase subscripts, respectively. The partitioned stiffness matrix can then
be written in the form
0
KA
KAa 0
0
0
(75)
K = *L Ka
0
0
K Bh
0
0
Ku h
The mass matrix can be partitioned similarly. We will write the kinematic
constraint between the nodes in a and b as
da = Gd b
(76)
If we then express the potential energy U and the kinetic energy T in terms
of the partitioned forms of the matrices and invoke the kinematic constraint
(eqn. 76), we obtain
U = i(d A K A d A + 2daTKAadA + d T KA
+ dK B d B + 2d a G 1 K b d B + djG T K b GdJ
T = i(d AM Ad A + 2d11M.,dA + d T M a d a
+ d T M B d B + 2d a G 1 M Bb d 1! + djG T M b Gd a )
(77)
IF" 1 = dA fA + dTfB + djGTfb + daTfa

The use of Lagrange's principle
d L\
dt\Jd)~
L _
rd" ~
L =T U+ W
(78)
117
then yields the equations of motion

M,
M, + G T M h G
M Bb G
KAa
G T Mlib
Vlf
,
Ka + G KbG
KBbG
G KBb
KB
(79)
To simplify the notation, we will write the explicit linear multistep formula
in the form
d" +1 =/J,d" + a 1 d"rh" 1 (d)
(80)
where and a, are scalar factors which depend on the method and time
step (i.e. [J] = 2 ,] = 2 for the central difference method) and h"(d)
designates the historical values of d and its derivativesthat is, those
pertaining to time steps preceding or including /;. By use of the same
notation, an implicit linear multistep method can be written as
d" +1 =/J 0 d n + 1 +h"(d)
(81)
If part of the mesh isto be integrated with a smaller time step, At/m (misan
integer), then we write the explicit difference formula as
(pm)
/M n
- ( P - Dim
+ ,d
n+ p - l ) / m
, un + l p - a i / m / j .
(82)
We first consider explicitexplicit partitions with the same time step.

These partitions are primarily of value when two different programs are
used for different parts of the mesh. For example, a common application is
to use an explicit Lagrangian finite difference program such as H E M P or
REXCO with an explicit structural finite element program such as
WHAMS. We first consider a jointed partition. If we substitute eqn. (80)
into eqn. (75) and rearrange terms, we obtain
1
h'
1
MAB
PI B
() + *
() + *ldH
'KA
Âl)
(83)
K.AB
If a consistent mass matrix is used, d'A+ ' and d'+ ', are coupled through the
118
T. BELYTSCHKO
terms M A B , and a totally independent solution of the two mesh partitions is

impossible. However, with a lumped mass matrix, M A B vanishes, so we
obtain
d A +1 = /5, A M A -'(f A - K A d A - K A B d B ) + * 1 A d A + h ' - ' ( d A )
< '
= '(fB - K l B d A - K B d B ) + 3 1 B d B + h ' - ' ( d B )
(84)
dA+
It can be seen that

' depends only on the ;'th time step of the solution in
(i.e. dB) and similarly dB+ ' depends only on d A . Hence, either partition may
be integrated first completely independent of the other partition.
F u r t h e r m o r e , since eqn. (84) c o r r e s p o n d s to a precise application of the
explicit time integration formulas to each s u b d o m a i n , stability is governed
by the usual criteriafor example, for the central difference m e t h o d t h a t
/<
<*>mai
(85)
F o r a disjointed partition the mass matrix becomes non-diagonal even

when the original mass matrix is lumped because of the presence of the
G 1 M b G term. However, because this part of the matrix is usually smaller
than the other p a r t s , it is still convenient to work with a lumped mass
matrix. Substituting eqn. (80) into eqn. (79), we obtain
dA+ ' = /? 1 A M A '(fA - K Au d' - K A d A ) + h ' " ' ( d A ) + s 1A d A
(85a)
<+ ' = j8.A(M a + G T M b G ) - ' + GTfb - K a d a - K A a d a

- G T K b d B - G T K b Gd;,) + h - ' ( d b ) + a 1 A d b
d'B+ ' = /5 1 A M B '(fB - KBd' - K B b Gd a ) + h ' " '(d B ) + a 1B d B
(85b)
(85c)
The displacements d|,+ ' are c o m p u t e d by eqn. (76).

It can be seen from the above that a completely separate solution is still
possible except for the need to transfer the mass of the interface nodes b to
p r o g r a m A before c o m p u t i n g d[f ' by eqn. (85a). N o t e that terms such as
K 1!b d B pose no problems, because they can be considered simply as forces of
B applied o n t o A. This procedure has the same stability properties as a
jointed partition.
To avoid a mass transfer, a c o m m o n procedure is to designate one side
the master side, with the other side the slave side. T h e slave side
displacements are c o m p u t e d first except at the interface. T h u s , if we
consider partition B the slave, A the master, we would first c o m p u t e dB+ ' by
eqn. (85c). T h e forces of B o n t o A are then taken to be
/ A B = C 1 "(KBbdj, + K X - M b d h - ' )
(86)
1 19
where the last term represents a d'A lembert force corresponding to the
accelerations of nodes in b. Note that it must be shifted back one time step,
because in explicit methods, db is known only after db+ ' is known. Equation
(85b) is then replaced by
da+ ' = Ma '(f, + CTfb + fA B Kada K A a d a ) + h " ' (db) + , AdA
(87)
so that a complete separation of algorithms is feasible. However, the
transfer of inertial terms with a time lag which is indicated by eqn. (86) has a
destabilising effect that is demonstrated by Belytschko and Yen. 6
We next consider an explicitexplicit jointed partition where domain A is
integrated in time with a time step At I in, while domain is integrated with a
time step A t. We will immediately restrict our attention to M A B = 0. The
governing equations are then
dB+ ' =
IBMBHF'B
KA Bd'B KBdB) + a1Bd'B + h " '(d B )
dA+"""> = /j 1 A M A "'(F A + , " 1 , ' m K A B d A
+,
(88)
"" /m K B d B +(n 1)/m )
+ aiBd B +,n """" +h'"' 1 1 ""(d B )
(89)
It can be seen that eqn. (88) can be performed independent of eqn. (89), so
that domain is easily updated first. However, for m > 1, eqn. (88) requires
the value ofdA+,"~ " "'forali nodes in A that are connected to elements in RB;
let these nodes be designated by the subscript . If dA + ' is obtained first,
then the fractional time step values can be obtained by interpolation.
Because these partitions with interface interpolation do not ensure that
all of the difference equations in time are met at all nodes, the usual stability
criteria no longer apply.
For a jointed explicitimplicit partition, (A , explicit; B, implicit), the
equations for updating the displacements are
' = /J1A IVI '( KA d"A KA Bd"B) f a 1A d A + h"" '(d A )
\"V
+
Ku + TT" M B 1 / mF++ 1' WKA Bd"
(90)
J / AI + 1" + ^ MBh"(dB)
V
POB
/
V
POB
/
Thus, if the explicit domain is updated first, the implicit domain may be
updated without any interpolation. A s expected, the stability of this
procedure then depends only on the highest frequency of the explicit part of
the mesh.
The essence of eqns. (85) can be graphically deduced by considering the
flow of information in explicit and implicit time integrations. For this
<
' =
120
T. BELYTSCHKO
A.
EXPLICIT-EXPLICIT
i+1
w~
r-^.
ft,
B.
EXPLICIT-IMPLICIT
/+;
INTERFACE
FIG. 4. Flow ofinformation in explicit-explicit and explicit-implicit partitions.

purpose we consider a one-dimensional mesh of constant strain elements as
shown in Fig. 4. Information only travels across one element during one
time step in an explicit method, but the entire mesh at the two time steps is
coupled in an implicit method. It can be seen from Fig. 4 that
explicit-implicit partitions involve no interpolation, but that
explicit-explicit partitions require an intermediate value to be determined.
In fact, two approaches can be taken to determine this intermediate
value: the equations at the interface node can be integrated at time step /
with both the small time step and the large time step to obtain u+ ' and
u\ + 2), or they can be integrated only once to obtain u, at the larger time
step with interpolation used to obtain u\ + il 2). The former approach has
significant disadvantages whenever the integer multiple ratio between the
time steps, m, is greater than 2, for the determination of u\+J/m,j = 1 to
(m 1 ). then requires the substep integration in RB of a subdomain of m 1
121
elements. This is obviously unwieldy in complex two- or three-dimensional

meshes, where this subdomain would have to be separately identified.
We have found the following properties of partition:' 1
(1)
(2)
(3)
In explicit-explicit partitions only linear interpolation is stable for

any At.
In master-slave node couplings the mass must be transferred rather
than the inertia.
In disjointed partitions the use of different time steps on A and is
unconditionally unstable unless an artificial viscosity is added.
Explicit-implicit partitions are far more difficult to program than

explicit-explicit partitions. The major advantage of explicit-implicit
partitions over the latter is that they permit the efficient use of consistent
mass matrices, which is advantageous when flexural elements or higherorder isoparametric elements are used. Thus, explicit-implicit partitions
are particularly useful for structure-media interaction problems, where the
implicit integration would be used only in the structural mesh.
The advantage of an explicit-implicit partition over a complete implicit
integration, of course, lies in the considerably smaller bandwidth of the
implicit part of the mesh. Since the computational effort in the solution of
the implicit equations varies with the square of the bandwidth, the number
of computations per time step is reduced significantly. In order to quantify
this qualitative discussion, we have given in Table I computational
estimates for a structure-media interaction problem similar to that shown
in Fig. 1, except that the dimensions are different. These computational
estimates are based on a moderately complex non-linear stress-strain law,
in which about half the computational effort in the explicit calculation is
devoted to evaluation of the stresses. In the computational estimates for the
implicit integration a direct solution procedure combined with a modified
Newton-Raphson method was considered, and an inversion of the stiffness
matrix at every time step was assumed. Only the computations for the
inversion were included in the implicit method, as they constitute the bulk
of the effort in larger meshes. In these estimates the frequency cut-offs are
also given, since the size of the time step in an implicit integration depends
on the desired spectral content. Since the differences between
explicit-explicit and explicit-implicit partitions are minute, only a single
column of figures is given.
It can be seen that for small meshes, if the cut-off frequency is small,
complete implicit integration is competitive. However, as the mesh is
122
T. BELYTSCHKO
refined, it becomes much m o r e expensive than explicit integration, which, in

turn, is more expensive than solutions by explicit-explicit or
explicit-implicit partitions. Because there is n o difference between the latter
in this type of p r o b l e m , explicit-implicit would only be preferable if a
consistent mass is desired for the structure.
T h e i m p o r t a n c e of choosing an effective scheme is quite obvious from
Table L ' A n i n a p p r o p r i a t e scheme a d d s immensely to the c o m p u t a t i o n a l
cost.
REFERENCES
1. Belytschko, T. (1975). 'Nonlinear analysisdescriptions and numerical
stability', in Shock and Vibration Computer Programs (Ed. W. and B. Pilkey),
pp. 537-62. Shock and Vibration Information Center. Washington. DC.
2 Belytschko, T., Holmes. N. and Mullen, R. (1975). 'Explicit integrationstability, solution properties, cost', in Finite Element Analysis of Nonlinear
Structural Behavior (Ed. T. Belytschko et al.), pp. 1-23. ASME, AMD
Vol. 14, New York.
3. Belytschko, T. and Hsieh, B. J. (1973). 'Nonlinear transient finite element
analysis with convected coordinates'. Int. J. Num. Meth. Eng.. 7, 255-71.
4. Belytschko.T. and Mullen, R. (1977).'Mesh partitions of explicit-implicit time
integration', in Formulations and Computational Algorithms in Finite Element
Analysis (Ed. J. Bathe et al.). MIT Press.
5. Belytschko, T. and Mullen, R. (1978). 'Stability of explicit-implicit mesh
partitions in time integration'. Int. J. Num. Meth. Eng.. 12, 1575-86.
6. Belytschko.T. and Yen. H. E. (1978).'Interpolation and extrapolation on mesh
interfaces in time integration'. FENOMECH
Conference.
7. Courant, R., Friedrichs. K. and Lewy, H. (1928). 'On the partial difference
equations of mathematical physics', Math. Ann., 100, 32-74.
8. Fujii, H. (1972). 'Finite element schemes: stability and convergence', in
Advances in Computational Methods in Structural Mechanics and Design (Ed.
J. T. Oden et al.), pp. 201-18, University of Alabama Press.
9. Holmes. N. (1976). 'Characteristics of transient finite element solutions'. Ph.D.
Thesis, University of Illinois at Chicago.
10. Hughes, T. J. R. (1978). 'Explicit-implicit finite elements". FENOMECH
Conference.
11. Key, S. W. (1974). ON DOA finite element computerprogram for the large
deformation dynamic response of axisymmetric solids', SLA-74-0039. April.
Sandia Laboratories. Albuquerque, New Mexico.
12. Krieg, R. D. and Key, S. W. (1973).'Transient shell response by numerical time
integration'. Int. J. Num. Meth. Eng., 7, 273-86.
13. Oden, J. T. and Fost, R. B. (1973). 'Convergence accuracy and stability of finite
element approximations for a class of nonlinear hyperbolic equations'. Int. J.
Num. Meth. Eng., 6, 357-65.
Implicit Finite Element Methods for the Dynamic

Transient Analysis of Solids with Particular
Reference to Non-linear Situations
D. R. J. OWEN
University of Wales, Swansea, UK

1.
INTRODUCTION
In many engineering applications the inclusion of dynamic terms in the

analysis of structures is essential. The term 'dynamic transient situations'
implies that the inertia terms must be included in the equations of
equilibrium. In certain instances widely different displacement and stress
predictions are obtained from a quasi-static analysis and a complete
dynamic transient solution. Examples where such inertia effects are
important include the impact loading of structures due to collision, shock
blast or explosive conditions, in which case the loading is of high intensity
and is applied for a short time period only ( % 1 ms). A further important
case is that due to seismic action, where the structural loading takes the
form of a prescribed acceleration history. Here the loading acts typically for
10-30 s and reverses in sign in analmost random manner. In no application
are the problems more serious than in the nuclear and chemical industries,
where "fail safe' conditions are essential. The type of structural loading can
have a profound influence on the choice of numerical method most suitable
for solution of the problem, and this aspect is discussed in Section 2.
In the last decade or so significant advances have been made in the
development and application of numerical methods to the solution of
dynamic transient problems. A primary factor in this progress has been the
parallel development of high-speed digital computers providing the faster
execution times required to make the solution of engineering problems a
feasible proposition.
The transient response analysis of structural systems generally involves
the discrete modelling of the structure by either finite element or finite
difference methods and the solution of large, second-order differential
123
124
D. R. J. OWEN
equations by direct time integration techniques. Although finite difference

solutions compare well in accuracy with those obtained by use of the finite
element method, the difficulties encountered in problems with complicated
geometries make its use unattractive, and in this study we shall concentrate
our attention on the use of finite elements in space with a finite difference
time solution.
Whatever the spatial discretisation scheme employed, the resulting
system of governing equations for the transient dynamic problem becomes
M + C + P(a) = F(/)
(1)
in which dots denote differentiation in time, and a stands for a set of

parameters describing the displacements which in the context of the finite
element method are the nodal displacement components. Matrix M is the
structural mass matrix (generally independent of time or displacement), C
is the damping matrix, which may well be a function of the velocities . and
F(7) are the applied or activating forces. The term is the internal set of
forces opposing the displacement of lhe structure. For linear elastic
situations
= Ka
(2)
where is the structural stiffness matrix, but for nonlinear situations

involving large deformations and/or plasticity behaviour must be
calculated from the stress distribution satisfying all nonlinear conditions,
so that
*
BVdr
(3)
in which are the nonlinear stresses and is the appropriate matrix
expressing the strains in terms of the nodal displacements. For large
deformation problems itself is a function of the displacements a.
The procedures for the numerical solution of (1) can be divided into two
classifications:1 direct integration methods and modal superposition
techniques. The direct integration methods can be further subdivided into
explicit and implicit methods. A lthough the direct integration and modal
superposition techniques may at first sight appear to be different, they are.
in fact, closely related and the choice of method is a question of numerical
efficiency. However, the modal superposiiion technique is normally used
for linear problems only.
In this chapter we shall first consider the use of implicit time integration
methods for the solution of linear elastic dynamic transient problems, and
IMPLICIT FEM FOR DYNA MIC TRA NSIENT A NA LYSIS OF SOLIDS
125
these will form the basis for nonlinear solution procedures. The effects of
geometric nonlinearities arising from large deformations will then be
introduced into the numerical algorithm. For materially nonlinear
problems the relevant expressions from the theory of plasticity are
presented, and, finally, the concept of viscoplastic solids, which is more
appropriate to timedependent plasticity analysis, is included in the time
integration process.
2.
PROBLEM CLA SSIFICA TION
In the direct integration method eqns. (1) are integrated by a numerical

stepbystep procedure, the term 'direct' implying that no transformation of
the equations to different basic functions is employed prior to the
integration process. The method is based upon two major ideas:
(a)
The dynamic equilibrium equations ( 1 ) are satisfied at discrete time

points within the interval of solutions. These time intervals are
assumed to be A t apart.
(b) A variation of displacements, velocities and accelerations within
each time interval / is postulated. Different forms of these
assumed variations give rise to different direct integration schemes,
each of which have different accuracy, stability characteristics and
cost of solution.
The basic requirement is to advance the solution from known conditions

a, and at time / to give the corresponding values at time t+, = t + A t.
If the solution at time / + , is obtained by considering equilibrium
conditions at time /, the method is termed an explicit integration method,
and such schemes do not require a reduction and inversion of the structural
stiffness matrix. K. However, explicil time integration schemes are only
conditionally stable, which requires the time step length / to be smaller
than some critical value, A tcr, for the process to converge.
On the other hand, if the solution corresponding to time tn+l is obtained
by considering the equilibrium equation (1) at time tn+[, the integration
process is said to be implicit, requiring the factorisation of the structural
stiffness matrix. K. at every time step. Such methods are unconditionally
stable, for linear analysis at least, and the maximum time step length that
can be employed is governed by the accuracy of solution and not by stability
of the integration process. Thus, the computer operations for an explicit
method are relatively few in number and are independent of the finite
126
D. R. J. OWEN
element mesh band or front width. However, there are severe restrictions on
the permissible time step length. Implicit methods demand many more
operations per time step but the time step length can be an order of
magnitude larger than for explicit algorithms. Both methods have
substantial advantages for certain classes of problems, and the merits of
various options have been compared and discussed by many investigators. 2 " 7 In forming guidelines for an appropriate choice of method,
it is important to note that certain problems require small time steps
regardless of the numerical stability requirements, whereas others do not.
Generally, problems can be categorised 8 as (a) wave propagation problems
and (b) inertial problems.
Wave propagation problems are those in which the behaviour at the wave
front is of engineering importance, and in such cases it is the high-frequency
components that dominate the response. Problems in this category include
shock response from explosive or impact loading. Dynamic problems
which are not wave propagation problems can be considered inertial
problems, and here the response is governed by the low-frequency
components. This class includes seismic response and large deformations of
elasto-plastic structures under ramp or step loading. The rise time and
duration of the load relative to the time required for a wave to traverse the
structure offers one approach to categorising the problem: if the rise time
and duration exceed several traversal times, the problem is generally of the
inertial type. As a broad guideline, wave propagation problems are best
solved by explicit techniques, whereas implicit techniques are more
appropriate for inertial problems. However, the relative economy of both
approaches is also influenced by the topology of the finite element mesh and
by the architecture of the computer to be employed.
3.
BASIC FINITE ELEMENT EXPRESSIONS
In order to evaluate the various matrices arising in governing eqns. (1), it is

necessary to specify the finite element process of discretisation employed.
Only the basic expressions are provided here, and the reader is directed to
references 9- ' ' fora more complete treatment. With reference to Fig. 1. the
displacements, u, at any point within an element are expressed in terms of
the nodal values, a. as
>"
Na
(4)
IMPLICIT FEM FOR DYNAMIC TRANSIENT ANALYSIS OF SOLIDS
127
-o-
$=-!
TWO AND THREE DIMENSIONAL

USED
1-0
1
)
0-577
'I
r^
INTEGRATING
ISOPARAMETRIC ELEMENTS
Io
0-577
POINTS FOR 2 x 2 GAUSS
RULE
FIG. 1. Isoparametric element configurations.

where N, are the element shape functions for each of the m nodal points of
the element. The change in strain associated with a displacement increment
Aa is defined as
= Ba
(5)
where B is the appropriate strain matrix, generally composed of derivatives

of the shape functions. For geometrically non-linear problems B is
displacement dependent and its explicit form isconsidered in Section 7. The
corresponding stress change is obtained by use of an appropriate
constitutive law as
=
(6)
The explicit form of D is considered in Section 8 for plasticity problems
128
D. R. J. OWEN
and in Section 9 for viscoplastic material response. Then the current

displacements and stresses become
a = a
(J
II
a = c
+ 3
+ ,
(7)
On imposing a virtual displacement system to the structure, the principle of

virtual work yields the governing equation ( 1 ), in which the various terms
are obtained from the following individual element forms:
N T pN dr
(8)
C =
N T CNdi'
(9)
Pn =
[B'fV.dr
(10)
F" =
N T bdt
ML
NTtdi
(11)
in which is the material density, and C stands for velocity proportional

damping forces, b for the body forces/unit volume and t for the boundary
tractions.
The overall structural forms of expressions (8)-(ll) are obtained by
summing individual element contributions according to the usual finite
element assembly process.
4.
IMPLICIT TIME INTEGRATION METHODS FOR LINEAR

PROBLEMS
In this section we describe the use of three implicit integration methods for
linear analysis. In particular, the Houbolt. Wilson 0 and Newmark
methods are presented. Use of these methods in linear situations also forms
a basis of non-linear solution algorithm by the same techniques.
4.1. The Houbolt Method
In this approach, 1 2 standard finite difference expressions are used to
129
express the acceleration and velocity in terms of the displacement. In

particular, the following backward difference formulae are used:
"+'
lla
7^T,
"+' /
18a
9a
,-i " - : }
+ , = ^ 2 { 2 a + 1 - 5 a + 4 a _ , - a _ : }
(12)
where the subscript /; + 1 stands for time i+i + t + At for which solution
is sought. In this implicit method equilibrium conditions are considered at
time /+, in order to obtain a + ,. so that eqn. (1) becomes
M + 1 + C (I+l + Ka + , = F +l
(13)
Substituting eqn. (12) into eqn. (13) and rearranging,

2
11
_ M
(5
+ C )a _ 1, ++[.;M+-C)
+ V2
27
(7
3,
(14)
Thus, the solution for a + ] requires knowledge of a, a,,., and a_ ; . The

process can be started from a knowledge of a 0 , 0 and 0 , but it is more
common and accurate to calculate the first two time steps by use of some
other algorithm such as the central difference explicit scheme with a fraction
of / as the time step.
The appearance of the stiffness matrix Kon the left-hand side of eqn. (14)
implies that a full simultaneous equation solution is required for every time
step. The H oubolt method, being an implicit method, is unconditionally
stable, and. hence, there is no critical time step limit and . can, in general.
be selected many times larger than for the central difference method.
If mass. M. and damping, C. terms are neglected, it can be seen from eqn.
( 14) that the H oubolt method reduces directly to a static analysis for time
dependent loads.
4.2. The Wilson Method
This method 1 3 is essentially an extension of the linear acceleration
technique in which a linear variation of acceleration from time tn to time . + ,
is assumed. In the Wilson 0 method the acceleration is assumed to be linear
from time / to time t + 0 . where 0 > 1, as shown in Fig. 2(a).
For 0 = 1-0 the method reduces to the linear acceleration scheme, but in
130
D. R. J. OWEN
t.at
tAt
WILSON-
b.
NEWMARK
t.eit
tet
METHOD.
METHOD.
FIG. 2. Evaluation of accelerations for the Wilson and Newmark methods.

this case the method is only conditionally stable. For unconditional
stability > 137. For any time so that 0 < < At we have from Fig.
2(a) that
a
r +i a; +
(2, + M , ,)
(15)
Integrating eqn. (15) gives

2
a( + t = a, + a, + ^ ^ (,+UA , ,)
(16)
and integrating again gives

a, + I = a, + a,T + 3, 2 + ^ _ ( 8( + (;., a,)
(17)
131
Or. for the particular time = OA t, eqns. (16) and (17) give
a
/ + UA/
OAt
, +
( a / + ()A/ +
(18)
;)
.
2
a, + ( M , = 3 , 4 - 0 - /a, + - (a, + ., 4- 2a,)
(19)
from which we can solve for , +(M , and l+0.A l in terms of a, + UA,:
6
3,+ =
" '
W (a,+"A' ~ a,) - f>/7 a ' ~~ '
3
/ + A, = g 7 ^ ( a , + A , - a , ) - 2a,
<20)
0-At
a,
(21)
To obtain the displacements, velocities and accelerations at time t + A t, the

equilibrium equations are considered at time t + OA t. This requires
projection of the applied load vector to time + Ai, which is performed
linearly as
F, + ^ , = F, + 0(F, + A, F , )
(22)
and then eqn. (1) at time t + 0A t becomes

1" 3 ; + (/ + Ca, + (M, + Ka, + ((.A, = F, + (M,
(23)
Substitution of eqns. (20) and (21) into eqn. (23) gives the following
system of simultaneous equations, which may be solved to give a, + UA/:
(b0M + ,C + K)a, +(M , = F, + 0(F, + A, F,) + M(/>0a, + b2, + 3 ,)
+ C(6,a, + 4 , + 5 ,)
(24)
where
b0 = 6/ 2 ;
, = 3/;
2 = 2ft, ; ft3 = 2;
4 = 2;
>5 = /2
(25)
Substituting a, + 0., into eqn. (20) gives ,+.A which is then employed in
eqns. (15)(17), all evaluated at = A t, to give
,+Ai = ^(a.+tA, - a,) + 7 , + 8 ,
a, + A/= , + N ( , + A, + a,)
3, +/ = a, + /, + 1 0 (, + d ; + 2,)
(26)
where
bb = b0iO;
h1=-b2/0;
=l-3/0;
9 = . / 2 ;
1 0 = , 2 /6
v27)
132
D. R. J. OWEN
It is noted from eqn. (26) that no special starting algorithm is required, since
a, + A l + & , + A, are all expressed in terms of the same quantities at time t
only.
4.3. The Newmark Method
This method 1 4 is also an extension of the linear acceleration method. The
following assumptions are used:
,+A, = , + [(1 e5), + l+A l]A t
(28)
3, + , = a, + ,A t + [({ a), f oc# + J A/
(29)
where and are parameters that can be determined to obtain integration

accuracy and stability. When = { and = , this method reduces to the
linear acceleration method. Newmark originally proposed as an uncon
ditionally stable scheme the constant average acceleration method, in
which case = | and = . For solution of displacements, velocities and
acceleration at time / + /. the equilibrium equations (1) are considered
also at time t + A t:
M, + &, + C, + A, + Ka, + A, = F, +A
(30)
Solving eqn. (29) for , + A, in terms of a, + A, and then substituting for , + A; in

eqn. (28), we obtain equations for , + A, and , + , each in terms of the
unknown displacements a, + , only. Substituting these expressions for , + A,
and , + A, into eqn. (30) gives a system of simultaneous equations which can
be solved to give a,+A r:
(b0M + ,C + )3, + , = F, + A, + M(b0a, + b2, + b3,)
+ C(,a, + 4 , + s ,)
(31)
where
0
'
aAt2'
'
2:
/'
b2 =/
-;
-e
2/2
(32)
By use of eqns. (28) and (29) all quantities at time / + A t can be finally
expressed as
, + M = bb(al + Al a,) + 7 , + b8,
I + A, = , + M . + toa< + 4,
a, + M = a , + (a, + A, a,)
(33)
133
where
bb = b0; 7 = b2;
bs = b3;
bg = / ( 1 ); bl0 = A t
(34)
Comparing eqn. (24) with eqn. (31) and eqn. (26) with eqn. (33), it can be
seen that the numerical time integration algorithms for the Wilson and
Newmark methods are identical, provided that the appropriate values of
the constants blb10 are employed: defined by eqns. (25) and (27) for the
Wilson method and by eqns. (32) and (34) for the Newmark method.
Thus, both methods can be easily incorporated in the same computer code.
5.
MODE SUPERPOSITION TECHNIQUE
The number of computer operations required in the direct integration

method is directly proportional to the number of time steps used in the
analysis. It should be remembered that, other than for the lumped mass
explicit central difference method, a complete equation solution must be
undertaken each time step. Of course, the more banded matrices M, C and
are. the less costly is this inversion process. Therefore, in the analysis of
linear structural systems the use of direct integration methods are generally
only effective when the response for a relatively short duration is required.
If solution is required for a long time interval, it may be more effective to
first transform equilibrium equations ( 1 ) into a form whereby the stepby
step time integration process is less costly.
To achieve this, we propose to transform the finite element nodal
displacements as follows:
a(/) = S X ( . )
(35)
where S i s a square matrix and X is a timedependent vector of order N.

Substituting eqn. (35) into eqn. (1) and premultiplying by S T results in
MX + CX + KX = F
(36)
where
M = S T MS;
C = S T CS. = S T KS, F = S T F
(37)
The objective of the transformation is to obtain new system mass, damping

and stiffness matrices M, C and which have a smaller bandwidth than the
original matrices and S must be determined accordingly. The most effective
transformation in practice is established by use of the displacement
134
D. R. J. OWEN
solutions of the free vibration problem in which damping is neglected. In

this case, eqns. (1) reduce to
Ma + Ka = 0
(38)
The solution to eqn. (38) can be postulated to be of the form

a =
- t0)
(39)
where is a vector of order N, r0 isa time constant and is the frequency of

vibration (rad/s) of . Substituting eqn. (39) into eqn. (38) gives the general
eigenvalue problem:
= 2
(40)
The solution to this yields the eigenvalues , , 2 , . . . , to each of

which corresponds an eigenvector 1,2,...
,. Defining matrices and
2 as below
(J
=[.
(41)
]:2 =
(I
it can be readily demonstrated that is a suitable transformation matrix S

in eqn. (37). since and M would then be diagonal matrices owing to the
orthogonal properties of the eigenvectors. In order to allow a complete
decoupling of the equation system (36). matrix C must also be diagonal in
form. This occurs only if the damping can be assumed to be proportional.
This is achieved if
!,
= 2,
' =./
0
'V;
(42)
where , is a modal damping parameter. The assumption in eqn. (42) implies

that the total damping in the structure is the sum of individual damping in
each mode. In this case eqn. (36) reduces to individual equations of the
form
x,(t) + 2,,(/) -f 2) =
^,
= W)
(43)
The solution to each equation in (43) can be obtained by use of the direct
integration algorithms described in Section 2. To obtain the complete
response, the solution to all equations in (43). / = 1. 2, 3 , . . . . N, must be
135
calculated and then the finite element nodal point displacements obtained
by superposition of the response of each mode, so that
a(0 = )
( 44 )
= 1
For many situations the applied loading only excites the lower
frequencies of the structure and therefore only a fraction of the total
number of decoupled equations (37) need be considered in order to obtain a
good approximation to the true structural response. Generally only the first
equations need be considered in eqn. (43), where <s: TV. This means that
we also need to solve only for the lowest/? eigenvalues and corresponding
eigenvectors in eqn. (40) and we only sum in eqn. (44) the response in the
first modes.
For example, in seismic loading situations in some cases only the ten
lowest modes need be considered, although the order of the equation
system may well approach 1000. On the other hand, for blast or shock
loading many more modes are generally needed and may be as large as
2/V/3. Thus, for selected situations the mode superposition technique offers
an economic solution alternative to direct integration.
The modal analysis technique can also be extended to nonlinear
dynamic problems and a summary of such applications can be found in
reference 34.
6.
NONLINEA R SOLUTION A LGORITHMS
In this section we extend the solution algorithms described in Section 4 to

account for nonlinear behaviour. Two basic modifications are required:
(1)
From expressions (2) and (3) it is evident that the linear restoring
forces = Ka must be replaced by the nonlinear value
P =
(2)
'adv
at each stage of the computation.

In order for the displacements and stresses to satisfy fully the
nonlinear conditions of the problem, it is generally necessary to
perform an equilibrium iteration sequence in each time step or
preselected time steps.
136
D. R. J. OWEN
In implicit methods equilibrium conditions are considered at the same

time for which solution is sought. If' solution is known for time t and we
wish to obtain the displacements, etc., for time fn4 , = tn + ,, then the
following equilibrium equations are considered for the nonlinear case:
M + 1 + C a + 1 f P + 1 = F + 1
(45)
where
P
n+ 1
[B + 1 ]V, + idf
(46)
The equivalent nodal forces, P + ,, at time / + , can be estimated as

P + 1 = P + K >
a = a n + 1 - a
(47)
where K is the tangential stiffness matrix evaluated from conditions at time

tn. A ssumption (47) implies a linearisation of the displacement increment
between times / and / + ,. Substituting eqn. (47) in eqn. (45) gives
M + , + C, I+ , + K<5a = F + , P
(48)
which is the form employed in references 1517. The solution of eqn. (48)
yields, in general, approximate displacement increments <5a. To improve
the solution accuracy and to avoid the development of numerical
instabilities, it is generally necessary to employ iterations within each time
step or selected time steps in order to maintain equilibrium. In this case it is
convenient to express equilibrium equations (48) in an alternative form.
With the superscripts /' 1 and ;' being used to denote values at two
successive equilibrium iterations, then the displacement change occurring
between these two stages is
Aa'" = a;,+ 1 *+\
(49)
Then consideration of eqn. (48) at iteration at time tn+ , gives, on use of

eqn. (49),
M;,+ 1 + C a ; + 1 +KA a' = F + 1 P ; , + 1 1 ;
/ = 1,2,3,...
(50)
The first iteration (/' = 1) in eqn. (50) corresponds to the solution of eqn.
(48), where 3 1 = <5a,a+1 =a,,! + 1 = a + 1 . a , + 1 = + , and P + , = P.
The vector of effective nodal forces, P'~+\, is equivalent to the element
stresses in the configuration corresponding to displacements a^ 1 ,. In the
solution of equation system (50) two basic options remain open. If apseudo
force formulation is followed, the stiffness matrix, KH. is kept at a constant
(initial) value, with dynamic equilibrium being maintained by successive
137
iteration with a varying pseudoforce which is the righthandside term in

eqn. (50). Alternatively, in the tangent stiffness method the stiffness matrix
K is allowed to vary throughout the computation, with the term Pj," ', being
replaced by an equilibrium correction term.
The approximations employed for the velocities , + ,, and accelerations,
,',+ ,, depend on the time integration scheme employed. The nonlinearities
in the problem enter equilibrium equations (50) through the tangential
stiffness matrix, KM, which generally must be reevaluated at the beginning
of each time step, and the equivalent nodal forces, P'~+\, which must be
calculated for each iteration within a time step. Section 7 is devoted to
evaluating these quantities for large deformation problems, while Sections
8 and 9 describe the same operations for classical plasticity and elasto
viscoplasticity, respectively. Details of the nonlinear solution process for
the integration schemes considered in Section 4 will now be presented.
6.1. Houbolt Method
Equation (14) can be rewritten as
M + 6^C + Kâ = ( M+ C ) a "~(^ M+

+ ( T 7 2M ++ ^ T 7 C ) a _ 2 + F + 1
3,
(51)
For each time increment a series of equilibrium iteration cycles

; = 1,2,3,
m are performed until the residual forces become tolerably
small. The basic steps of the process are as follows.
Step (1) Initialise, a 0 , 0 and 0 .
Step (2) Form the matrices K, M and C for the structure. The element
contributions are first formed and the overall matrices
accumulated during the equation assembly process.
Step (3) Form the effective stiffness matrix initially, assuming linear
behaviour:
K
M + rr.c
= in
2
( 52 )
/
6/
Step (4) If a new stiffness matrix is to be formed, update K to obtain K
by computing the current value of
K=
B,[DBdi
J
in eqn. (52).
,38
D. R. J. OWEN
Step (5) Form the effective (or pseudo) load vector

F + i = F + i + M ( 3 a 4 a _ , + a _ 2 ) / A r
+ C ( a K , +a2)M?
(53)
Step (6) Solve for the displacement increments

a = [K] 1 F + 1
(54)
Step (7) Iterate for dynamic equilibrium:

(a) i = / + l
(b) Evaluate the (/' 1) st approximation to the accelerations,
velocities and displacements, using eqn. (12), as
ai,;1! =a + a'-1
a!,;\ =
i 11 al,;'. - I8a + 9a. , - 2a^2}
(55)
.+ 1 = ^ 7 2 i 2a fi 5a + 4a, 1 a _ , |
(c) Evaluate the (/' 1) st residual forces, ^, + , as
',,- = F n + , - [M;,; 1 + Caj,; 1 , + ;,; 1 ]
(56)
(d) Solve for the j'th correction to the displacement increments

Aa'= [&]#;
(57)
(e) Evaluate the corrected displacement increments

a' = a ' " 1 + A n'
(58)
(f) Check whether the iteration sequence has converged. It is usual

to compare some measure of the displacement correction
vector. ' (usually the norm), with the same measure of the
total displacements, that is, check whether
NO
Return to Step (7a)
\A'\
+ a'I
tolerance
factor,
(59)
YES
Step (8) Return to Step (4) to process the next time step.
The solution scheme described above can be interpreted as a generalised
Newton process.
139
6.2. The Wilson and Newmark Methods

It is convenient to consider these methods together, since, as was seen in
Section 4, the numerical processes coincide with appropriate choice of
numerical constants. The essential steps in the numerical algorithm are
outlined below.
Step (1) Initialise, a 0 , 0 and a 0 .
Step (2) Calculate the following constants:
Wilson 0 method
A
* ~ (0,)
Newmark
bo
' ~ 0/
1 .b
/2'
'
oc
A t
b2 = 2 , ; 3 = 2
A4 = 2; 5 = 0//2
4 = 1 ; 6 5 = , ( 2
6 = o / 0 ; 7 =
b6 = 0 ; 7 = 2
2/0
bs = 1 3/0; bg = //2
2
, = / /6
oAt
b8 = b3;
b
io
b9 = A t(\ )
= A t
For the Newmark method set 0 = 1, since then the effective

loads for both methods have a common expression.
Step (3) Form the matrices , M and C for the structure.
Step (4) Form the effective stiffness matrix, initially assuming linear
behaviour:
K = 0 M + ,C + K
(60)
Step (5) If required, update
K = JB T D n B n di;
to give new effective stiffness matrix, Kt.
Step (6) Form the effective (or pseudo) load vector, from eqns. (24) and
(31), as
W A , = F, + 0(F,+A I F,) + M( 2 , + b3,)
+ C(M, + M,) - P,
(61)
140
D. R. J. OWEN
Step Cl) Solve for the displacement increments

a = [K,]F, + (M,
(62)
Step (8) Iterate for dynamic equilibrium:

(a) / = / + 1
(b) Evaluate the (;' 1) st approximation to the acceleration,
velocities and displacements, using eqns. (20) and (21), as
',;A, = b0a'-1
+,,
3
/ + /
- b2, - b 3 ,
= , <5a'" '
4 ,
5 ,
(63)
= a, + a
(c) Evaluate the (i 1) st residual forces as
*\lL
= F, + 0(F,+tA, - F,)
|Ma, + 4 , + Ca t + + r , + ( ) A ,J
(64)
(d) Solve for the /th corrected displacement increments
Aa = [K I ]-V;; e , Al
(65)
(e) Evaluate the corrected displacement increments

a' = a' " ' + A a'
(66)
(f) Check for convergence of the iteration process

NO
Return to Step (8a)
lAa'l
<
la, + a'|
YES
Step (9) Return to Step (5) to process the next time step.
LARGE DEFORMA TION PROBLEMS

In considering the effects of gross deformation on structural performance it
is first necessary to choose an appropriate description of the stresses and
strains. The first alternative is to employ the Cauchy or true stresses and the
corresponding A lmansi strain tensor. The Cauchy stress is always oriented
141
in the global x directions and does not follow the particle; this makes it
difficult to write the appropriate constitutive equations. Furthermore, since
we are concerned with true stresses, all integrations in the finite element
process must be made over the deformed volumes.
For these reasons it is far more common to work in terms of the second
PiolaKirchhoff stresses whose reference coordinates move with the
particle and are based on the undeformed area. Thus, for small strains the
second PiolaKirchhoff stresses are the engineering stresses. The
corresponding strains are the G reenLagrange strains, defined as
cu,
cu
cukcuk
CX:
CX:
CX: CX:
, '
_ cu_
oxj
' ,
1 cuk cuk
2 ex, xj
>
(67)
where u are the displacement components and x are the undeformed

coordinates. However, it is difficult to use this description for large strain
problems, since the Green strains are not true strains in this case and the
PiolaKirchhoff stresses lose their physical interpretation. Geometrically
nonlinear problems may be formulated in either:
(1)
(2)
a total Lagrangian coordinate systemwhich is expressed in terms

of the initial position; or
an updated Lagrangian coordinate system in which the initial
position changes and becomes the current equilibrium state prior to
some incremental change.
An updated Lagrangian formulation restricts displacements from a

current initial position to satisfy the criterion of small rotationsand, hence,
can be employed for the solution of large strain problems. The
straindisplacement relationship can be incrementally expressed as
de = Bda
(68)
where can be separated into linear and nonlinear parts:

= B 0 + BL
(69)
In this, B 0 is the usual operator for infinitesimal strains and B, is the non
linear contribution which is a function of the current displacement state.
Matrix B, can be explicitly expressed but it is often more convenient to
express directly. The matrix partition associated with node / can be
142
D. R. J. OWEN
written, for a general threedimensional solid,

ex c N
,
B,
^
cy cz
cx DN,
c cx
cx c N,
- ^
_ cy cx
1
+
cx
cx cN
cx cy
cx
) c N,
+
(
cx
cy C
-^ + ^ oy
<~)
cx
"
c cN
^ ^
cz cz
cy CN;
cy c N,
~^ + ^~
dz cy
cy c
c y cN
cy oN
oy cy
cy cy
oy c N,
~^
~z
cx cx
czcNi
>
cy cy
cx cN
_ _.
c c
cx c ,'
cx c N,
cz cy
cxcNj
cx cx
cycN,
cx cx
cx c N
>
czcN,
cycNj
cx
c c
ON

cz cy
(70)
c c/V,
h
cy c
cN
( cx
czcN
f cN

oy cx
cx c
czcN
;
cx c y _
The only influence that large displacements have on the numerical solution
sequence described in the previous section is in the form of the matrix B to
be employed. In the total Lagrangian approach the original coordinate
system is employed throughout the analysis and the full nonlinear strain B
defined by eqns. (69) and (70) is employed. In an updated formulation the
nodal point coordinates are recalculated (or updated) at the end of each
time step and the infinitesimal strain matrix, B 0 , is employed.
8.
ELA STOPLA STICITY
By far the most common source of material nonlinearity in structures is

that arising from plastic behaviour. In this class of problem we have, as well
as the conventional case of metal plasticity, the ultimate loadcarrying
capacity of concrete structures and the failure of soil masses according to
critical state theories. The basic expressions of elastoplastic theory will now
be presented in a form suitable for finite element analysis. 18 20
Plastic deformation is essentially an irreversible straining which is not
time dependent and begins once a certain level of stress has been reached.
This level is expressed by means of a yield criterion of the form
F(a,ED,d)
=0
(71)
143
where is the plastic strain and 3C is a hardening parameter. The changes of

strain in the material may be separated into elastic and plastic components
as
(72)
d = d ,. + d
where the elastic part is given by
ds =
O~lda
(73)
and the plastic part is assumed to be derivable from a plastic potential, Q,

by the flow rule 21
= d/.
cQ
(73a)
in which is a non-determinate constant. Associated plasticity is defined

by the identity Q = F, in which case
cF
ds D = d/.^r- = d/.q
'
co
(74)
and is termed the normality rule, since it directs the plastic strain increment
normal to the yield surface in stress space. While plastic deformation is
taking place, stresses remain on the yield surface and, consequently, from
eqn. (71)
dF
OF
(
da +
(75)
Introducing
A =
'cF ,
d
idd
ds
(76)
and
cF
( a
=q
(77)
which define the yield surface after initial yield, eqn. (75) may be written
qTd<x-/ld;. = 0
(78)
Also, eqn. (72) can be expressed as

D-'dff + d / ^ oa
(79)
144
D. R. J. OWEN
Premultiplying both sides of eqn. (79) by dT = q1 D and eliminating q ' da by

use of eqn. (78), we obtain the plastic multiplier, d/., to be
'
qTddE
(80)
M +q'Dq]
Or substituting eqn. ( 80) into eqn. (79), we obtain the complete elastoplastic
incremental stress-strain relations to be
dA =
dff = D c p d
(81)
with
D
- = D -ZTdv d = Dq
(82)
It can readily be shown that A = H', where H' is the local slope of the
uniaxial stress-plastic strain curve.
Elastoplastic material behaviour affects the numerical solution algorithms described in Section 6 in two ways:
(1)
(2)
In Step (4) of Section 6.1 (Step (5) of Section 6.2) D becomes D ep

for the evaluation of K. It is noted from eqn. (82) that D e p is stresslevel-dependent, since q = cF/ca.
The stresses a at each plastic point must satisfy the yield criterion.
During application of load, either at the beginning of a time step or
during equilibrium iteration, the stresses at some points may exceed
the permitted level. These stresses must be brought back to the yield
surface by the numerical process shown schematically in Fig. 3.
Thus in Step (7) of Section 6.1 (Step (8) of Section 6.2) an additional
operation is required after operation (b):
Step (Tb)(i) (Step 8(/)) For all yielded points, reduce the stresses to
satisfy the yield surface condition.
These modified stresses are then employed in the evaluation of the
equivalent nodal forces P. + 1, in Step (7c).
A major disadvantage in using classical plasticity theory for dynamic
problems is that time rate effects in the material deformation process are
ignored. This can be partly overcome by allowing the instantaneous yield
stress of the material to be a function of the current rate of straining.
However, time-dependent plastic flow phenomena are physically better
modelled by means of viscoplastic theory, and such an approach is
described in the next section.
145
bdXS.=Dde^
FIG. 3. Reduction of stresses to the yield surface for elastoplastic situations.

9.
VISCOELASTIC-PLASTIC MATERIAL RESPONSE
In order to introduce a material response which is strain-rate-dependent

into the dynamic model, the use of viscoplastic theory is attractive. The
concept of viscoplastic material behaviour is best introduced by means of
the one-dimensional rheological model, illustrated in Fig. 4. The friction
slider component develops a stress , becoming active only if > Y, where
is the total applied stress and Y is some limiting yield value. The excess
stress, , is carried by the viscous dashpot. Instantaneous elastic
response is, of course, provided by the linear spring. The presence of the
dashpot allows the stress level to instantaneously exceed the value predicted
by plasticity theory, the solution tending to this equilibrium level as steady
state conditions are achieved in the system.
Generalising to the case of a continuum, 2 2 " 2 4 the total strain rate can be
expressed as
= c + vp
(83)
where the subscripts e and vp denote, respectively, the elastic and
146
D. R. J. OWEN
Inactive if
vp
il
nrrrrrrm
Fie. 4.
Basic viscoplastic model.
viscoplastic components. The total stress rate depends on the elastic strain
rate according to
= D c
(84)
The onset of viscoplastic behaviour is governed by a scalar yield condition
of the form
F(a, ) F0 = 0
(85)
in which F0 is a reference yield value.
The rate of viscoplastic straining can be defined by the following
associated viscoplastic flow rule:
= -,'<0(F)>
OF
( a
7<<P(F)>q
(86)
in which y is a fluidity parameter controlling the plastic flow rate. The term
(.) is a positive monotonie increasing function for > 0 and the notation
< > implies
<(.)> = (.)
<(.)> = 0
for > 0
. < 0
(87)
147
Different choices have been recommended for the function . The two most
common versions are
in which M and are arbitrary prescribed constants.

At this stage we must specify a time stepping scheme for integrating the
viscoplastic strains. In previous work the Euler process was employed, 24
which is only conditionally stable. In order to take advantage of the larger
time step lengths permitted in the dynamic implicit time integration
schemes, an implicit algorithm can also be used for the integration of the
viscoplastic strains. 23
With the strain rate law expressed by eqn. (86) we can define a strain
increment occurring in a time interval At,, = tn+, tn as
. = /[(1 -/?) . + /3 . +1 ]
(89)
For = 1 we have a 'fully implicit' (or backward difference) scheme, while

= \ results in the so-called 'implicit trapezoidal" scheme, which is also
known as the Crank-Nicolson rule in the context of linear equations. To
define + , we can use a limited Taylor series expansion and write
vp.+i = vp. + ,,,,
(90)
where
H = [~A
= H>)
(91)
and is the stress change occurring in the time interval tn+ , tn. Thus,
eqn. (89) becomes
where
.= + 0
Cn = -Atn-Hn
(92)
(93)
Matrix depends on the stress level and no difficulty arises in its

evaluation. Using the incremental form of eqn. (84), we obtain
= DAiJ1 = 0( - )
(94)
Or expressing the total strain increment in terms of the displacement

increment and substituting for
from eqn. (92) gives
= D(B,,a - /)
(95)
148
D. R. J. OWEN
where
D = ( D ' + C J '
(96)
Matrix D is a symmetric matrix when the viscoplastic law is associative. In

eqn. (95) the notation B is employed to denote the possible inclusion of
large deformation effects.
During a time increment the equilibrium equations which must be
satisfied are given by the following incremental expression:
=
(97)
in which A F represents the change in loads during the time interval A t.

Using eqns. (92) and (95) the displacement increment occurring during time
step A t can be calculated as
(5a
= [K]'A V
(98)
where the incremental pseudoloads are

= J B;,Dvp.A /df + A F
(99)
and K is the tangential stiffness matrix with the following form:

K =
BlDBdt
(100)
The displacement increments a when substituted back into eqn. (95) give
the stress increments A a and thus

/ It
+ = ff,
+
a
M+ 1
= a + J
(101)
BtJa D '
(102)
= . + ,
(103)
Use of eqn. (95) gives
. =
and then
.+ 1
The changes required to include viscoplastic material response in the non

linear solution algorithms described in Section 6 are given below.
(1)
(2)
In Step (4), Section 6.1 (or Step (5), Section 6.2) the form of K
given by eqn. (100) must be employed.
A dditional operations are required to evaluate the stresses as
follows:
149
Step 6.1 (Step 7.1)

(a) Evaluate the viscoplastic strain rate based on stresses :
vp. = ,() = >
(b)
(104)
Calculate the stress increment from eqn. (95) as

= D ( B > - . 0
(105)
Determine the total stresses as

+ = +
(106)
More sophisticated viscoplastic material models can be generated by use of

the overlay principle (references 26, 27).
10.
GENERAL REMARKS AND DISCUSSION
For linear problems the three methods described in Section 4 are

unconditionally stable provided that certain conditions are met on the
parameters involved. For the Newmark method the integration scheme is
unconditionally stable provided that > 0-5 and = 0-25(<5 + 0-5)2 and
the Wilson 0 method is unconditionally stable for 0 > 1-37.
Various studies of the integration accuracies of the method under linear
conditions have been reported, and since solution accuracy depends on
several factors, we shall only quote the optimum parameter values for
general use. For the Newmark method the most desirable accuracy
characteristics are given by = 0-5 and = 0-25, and the corresponding
value for the Wilson 0 method is 0 = 1-4.
We now address some remarks to the question of computational
efficiency of the implicit time integration algorithms previously discussed
for the solution of non-linear problems. Comparison studies have been
reported by several investigators. It is difficult to arrive at general
conclusions, since the suitability of a particular time integration algorithm
depends on several factors, such as the type and degree of non-linearity and
the loading characteristics. Comparison studies on a geometrically
non-linear shell cap 2 8 by time integration schemes, which included the
Houbolt and Newmark methods, indicate that the Houbolt method was,
overall, the best for highly non-linear problems, providing reasonably
accurate solutions in realistic computer times. The Houbolt method was
found to be only conditionally stable, with the critical time step depending
on the degree of geometric non-linearity. Reference 29 reports that the time
150
D. R. J. OWEN
step length depends on the degree of geometric non-linearity and not on

material non-linearities. Reference 3 shows that the Houbolt method is
stable for non-linear problems provided that iteration is used at each time
step. The above studies also indicated the Houbolt method to be, overall,
superior to the Wilson 0 method.
The performance of various time integration methods have been
theoretically compared in reference 30 for inertial problems, and it is shown
that methods with high numerical damping do not necessarily possess
stability characteristics superior to those with less damping for non-linear
problems. In particular, it is demonstrated, by numerical examples, that the
Park method shows improvement over the Houbolt method in spite of the
higher damping inherent in the latter scheme.
At present, explicit methods are considered to be more economical for
non-linear transient problems than implicit techniques. The main reason
for this is that, in path-dependent materials, such as elastoplastic materials,
the time step required to follow the stress history adequately is often of the
order of the Courant stability limit for explicit integration. Indeed, even for
linear problems it is conceivable that explicit methods can be economically
superior to implicit schemes if solution is to be carried out on 'virtual
memory' computers where access of peripherals (required for full equation
solutions by implicit methods) is relatively expensive.
Finally, we consider the choice of an appropriate strain- or workhardening model for elastoplastic calculations. It is logical to expect the
kinematic hardening model, which can accommodate a Bauschinger effect,
to provide more realistic results in view of the stress reversals that can
generally occur in a material under dynamic conditions. The results of a
comparative study of various hardening laws are reported in reference 31
and arrive at the following conclusions. Isotropic hardening generally
provided the best agreement with experiment and the kinematic model the
worst. The mechanical sublayer 32 and Mroz 3 3 models predicted much the
same response. Consideration must be given to the type of material and the
conditions of application before a final choice of hardening law can be
made.
REFERENCES
1. Bathe. K. J. and Wilson. E. L. (1976). Numerical Methods in Finite Element
Analysis, Prentice-Hall. Englewood Cliffs, N.J.
2. Goudreau, G. L. and Taylor. R. L. (1972).'Evaluation of numerical integration
methods in elastodynamics". Comp. Meth. Appi. Mech. Eng.. 2, 59-67.
151
3. Weeks, G. (1972). 'Temporal operators for non-linear structural dynamic

problems', J. Eng. Mech. Div., Proc. ASCE, 98, 1087-1103.
4. Wu, R. W. H. and Witmer, E. A. (1973). 'Non-linear transient responses of
structures by the spatial finite element method', AIAA Jl, 11, 1110-17.
5. Nickell. R. E. (1973). 'Direct integration methods in structural dynamics', J.
Eng. Mech. Div., Proc. ASCE, 99, 303-17.
6. McNamara, J. F. and Marcai, P. V. (1973). 'Incremental stiffness method for
finite element analysis of the nonlinear dynamic problem" in Numerical and
Computer Methods in Structural Mechanics (Ed. S. J. Fenves, N. Perrone.
A. R. Robinson and W. C. Schnobrich), pp. 353-76, Academic Press, New
York.
7. Tillerson, J. R. and Haisler, W. E. (1971). 'DYNASOR IIa finite element
program for the dynamic nonlinear analysis of shells of revolution", SC-CR-706169, Sandia Laboratories, Albuquerque, New Mexico (also Rpt. 70-19,
Aerospace Engineering Department, Texas A&M University).
8. Belytschko. T.. Holmes, N. and Mullen, R. (1975). 'Explicit integration
stability, solution properties, cost' in Finite Element Analysis of Transient
Nonlinear Structural Behaviour (Ed. T. Belytschko el al.), ASME, AMD-Vol.
14, New York, pp. 1-21.
9. Zienkiewicz. O. C. ( 1977). The Finite Element Method, 3rd edn, McGraw-Hill,
London.
10. Hinton. E.and Owen, D. R.J. (1977). Finite Element Programming, Academic
Press, London.
11. Desai. C. S. and Abel. J. F. (1972). An Introduction to the Finite Element
Method, Van Nostrand Reinhold, New York.
12. Houbolt. J. C. (1950). 'A recurrence matrix solution for the dynamic response
of elastic aircraft". J. Aeronaut. Sci., 17, 1450-1.
13. Bathe. K . J . and Wilson, E. L. (1973). 'Stability and accuracy analysis of direct
integration methods'. Int. J. Earthquake Eng. Struct. Dynam., 1, 283-91.
14. Newmark. N . M . (1959).'A method of computation for structural dynamics', J.
Eng. Mech. Div., ASCE, 85, No. EM3, July, 67-94.
15. Bathe. K.J.. Ramm, E. and Wilson, E. L. ( 1975).'Finite element formulations
for large deformation dynamic analysis'. Int. J. Num. Meth. Eng., 9, 373.
16. Bathe. K. J. and zdemir. H. (1976). 'Elastic-Plastic large deformation static
and dynamic analysis'. Computers Structures. 6(2). 81.
17. Bathe. K. J. (1976). 'Static and dynamic geometric and material nonlinear
analysis using ADINA'. Report 82448-2. Acoustics and Vibration Laboratory,
Mechanical Engineering Department. Massachusetts Institute of Technology.
18. Zienkiewicz, O. C. and Nayak, G. C. (1971). 'A general approach to problems
of large deformations and plasticity using isoparametric elements". Proc. 3rd
Conf. Matrix Methods in Structural Mechanics, Wright-Patterson A.F. Base.
Ohio. pp. 881-928.
19. Wood, R. D. (1973). 'The application of finite element methods to
geometrically nonlinear structural analysis', Ph.D. Thesis, University of Wales.
20. Owen. D. R. J. and Hinton, E. (1979). Nonlinear Finite Element Analysis:
Theory and Practice, Academic Press, London. (In press.)
21. Hill, R. (1950). Mathematical Theory of Plasticity, Pergamon Press, Oxford.
22. Perzyna, P. (1966). "Fundamental problems in visco-plasticity', in Recent
Advances in Applied Mechanics. Academic Press, New York.
152
D. R. J. OWEN
23. Zienkiewicz. O. C. and Cormeau. I. C. (1974). 'Viscoplasticity, plasticity and

creep in elastic solidsa unified numerical solution approach'. //;/. J. Num.
Meth. Eng., 8, 821-45.
24. Zienkiewicz, O. C , Owen, D. R. J. and Cormeau. I. C. (1974). 'Analysis of
viscoplastic effects in pressure vessels by the finite element method", Nucl. Eng.
Des., 28(2), 278-88.
25. Kanchi, . ., Zienkiewicz, O.C. and Owen, D . R . J . (1978).'The visco-plastic
approach to problems of plasticity and creep involving geometric nonlinear
effects', Int. J. Num. Meth. Eng., 12, 169-81.
26. Owen. D. R. J.. Prakash, A. and Zienkiewicz. O. C. (1974). 'Finite element
analysis of nonlinear composite materials by use of overlay systems'.
Computers Structures, 4, 1251-67.
27. Pande. G. ., Owen, D. R. J. and Zienkiewicz. O. C. (1977). 'Overlay models in
time-dependent nonlinear material analysis', Computers Structures, 7,435-43.
28. Tillerson. J. R. and Stricklin, J. A. (1970). 'Numerical methods of integration
applied in the nonlinear dynamic analysis of shells of revolution". NASA-CR198639. National Aeronautics and Space Administration. Washington. DC.
August.
29. Wu. R. W. H. and Witmer. E. A. (1972). 'Finite element analysis of large
transient elastic-plastic deformations of simple structures, with application to
the engine rotor fragment containment/deflection problem'. NASA-CR120886. National Aeronautics and Space Administration, Washington DC
(also ASRL TR 154-4), January.
30. Park, K. C. (1975). 'Evaluating time integration methods for nonlinear
dynamic analysis'. Infinite Element Analysis oj Transient Nonlinear Structural
Behaviour (Ed. T. Belytschko et al.). ASME. AMD-Vol. 14. New York.
pp. 35-58.
31. Hunsaker. ., Jr, Vaughan. D. K. and Stricklin. J. . (1979).
'Theoretical-experimental comparison of four hardening rules', J. Pressure
Vessel Techno/.. ASME. (In press.)
32. Duwez. P. (1935). 'On the plasticity of crystals'. Phys. Rev.. 47(6), March 15,
494-501.
33. Mroz, Z. (1969). 'An attempt to describe the behaviour of metals under cyclic
loads using a more general workhardening model". Acta Mech., 7(2-3),
199-212.
34. Stricklin. J. A. and Haisler. W. E. (1977). 'Formulations and solution
procedures for nonlinear structural analysis'. Computers Structures, 7, 125-36.
Application of Transient Dynamic Numerical

Methods to Problems in Fracture Mechanics
D. R. J. OWEN
University of Wales, Swansea, UK
1.
INTRODUCTION
In recent years the problem of dynamic crack propagation in solids has

attracted considerable attention. Quantitative studies were originally
restricted to simple situations in which a one-dimensional model was
employed for the simulation of fracture conditions. Recent developments in
numerical techniques for dynamic transient stress analysis coupled with the
availability of high-speed digital computers have ensured that more
realistic models can now be employed in crack propagation studies.
The simplest model for dynamic fracture investigation is the doublecantilever beam (DCB) employed in the standard testing of the fracture
toughness of metals (ASTM E399-70T). Such a specimen has been
employed by most investigators concerned with dynamic fracture
phenomena, and a brief review of work reported to date is provided in
Section 3.
A problem of considerable practical interest in dynamic fracture is that of
crack propagation in pressurised pipelines. Two non-linear effects can
contribute significantly to the crack propagation and arrest phenomena in
such structures. Firstly, plastic flow of the material in the vicinity of the
crack tip can influence the fracture conditions markedly; and secondly, the
gross deformation of the pipeline geometry in the cracked region generally
has a profound effect on the gas depressurisation response.
Dynamic transient finite element analyses are necessarily expensive with
regard to computer costs, and any savings that can be made in this area are
of prime interest. This is particularly important in the fully threedimensional situation of a propagating crack in a pipeline. The largest time
step which can be employed in explicit time integration of the equations of
motion is directly related to the least distance between any two adjacent
153
154
D. R. J. OWEN
nodal points in the structure and introduces a severe limitation when

relatively thin walled pipes are solved for by the use of threedimensional
elements. This difficulty can be alleviated by the use of shell elements, and
the recently developed semiloof thin shell element can be used to advantage
in this context. Much larger time steps can be employed with this element,
and accurate results in both linear and nonlinear situations can be achieved
with relatively coarse meshes.
The main disadvantage of employing conventional plasticity expressions
in dynamic transient calculations is that the classical theory does not permit
stress levels to instantaneously rise above the yield value. This
experimentally observed phenomenon can be included numerically by the
use of viscoplastic theory in which timerate effects are accounted for.
In this chapter we first consider the basic concepts of fracture mechanics
and then review briefly the dynamic transient crack propagation studies
reported to date. A fter summarising the main details of the central
difference explicit time integration method, as applicable to isoparametric
elements, we describe application to, firstly, the case of a DCB specimen
and then to crack propagation in a pressurised pipeline by use of three
dimensional solid elements. Use of the semiloof thin shell element in large
deformation viscoplastic transient dynamic situations is then described,
and finally application of the shell formulation to two pipeline fracture
situations is discussed.
2.
BA SIC CONCEPTS OF DYNA MIC FRA CTURE MECHA NICS
A review and critical appraisal of solution techniques in dynamic fracture

mechanics has been undertaken by Kanninen' and much of the material in
this section is drawn from that work.
For the initiation of crack growth, the procedures of linear elastic
fracture mechanics (LEFM) can only give approximately correct
predictions for materials, and plasticity effects must be introduced to give
precise results. However elastoplastic techniques have not yet been
developed in a completely acceptable manner even under static conditions.
Consequently the process of rapid crack propagation and arrest in
structures is generally treated only in terms of LEFM concepts.
We commence by introducing a timedependent stress intensity factor,
K, defined, just as in the static case, as
K = lim yfnr ay(r, 0, t)
(1)
TRANSIENT DYNAMIC NUMERICAL METHODS
155
where (/\ , t) is the time-dependent normal stress component expressed

in terms of polar coordinates centred at the crack tip. Equation ( 1 ) holds for
dynamic conditions, since the nature of the singularity does not change
when crack propagation occurs. The conditions governing crack growth
can be expressed in terms of an experimentally determined critical value of
K, which is assumed to be a material property. Under static conditions we
have K = Kcas the condition for crack instability. In the dynamic case Kc
has two counterparts. First, for the initiation of crack growth,
(,,0
= <)
(2)
where a represents the crack length, is the applied traction, / is the time
and the loading rate. Then, for a propagating crack,
K(a, a, t) = K0()
(3)
where denotes the crack velocity. If the parameters Kd and KD are known,
no further experimental fracture data are needed to obtain dynamic
fracture mechanics solutions.
An equivalent formulation of the conditions of crack propagation can be
made in terms of the strain energy release rate, G. Then the criterion for
dynamic crack propagation can be expressed as an equality between G and
its critical value, the energy dissipation rate required for crack growth.
Denoting this latter quantity by R, we have the crack propagation
condition:
G(a, , /) = R()
(4)
For a propagating crack, arrest will occur at time t. when, and only when,
G < R for all / > E,. The dynamic energy release rate (or the driving force
for crack extension) can be expressed as
1 (dW
du
dT\_
1 (dW
du
~b{cia~~dli~fa]~fa\cTr~dl~~&\
dT)
()
where u is the strain energy, Fis the kinetic energy, IF is the work done on
the structure by the external loads and b is the plate thickness at the crack
tip. Although eqn. (5) represents a global condition, G can be given a local
crack tip interpretation, since the dynamic energy release rate can be
directly connected to the dynamic stress intensity factor as
C=
-^-()2
(6)
156
D. R. J. OWEN
where E and are, respectively, the elastic modulus and Poisson ratio, and
plane strain conditions have been assumed. Also, () is a geometryindependent function of the crack speed, given by
a^2
(1 _
v)[4(i
(1
2
-a2/c12Y'
.
_ a2/c2)12(\ - a2lc\y<2 - (2 - 2/c)2]
(7)
where c, and c2 are, respectively, the dilational and shear wave speeds in the
material.
As a consequence of eqn. (6) we have, for linear elastic conditions,
f
M^W
V'2
(8)
Thus both Kt and R express the material's resistance to cracking and are
taken to be material properties which are independent of geometry and
applied loads. Equation (4) is attractive, since its basis lies in a fundamental
energy balance and can be readily extended to nonlinear situations. An
alternative criterion which has been employed for dynamic crack
propagation, in view of its simplicity, is one based on a limiting value of the
normal tensile stress at the crack tip. 2 Crack initiation is assumed to occur
at a stress level , and continuing propagation takes place if the stress
component normal to the cracking plane at the crack tip attains a value
uc = <7 The factor 5 can be interpreted as a measure of the initial
bluntness of the crack tip, and a c quantifies the material toughness.
3.
REVIEW OF PREVIOUS WORK
In this section we briefly review some of the most prominent numerical

investigations that have been undertaken on dynamic fracture problems.
One of the earliest attempts at simulating dynamic crack propagation in a
DCB specimen was by Kanninen. 3 4 who included transverse shear as well
as lateral inertia effects in a Timoshenko beam model supported on a
generalised elastic foundation. The energy balance criterion, eqn. (4), was
employed for crack propagation. Schmuely and Peretz, 2 adopting a critical
stress criterion, employed finite differences for both time and spatial
discretisation in their twodimensional DCB solution. Kobayashi and
157
Mall 5 investigated the problem with the aid of a dynamic photoelastic

model and found a significant degree of consistency between the numerical
and experimental results.
Keegstra, Head and Turner 6 developed an implicit time integration
transient dynamic finite element model and applied it to simulate a
propagating crack in a DCB specimen. The crack growth criterion was
based on the crack tip nodal force, which is proportional to the stress
intensity factor of eqn. (1). Elastic material behaviour is assumed, and to
provide an energy sink, the crack tip node is not suddenly released when a
critical value is achieved. Instead the force is reduced gradually so that work
is done (and energy dissipated) at the crack tip. The authors physically
justify use of these 'holding back' forces, which may exist at several nodes
behind the crack tip, by comparing them to the plastic links in Dvorak's
brittle fracture model. 7 Keegstra et al. claim encouraging agreement
between the computer predictions and the experimental measurements.
A similar study has been reported by Malluck and King, 8 who also
introduce energy dissipation through a gradual release of the crack tip
node. The model differs from Keegstra's approach in that crack tip
positions between two adjacent nodal points are considered and the
restraining force associated with crack tip node release is limited to one
nodal point. A discussion of the factors, including mesh size and nodal
release process, influencing the stability of dynamic fracture solutions has
been presented by Rydholm, Fredriksson and Nilsson. 9
For the practical problem of crack propagation in pressurised pipelines,
Fearnehough 10 reviews the available literature on both brittle and ductile
situations. Among the full-scale test results available are those undertaken
at Battelle 11 and British Gas Corporation. 1 2 Experimental work on
aluminium and steel models for simulating ductile crack propagation has
been carried out by Shannon and Wells, 13 while Alder' * performed model
studies for determining the pressure distribution at and behind the crack tip
to corroborate his numerical predictions obtained by gas dynamics
principles. 15 Theoretical models have been presented by, among others,
Hahn et al.16 and Poynton and co-workers.' ' , 1 8 Further full-scale tests are
reported by Poynton and Fearnehough, 19 who discuss the stability of
constant-velocity propagating cracks. All analytical models referred to
above are based on energy balance criteria and pertain to constant velocity
crack propagation situations, and as such are quasi-static. Kanninen,
Sampath and Popelar 20 consider the steady state constant velocity
problem, including inertia and gas dynamic effects as well as plasticity
behaviour in the vicinity of the crack tip. Ayres 2 ' has predicted the transient
158
D. R. J. OWEN
opening profiles of a pipe having a stationary crack tip using a thin shell,
finite element, large elastoplastic deformations computer program.
4.
TRA NSIENT DYNA MIC FINITE ELEMENT A NA LYSIS BY

EXPLICIT TIME INTEGRA TION
In this section we summarise the basic expressions for explicit time

integration by the central difference scheme. The dynamic problem
invariably gives rise to a set of ordinary differential equations of the form 22
M + C + P(a) + F(/) = 0
(9)
in which dots denote differentiation in time; a stands for a set of parameters

describing the displacement; M is the mass matrix (generally independent
of time and displacement); C is the damping matrix, which may well be a
function of ; is the internal set of forces opposing the displacements of
the structure; and, finally, F is the activating force set.
The ordinary differential equations (9) can be solved in a variety of ways,
including various finite difference and finite element (in time) approxi
mations proceeding in small time steps A t. The latter have not yet shown
their merit and, in general, finite difference schemes of various kinds are
most widely used. Most of these schemes require a repeated solution of a
full set of simultaneous equations at every time step, and, especially if these
equations are nonlinear, a large amount of computing time and complex
program organisation is required. There is one notable exception, however.
This is the most obvious timecentred finite difference approximation to
eqn. (9), which can be written as
M(a + , 2a + a_ ,)/(/) 2 + C(a + , a_ ,)/(2/) + P + F = 0
(10)
where suffixes /; + \.n,n 1 stand for three successive time intervals. If
values of a and a_ ,. etc.. are known, the value of a,,+ , can readily be found
as
a + , = ( + //2) 1 [(/) 2 ( F) + 2Ma (M CA //2)an_ ,]
(11)
If the matrices M and C are diagonal, the calculation at every time step is
trivial and in the finite difference parlance this forward integration scheme
is called 'explicit'. The 'explicit' scheme, while possessing considerable
merit, can be shown to be unstable when / exceeds a certain critical value.
TRANSIENT DYNA MIC NUMERICA L METHODS
159
and its advantages are counterbalanced by the large number of time steps
required. Further, in the usual finite element approximation the matrices M
and C do not appear generally in a diagonal form and a further
approximation known as 'lumping' must be introduced to make the scheme
viable.
For many problems involving short transients (i.e. very rapid
loadingdeformation sequence) extremely efficient solutions using such a
scheme can be obtained, and fast fracture problems generally fall into this
category. The internal forces for use in eqn. (11) are most conveniently
determined from
B T adr
P., =
(12)
in which <r are the current stresses which satisfy all nonlinear conditions in
the structure. For linear elastic situations obviously
P = Ka
(13)
in which K is the structural stiffness matrices.

4.1. Time-stepping Scheme and Organisation of the Program
The basic recurrence algorithm is given by eqn. (11), and it is immediately
noted that if the values of a, a_ ,, P and F are known, calculation will
progress in a direct manner.
Indeed, if the matrices M and C are diagonal and if a, denotes the /th term
of the a vector, etc., we can rewrite eqn. (11) simply as
1
M + C'A t/2
l(At)2(P^
F',) + 2M,a'l (M, C",A l/2)an, ']

(14)
and the computation is direct provided that the forces P" and F" can be
determined. In the finite element context (and, indeed, in any discrete
structural assembly) the total 'force' is the summation of element
contributionsthat is,
m
rn
P,.= Y/>,,;
Ft = Y Ft.,
<= 1
(15)
where the suffix e stands for a particular element. It is therefore immediately
160
D. R. J. OWEN
obvious that it is not necessary to assemble the full vector of forces before
the calculation beginsand that the computation can be carried out'point by
point' to determine a"+1.
At any time stage the sequence is, therefore:
(1)
(2)
(3)
(4)
Calculation of and assembly of forces P" and F" for a node ; = 1.

Computation of a" +1 from known values a" and a"'1.
Repetition of the process for i = 2, etc.
Computation of P"+ ' and F"+ ', etc.. for the next time step.
The storage requirements are thus directly proportional to the number of

unknowns. /and not as when an inversion is necessary to r o r / 3 . Further,
the computational time is also directly proportional to ;. Herein lies the
great computational simplicity of the process.
It is noted that to start algorithm (14) the values of a" and a"~ ' are
necessary. Often at the start of a dynamic problem the values of af and J'
(the velocity) are prescribed and apparently a starting difficulty exists.
However, writing
a = (a -~1)/(2)
(16)
allows a special algorithm to be written for the first time step by eliminating
1 from eqn. (14).
In practice we seldom find it necessary to use this starting algorithm if.
say before the commencement of the transient, the structure was at rest and
all the displacements prescribed.
4.2. Evaluation of the External and Internal Forces
With standard finiteelement notation employed, thedisplacement at any
point of the continuum is approximated by the expansion
a = "',., =
(17)
the incremental strains incurred during a time step can be written

= 3
(18)
and the accumulated (total) stresses are

,^,,^+,,
(19)
where
=
(20)
161
On use of the principle of virtual work, the appropriate terms of eqn. (9) are
found to be
N T pNdu
(21)
N T CNdi'
(22)
P.. =
(23)
F., =
N'bdr +
.du
NTd dS
(24)
(25)
Aa"
in which is the material density, C stands for velocity proportional

damping forces, b are the body forces per unit volume and d are the
boundary tractions. All integrations are performed numerically with Gauss
quadrature formulae.
4.3. Large Deflection and Plasticity Effects
When large deflections occur, but the strains still remain tolerably small,
constitutive equation (20) is still valid. If a Lagrangian strain description is
adopted, in which quantities are referred to the original reference system,
then the strain matrix, B, takes on a non-linear form dependent on the
current configuration but all integrals can be based on the original volumes
that is,
(26)
B = B(a)
The exact form of the matrix B for three-dimensional situations can be

found in reference 23.
If elastoplastic material behaviour is allowed, the matrix D in eqn. (20)
becomes stress dependent and is replaced by ~-ep>
D t p , whose precise form is 2 3
~ t"
D.
D-D
OF
ca
OF
ca
~F~ ~F
< + ca
D
t'a
(27)
in which A is a known strainhardening parameter. The quantity F i s the

yield function, which is of the general form
F(ff,ep,ffy) = 0
(28)
in which a is the uniaxial yield stress and <Tp the effective plastic strain. For a
162
D. R. J. OWEN
shell formulation the yield function is based on the stress resultants, and
particular forms are described in Section 7.2.
The process of computation proceeds, therefore, as follows:
From steps (l)(3) of the previous sequence, a is found by use of
the elastic matrix, D.
(2) A check is made to determine whether a is above the yield limit. If it
is not, computation proceeds to step (4) of the previous sequence; if
yes:
(3) The elasticplastic matrix, D e p , is computed by use of _,.
(4) Steps (l)(4)ofthe previous sequence are carried out with D = D t p .
(1)
A slight elaboration is necessary in principle if _ , is below the plastic

limit. The matrix DL.p is then computed with a part of the increment allowed
to be made in elastic mode and part in elastoplastic mode. Details of such an
elaboration are given in reference 24, but it is found that in using small time
steps this procedure is not essential.
Little has to be said concerning the computation of F. If the forces band
d are conservative, then their magnitudes are simply calculated by eqn. (24).
the original element coordinates being used. However, if the forces are not
conservative (i.e. if, for example, a pressure loading is applied to grossly
deforming boundaries), then the computation given by eqn. (24) must be
performed on the deformed geometry of the element. This modifies not only
the regions of integration, but also the shape functions themselves, which
will depend now on updated element coordinates.
4.4. Lumping of the Mass and Damping Matrices
The inertia force vector is found, for given shape functions, by the matrix
M given in eqn. (21). This matrix, unfortunately, is not diagonal and the
essence of the explicit marching scheme requires it to be diagonalised. It is
possible to achieve such a diagonalisation by the use of substitute shape
functions. , which have values of unity over some section of the element
adjacent to a node and are zero elsewhere. With no overlap of such
functions all terms containing products are zero and only diagonal terms
remain. This proposal for 'lumping' was made by Clough.2"1 and later Tong.
Pian and Bucciarelli 26 proved it to be convergent. The important condition
which has to be satisfied, however, is that at all points within any element
li
= 1
(29)
This is equivalent to requiring the total mass of the element to be preserved

and within this constraint permits many possible systems of lumping.
163
For the quadratic elements used here several alternatives were

investigated by Rock 2 7 and Hinton, Rock and Zienkiewicz, 28 and the most
efficient one at the present time appears to be offered by lumping in
proportion to the diagonal terms of the original (consistent) mass matrix.
We proceed thus as follows: First the consistent mass matrix of each
element is evaluated by expression (21), the diagonal terms M)] being noted.
From this the element diagonal terms, M, are computed as
M, =
M;;.=
(30)
in which
Ml, =
,,;
dv
(31)
When all element diagonal terms have been determined, we find the total
diagonal term by summationthat is,
M, =
> M"
(32)
= 1
This form of lumping has been shown to result in an accuracy improvement

for many vibration problems and also has been shown to perform equally
well in transient situations. 22 In order to take advantage of the explicit time
stepping scheme, it is convenient to assume that the viscosity of each
element is proportional to its density. Then
c, = ,
(33)
where is a proportionality constant.

4.5. Time Steps Required for Stability
As mentioned before, the stability of the central difference scheme is
conditional. It is well known that this should be a fraction of the smallest
natural period associated with the discretised structure. An estimate of such
a period can be obtained by studying the eigenvalues of elements, and it is
related to the smallest time taken for elastic waves to traverse the element.
If we note that for an isotropic elastic body the wave speeds are
c2=
(34)
164
D. R. J. OWEN
for dilatation waves and shear waves, respectively, and since always the first
of these is the higher one, the time step is given by
, = y'L"lcx = )'"
p(\ + v)(l - 2 v )
(1 - v) "
(35)
where y' is some number less than unity and L" is the smallest distance
between adjacent nodes of any parabolic element used. Rock 27 finds that
y' = 0-45 is a safe value to use, and the present studies corroborate the
point. Indeed, for many problems y' values in excess of this limit provided
stable solutions, the maximum observed being 0-5.
If linear elements are used, then for stable solutions a value of 0-91 for
y' was found suitable.
These limitations can often be imposed with some severity even if one
small size element is included in the analysis. The choice of mesh
subdivision is thus critical.
5.
NUMERICAL ANALYSIS OF THE DOUBLE-CANTILEVER

BEAM
Figure 1 shows the DCB specimen to be considered where one of the finite
element meshes employed is illustrated. Owing to symmetry only one half
of the solid need be analysed. The loading-pin holes are ignored in solution,
and instead selected nodal points are assumed to be displaced a prescribed
L = 360 72 m m
@>
23 1mm
12
. : 92 <mm
CD
22 36 m m
= 268 32 mm
each
Mesh "B"
8 parabolic elements ; 183 nodes
E = 2 128 IO 4 k g / m m 2 . = 0 273
= 0 7 9 I O - 9 kg s 2 / m m '
At = 0 75 -
FlG.
Details of mesh for DCB problem (plane stress).
165
distance apart. As the loading rate is slow, static conditions can be assumed
before crack propagation commences. Numerically the steady state
deformed configuration is obtained by performing a transient dynamic
analysis and employing a suitable mass-proportional damping term as
described in Section 4.4. Node 100 at position (H/2,L) is subjected to a
prescribed lateral displacement of H. This lateral steady state displacement
is consistent with the values assumed in references 2 and 4, and is employed
here for comparison purposes. In order to avoid numerical instability
associated with the large mesh distortion which takes place, this
displacement is assumed to be achieved in a selected number of time steps
(100 for the problems presented) and to be applied at a constant rate. Steady
state conditions are deemed to have been achieved when the change in
displacement norm for successive time steps of all nodal points is
sufficiently small.
5.1. Stress Criterion of Crack Propagation for DCB Specimen
Once steady state conditions have been achieved, the normal stress
component. , at the Gaussian integrating point nearest the crack tip is
designated . To initiate crack propagation the nodal restraint at the crack
tip (node number 25) is removed and the damping coefficient now set to
zero. The crack tip is assumed to have now progressed to the next node
(node number 24) and propagation is governed by the ax stress value at the
integrating point immediately adjacent to the new crack tip position. If this
stress exceeds ac = JS, then the crack is assumed to propagate to the next
node. In the problems presented values of S ranging from 1-0 to 2-5 have
been considered. In each case a plot of crack growth against time is
obtained and a best straight line fit employed to determine the average
velocity.
Two different geometries and finite element meshes are considered.
Figure 2 shows the relationship between crack growth and time for the
geometry and material properties indicated in Fig. 1. For the steady state
solution a damping coefficient of 12000s^ 1 was employed and three
different fac values are considered. The relative arrest length, a/a0, and
the non-dimensionalised crack velocity, v/c0 (where c0 = ^JEjp), are
recorded in Table I and plotted in Fig. 3, where comparison is made with
other available solutions. 2,4
5.2. Energy Balance Criterion for DCB Crack Propagation
In this approach three energy terms must be essentially considered. These
are the internal strain energy, the kinetic energy of the system and the
166
D . R. J. O W E N
(Wo c =2 S)
"
t_ (Assumed
A
jS
a
<S I(""Vo
c=
>r
j f s ^ *
arrest point )
17 )
%= stress
at
oc=
stress
critical
crack
(propagation)
for
subsequent
initiation
propagation
(Arresi)
< ^ ^ l 0 / 0 < : = 1 Ol
0
40
ao
120
160
200
20
Time ( M- s)
FIG. 2. Computed crack growthtime plots using stress criterion for different
aqaQ values (DCB. mesh B).
energy absorbed at the crack tip as the crack propagates. It is noted that the
external work done is zero in this case, since the loading takes the form of a
constant prescribed displacement.
The strain energy, u, at any instant can be calculated from
,.., = + (P+, +)(3)
(36)
The kinetic energy contribution, K, at any instant can be directly calculated
from
* + i = ( + 1 ) T M n + 1
(37)
TABLE I
SUMMARY OF RESULTS FOR DCB PROBLEM
No.
1
:
3
4
5
6
7
8
Result
(as piott ed
in Fig. 3)
O
O
C
D
E
F
Ci
Mesh
used
A
A
B
B
B
B
B
B
Criterion
Stress S= 1125
Stress 5 = 1375
Stress 5 = 1 0
Stress 5 = 17
Stress 5 = 25
Energy R = 16 714 kg/mm
Relative
arrest
length.
a/a0
Crack
velocity.
v/c0
214
266
036
317
378
148
197
245
0125
1160
0096
0174
0227
0060
0118
0174
167

Experimental
[2]
Kanninen - energy
Sn
h im
i iu
ue
e li yy
Present
&
&
formulation i
(See
table
criterion
[21
Peretz
r e r e t z -- ss tt rr e
e ss ss
O
o
o
--
for
Energy
S
S tt rr e
e ss ss
Stress
c rnitteerr i o n
e
c r i t e r iion
ce rni tt ee rr ii co n
e n t e r iion
c
(Mesh
'
(Mesh
'A')
(Mesh
'B')
details)
0 15
0
Relative
20
arrest
3 0
length
/a.
FIG. 3. DCB problem. Plot of crack velocity against relative arrest length.
Then from eqns. (4) and (5) we have
d
dK
- + + bR = 0
da
da
(38)
In the present numerical analysis da is interpreted as the distance between

two adjacent nodes, and the crack will propagate from one node to the next
provided that
(du + dK)
(39)
> R
b da
The computational procedure is as follows. The steady state strain energy,
168
D. R. J. O W E N
u 0 , and kinetic energy, K0, are calculated. (In fact, K{l = 0.) The damping
term is then eliminated, the restraint at the crack tip removed and u and K
determined for each time step. The quantity ((w0 + K()) (u + K))/b da is
calculated and propagation of the crack to the next node allowed when this
term attains the specified value of R. Similar to <xc in the stress criterion, is
a measure of the specimen toughness under dynamic conditions.
250
-i
(Arrest )
150
(Arrest)
o
g"
100
(R= 11K3 k g / m m
iO
(R = 1 6 7 U
kg/mm)
80
120
(Arrest)
160
200
Time ( s)
FIG. 4. Computed crack growthtime plots using energy balance criterion for
different toughness (R) values (DCB mesh B).
The energy change (du + dK) required for crack advance by the energy
criterion can be produced in two ways:
(1)
(2)
If the material behaviour is assumed to be elastic, an energy sink

can be provided by gradually releasing the crack tip node as soon as
a critical energy level is reached. In this way work is done (and
energy dissipated) at the crack tip.
If elastoplastic material behaviour is assumed, plastic deformation
in the crack tip region automatically provides the necessary energy
sink.
In Fig. 4 the relationship between crack growth and time is presented for
three toughness values, and assuming elastic material behaviour. A gain
a/a0 and v/c0 values are recorded in Table I and incorporated in Fig. 3.
Good agreement between the eight points A from the present
169
computation and the experimental and numerical results of references 2

and 4 is evident.
6.
DUCTILE CRACK PROPAGATION IN PRESSURISED

PIPELINES
The situation leading to dynamic crack propagation in a pressurised

pipeline can be summarised as follows. A short axial crack is initially
developed along a generator of the pipe. Under the action of the internal
pressure, P{, the crack begins to open out, as shown in Fig. 5, which results
in a pressure drop to a value P0 at the crack tip. Although relationships
between P] and P0 have been developed, little information regarding the
time in which this pressure drop occurs is available. A pressure drop also
occurs along the cracked zone, the value of which is a function of the axial
distance, Z, and the angle of crack opening, 20. This pressure variation has
been quantified 1 5 1 8 ' 2 0 in previous investigations but only under constantvelocity steady state assumptions.
Thus, in reality the system is a coupled fluid-solid problem, with the pipe
wall deformations and the gas pressure variation being interdependent. In
10
p,f'"
V
^ ^ V s _
X- 0 5
P.e-'
0
I1
3
t(m.s)
FIG. 5. Opening profile of a pressurised pipe with an initial crack and the pressure
distribution in the pipe.
-J
C
3
^C
( 20 noded I elements
er of nodes 254
rule 2 H 2 2 gauss
FIG. 6. Mesh used for dynamic large-deformation clastic-plastic three-dimensional analysis of a pipe (with a stationary or
propagating axial crack) under pressure loading.
171
this work the problem is uncoupled and the steady state gas pressure
distribution is assumed to exist at all times, which clearly introduces an
element of approximation, particularly at commencement of crack
propagation. Furthermore, the pressure distribution is assumed to be
axisymmetric with respect to the pipe axis. The initial pressure, P is
assumed to drop linearly to a value P0 in a specified time and to remain
constant at this value thereafter. This is achieved by use of an appropriate
function of time J(t) (see Fig. 5) in the relation
P0 = Pc JO)
The variation in pressure in the cracked zone is taken to be
= p o e '^.,)
(40)
18,20
(4 j)
where is a constant (assumed as 228 in the present study), 0 is the half

angle of crack opening measured in radians, R, the internal pipe radius and
Z c the axial distance measured from the current position of the crack tip.
The pressure at and ahead of the crack tip is assumed to be P0.
The case of a propagating crack in a pipeline was first studied by use of
the finite element mesh shown in Fig. 6. The pipe geometry is indicated and
the quartersegment to be analysed was modelled by a total of 28 three
dimensional isoparametric elements. The solution sequence followed is
summarised below.
(1)
(2)
The static solution for the pressurised pipeline in the absence of any
crack is first obtained. This is achieved by means of a transient
dynamic analysis with a damping factor approaching the critical
damping value.
The damping factor is then set to zero and an initial (or starter)
crack of predetermined length is introduced along the axis of the
pipe. Numerically this is achieved by the sudden release of selected
nodal points. Physically this situation corresponds to accidental
pipeline damage and is the method by which a propagating crack is
usually introduced in experimental studies. A t this stage depres
surisation as well as axial pressure variation are accounted for
according to expressions (40) and (41).
The length and location of the starter crack are indicated in Fig. 6 and
elastic largedeformation behaviour was assumed. A stress criterion for
crack propagation was first assumed, and Fig. 7 shows the crack growth
with time for three different cases.
The problem was then reanalysed, an energy balance crack propagation
172
D . R. J. O W E N
32.000 I b / i n *
tip reached end ot pipe section)
(crack
aq = 70.000 lb/in 2
q/oc
=15
(Arrest)
p.= 2250 l b / i n 2 . p i t ) = p ( i . e . t l
E = 27 5 IO 6 lb/in 2 . V--03
= 0 734x10" 3 lb s 2 / m ' , e. =-/^7,
R = 15' ; t = 2 5" , L = 7 5 " : a.= 30"
= 2 5ps
o .y = stress at crack (propagation) initiation
Oc = critical stress for subsequent propagation
It) = half angle of crack opening
WO
crack
500
growth
600
Initiation
FIG. 7. Dynamic largedeformation elastic (threedimensional) analysis of a pipe

with an axial crack under pressure using stress criterion for crack propagation (see
Fig. 6 for details of mesh used).
criterion being used, and the results are shown in Fig. 8. The calculated
crack velocities in this case are in agreement with experimentally measured
speeds for pipelines. On the other hand, the values based on a limiting stress
propagation criterion are excessive and can only be attributed to the values
of aq and assumed.
Crack propagation in a pressurised pipeline is a problem in which
5
to
35
30
E = 27 S I O 6 I b / i n 2
v = 03
= 0 734 1 0 " 3 l b . s 2 / i n 4
At= 2 5 s
Rj = toughness for initiation of crack
propagation
R d y n = toughness for subsequent
propagation
(Arrest )
0-057
200
00
Time ( s i
600 B OO
measured
from
1000
1200
crack growth
initiation
FIG. 8. Dynamic largedeformation elastic (threedimensional) analysis ol'a pipe

with an axial crack under pressure loading using energy balance criterion for crack
propagation (see Fig. 6 for details of mesh used).
173
plasticity and large-deformation effects are important. It can be shown that

the extent of deformation is significantly greater when elastoplastic
material behaviour is permitted than for the purely elastic case.
Furthermore, the deformations in the cracked region are large and the
gross changes in geometry which take place influence the gas pressure
loading considerably.
The theory of viscoplasticity offers the most suitable method of including
plasticity effects in dynamic transient models, since such a model includes
automatically the effect of loading rate on the material performance. Also,
considerable computational economies are offered by solving pipeline
problems by means of shell elements, and to this end we describe the use of
the semiloof shell element in dynamic transient situations when both large
deformations and viscoplastic effects are included.
7.
THE SEMILOOF SHELL ELEMENT IN NON-LINEAR

TRANSIENT DYNAMIC ANALYSIS
7.1. The Basic Element

Complete details of the semiloof element formulation can be found in
reference 29. The element has'initially' the 45 degrees of freedom shown in
Fig. 9. The degrees of freedom corresponding to the displacements of the
Conventional nodes
AZ
FIG. 9. Nodal configuration of the semiloof shell element.
174
D. R. J. OWEN
central node are combined to create a deflection normal to the element. The
rotations and normal displacement at the centre, as well as the 8 rotations
0Jy: at the loof nodes are then eliminated by imposing 11 constraints on the
shear behaviour. The quantities employed in the definition of the various
matrices involved can then be expressed in terms of the degrees of freedom
retained, a", as
(42)
G = Fa''
in which
CU CU CV CV cu cv
cu
C2u
-.2
cv
c X cY c c YCZ CZ C XC c YCZ c XC C YCZ
(43)
The terms CujcX, etc., are the displacement derivatives with respect to the
local axes. Matrix F is a 13 32 array and contains the final shape
functions for use in the evaluation of the strain and mass matrices.
Subject to the conditions of small rotations and finite membrane strains.
the strain-displacement relationship for large-displacement problems can
be written as
( ;/
< u
CX
'Z
Cv
CY
Cu cv
+
CY CX
u cv
Tzc~z
(B 0 + BL)a'' = Ba
=
2
c U
CXCZ
(44)
-< 7
cv
cYcZ
cu
CYCZ
-,
>
+-
xcz
Matrix B 0 can be formed from rows 4-7 and 10 3 of F, and B L , which

contains the non-linear terms, is formed from rows 8 and 9. For
geometrically non-linear behaviour the strain matrix, B. has to be
determined at every time step according to eqn. (44) if a Lagrangian
formulation is adopted. For the updated Lagrangian procedure new values
175
of the shape functions and their derivatives have to be determined with

respect to the local axes associated with thecurrent element position and the
linear matrix, B 0 , employed at each stage.
The consistent mass matrix is defined as
MF
N T .mNdi; +
NXNfldi'
(45)
where m = p l 3 a n d m = (ph2/12)I3; in which is the material mass density

and // the shell thickness.
Matrix NM, is a 3 32 submatrix, which contains the terms associated
with the displacements at the corner and midside nodes with the remaining
positions taken as zero; and Nu is a 3 32 submatrix, which contains the
rotation terms at the loof nodes and null values in the remaining positions.
These submatrices are such that
= N. + N u
(46)
where is the 3 32 matrix composed of the first three rows of F according

to
u , t , n | T = Na''
(47)
The lumped mass matrix is obtained according to the procedure established

in reference 28 and successfully applied in the analysis of some structural
problems using isoparametric elements. 22
7.2. Elastoviscoplasticity and the Yield Criterion
The total strain rate, , is assumed to be the sum of the elastic, c, and
viscoplastic, , components. The viscoplastic strain rate is defined
according to Perzyna's associated form of the viscoplastic flow rule 3 ' as
= )'<(>(^
<a
(48)
where a is the stress vector composed of the membrane stresses and bending
moments and F is a scalar yield function. Also,
<()>=\(
"~/;(F)>0
(49)
[0
cj)(F)<0
and y is a function of the material viscosity coefficient.
A general form of the yield function has been proposed by Crisfield 32
176
D. R. J. OWEN
which is expressed in terms of the membrane stresses (Nxx, N , Nxy) and

bending moments (Mxx, M. , MV) as
F=Q,+%
+ L I , J 1
(50)
in which
. = 4 ( ^ ; , + '2-- NxxNyy + lNl)

iv
- MxxMyy + 3M2y)
1
/O
t M
\l
..
//
(51)
2
where // = is the uniaxial yield force, M0 = 0 /4 is the uniaxial

yield moment and 0 is the uniaxial yield stress of the material. The
parameter is defined in terms of the viscoplastic curvature y^ as
a=l_ie^;
/7 = f ^
(52)
30
where fi is a constant. E is the elastic modulus and
Xsp =
i r 2 + (/;P 2 + X?xX?y + () 2 /4 ! '
(53)
For = 1 (/s p cc ) this yield function reduces to the surface proposed by

Ilyushin as an approximation to the von Mises yield surface (see reference
33). If the elastic strain rate is computed according to the generalised
Hooke's law and 0(F) = F i s assumed in eqn. (48), the total strain rate is
obtained to be
= D - ' r i + }<>
(54)
where D is the usual elasticity matrix and M is a 6 6 matrix whose explicit

form can be readily determined. Considering the Euler scheme for time
integration, the accumulative viscoplastic strain can be computed as
?+1 = ;' + ;' -,
(55)
where and /; + 1 stand for two successive time intervals and / is the time
step length.
177
7.3. Time Step Stability Limits

Stability of the solution process is governed by the limiting time step
lengths established, respectively, for the central difference process
employed for time integration of the equation of motion and for the Euler
scheme considered in the viscoplastic algorithm. The critical time step
length for bending of thin plates has been given by Leech, 34 and a modified
version is employed here. The maximum permitted step length is
L(3p(i
v2)
^*'2""|
(56)
where Ls. is the least distance between any two adjacent nodes, is
Poissoffs ratio and a, is a proportionality constant less than unity. For
membrane behaviour the critical time step length is related to the
dilatational wave speed and given by
A,
"
=g
(p(l + v)(l - 2 v ) | 1 2
^{
(i-v) "j
(57)
where a 2 is a constant less than unity (a 2 = 0-45 for parabolic elements 22 ).

The critical time step length associated with use of the Euler time
integration scheme for viscoplasticity was defined by Cormeau 3 5 for a von
Mises yield surface. Since the Ilyushin criterion, employed in the numerical
examples presented in the next section, is an approximation to the von
Mises surface, a modified form of this limiting time step length is used here.
Thus
^ - f - ^ - F .
(58,
where F0 is a reference value of the yield surface and is a constant. The

time step length employed in solution is limited to the least value obtained
from eqns. (56)-(58).
7.4. Illustrative Examples
In this section some numerical examples are provided to illustrate use of
the semiloof element in non-linear analysis. Figure 10 shows the elastoviscoplastic response of a simply supported square plate subjected to a
uniformly distributed step loading. By increasing the viscosity coefficient, y,
progressive amounts of plastic deformation are obtained but with little
change in the response frequency.
The dynamic response of an elastic spherical shell cap is shown in Fig. 11,
178
D. R. J. OWEN
F l t l = 0 005
U.DL.
i i i i i u i i i i i i m i i i i i i n u m u i i n uiiiiiii
Material properties
E =105
v= 0 3
/
h=05
;
Mo= 0
a
21 =32 0
/
M =0 035
/'
o Y' = 0 0001
y = O00C05
_ . D _ . '=
/
/
20
-0
60
80
0 0
2-0
l i -0
16 0
ISO
Time I sec I
FIG. 10. Elasto-viscoplastic dynamic response of a simply supported square plate

under uniformly distributed step loading.
where the results obtained are compared with those of reference 30. Largedeformation effects are included and solutions for both a total Lagrangian
and an updated Lagrangian formulation are presented. The considerable
difference between the present results and those of reference 30 is
attributable to the boundary conditions assumed. In reference 30 a 10-node
isoparametric element was employed with two out of three nodal points
being allowed to slide along the inclined support, whereas a fully fixed
condition was assumed in the present analysis. Further analyses employing
different boundary conditions confirmed this observation, but a direct
comparison was not possible, since the nodal distribution along the
boundary differed for the two studies.
For engineering problems the limiting time step length for the
viscoplastic algorithm, eqn. (58), is generally far greater than either of the
limits imposed by eqns. (56) and (57). For thin shells the time step length
restriction imposed by eqn. (56) is not severe, owing to the presence of// in
the denominator, and the critical time step is provided by the membrane
action restriction, eqn. (57), for most practical problems. It should be noted
that for pure fiexural action the time step limit imposed by eqn. (57) must
still be observed even though the membrane stresses will be sensibly zero.
For this reason it is advantageous to develop a plate bending element for
Large deformation . updated

semiloof shell element
Large deformation . Lagrangian

semiloof shell element
Solution
179
formulation
formulation
from ref 30
m
o ,-. o
.'
'
*
"\\
6 '<\ '.
Time
3
'
O 'i1 Ilsec.10- l
?\
.1
* I
\
\^%
9
I
\
\\
\ \
i'
/ /
rP
FIG. 11. Large-deformation dynamic response of a spherical shell cap.

dynamic analysis, since in this case eqn. (56) provides a much larger limiting
time step for plates which are appreciably thin. In fact, the solutions
illustrated in Fig. 10 were obtained by use of such a plate element and the
permissible time step length was some seven times greater than that allowed
by a full shell formulation.
8.
APPLICATION OF THE SEMILOOF ELEMENT TO

PIPELINE FRACTURE PROBLEMS
In this section we describe the numerical solution of two pipeline crack

propagation models for which some experimental results were available to
'
/ /
9.20
cc
30 00
2-2005
t 0
2000
9-60
ZONES OF VISCOPLASTIC
FLOW
t 0 - 4 I N.
303
PS
I.
rr.
7.
8-120
-180
Z.O
.J
5
10
15
I
JO
25
30
35
I I
40
45
50 I N.
FIG. 12. Finite element mesh and development of viscoplastic zones for pipe fracture modelCase I
181
TABLE II
Case I
Pipe internal radius. R,(\r\)
Pipe wall thickness, /(in)
Starter crack length, 2a0(in)
Internal pressure (lb/in2)
Elastic modulus, E (lb/in2)
Poisson's ratio,
Yield stress (lb/in2)
Mass density (lb.s2/in4)
Toughness parameter, (lb/in)
Case I
116
116
04
0389
24
22
303
175
29 IO6 29 10"
03
03
40000
40 000
0734
0734
2900
2 900
No strainhardening
the author. Details of the geometry and material properties are included in
Table II. In each case both elastoplastic material behaviour and large
deformation effects are considered. The drop in crack tip pressure is based
on the plan area of the crack configuration. Equation (40) is replaced by the
following semiempirical relationship:
A
^ = 1 _ 07
(59)
AV
in which A is the crack opening area and A p is the crosssectional area of the
undeformed pipe.
The axial variation in pressure in the cracked zone is again assumed to be
given by eqn. (41). In each case the symmetric quadrant of the model was
idealised by a total of 44 semiloof shell elements, and the developed mesh is
shown in Fig. 12.
The static solution was first obtained by carrying out a damped transient
dynamic analysis, and Fig. 13 shows that steady state conditions are rapidly
achieved. The starter crack of prescribed length was then introduced by the
release of selected nodal points, and at this stage the damping constant was
set to zero. The energy criterion of crack propagation was employed, and
the energy dissipated in the crack tip zone was directly calculated as
a1,pdvdt
(60)
in which a are the total stresses, vp the viscoplastic strain rate and the
volume integration taken over the material in the crack tip region
contributing to the cracking process. For neartip plasticity problems this
sc
t J
t = 10
SE
C S
D A M P I N G C ONSTANT - - 1 2 0 0 0 S E C '
STEADY STATE C ONDITIONS

REACHED
je
400
TIME
500
STEPS
FIG. 13. Static solution of uncracked pressurised pipe by means of a damped transient dynamic solution.
CRACK PROPAGATION
564 -
ENERGY LEVEL FOR CRACK

PROPAGATION
CASE II )
>
/.
<
>
s
'.
m
>
r
i:
m
C
D
:
3000
OF
TIME STEPS
Fia. 14. Development of energy available for crack propagation with time.
GC
SC
-U
CASE
0-4 I N
303
PSI .
t-
p
AVERAGE CRACK
VELOCI TY
(3
470
FT/SEC.
- 0-02
<4
<*H
-6 6 O
52
L
2300
2400
2500
2600
N
2700
OF
2800
TI M E
2900
3000
3100
3200
3300
3400
STEPS
Fio. 15. Computed crack growth-time plot for Case 1 pipe study.
10560
FT/SEC.
TRANSIENT DYNAMIC NUMERICAL
CASE
185
METHODS
CASE
I.
II.
t = 0 - 3 8 9 IN.
P = 1 7 5 PSl.
2150
TIME STEPS
2fl00
1IME STEPS
t = 10
2800
TIME STEPS
SEC
ALL DISPLACEMENTS M A G N I F I E D
BY 5
FIG. 16. Pipe opening profile of central section with time.

region can be defined, but when plasticity is widespread, difficulties arise in
assessing the extent of the plastic zone contributing to the fracture
mechanism. In the present calculation the fracture energy, E, is calculated
from the stress and viscoplastic strain rate at the Gauss point nearest to the
crack tip node, but for future work it is clear that more attention needs tobe
given to this evaluation process. Propagation of the crack to the next node
is allowed when
b da
(61)
in which b is the pipe wall thickness, da the internodal distance and R the
fracture toughness of the material. The variation in fracture energy with
time is shown in Fig. 14 for both cases. For Case 1 the energy builds up until
the critical level is attained and crack propagation occurs. However, for
Case II the fracture energy reaches an upper limit well below the critical
value and the induced starter crack does not propagate at all. These results
are in agreement with experimental observations, where Case I showed
considerable crack propagation and Case II none whatsoever. Figure 15
SC
TANGENTIAL
FORCE/UNIT LENGTH
12000
10000
0-A IN.
8000
3 0 3 PSI.
-6
6000
THEORETICAL
INITIAL
4000
FIG. 17. Tangential stress distribution along crack axis for Case I.
187
shows the extent of crack propagation with time for Case I, and solution
was terminated at 3000 time steps for reasons of economy. The average
crack velocity is calculated to be r/c 0 = 0 028.
The development of zones of plastic flow with time is included in Fig. 10
for Case I. and it is seen that the initially localised plastic flow becomes
w idespread as the flanks of the starter crack deform appreciably. The extent
of deformation in the cracked zone is illustrated in Fig. 16, where the
deformed shape of the central section (Z = 0) is shown for various times.
The greater deformation of Case I is immediately apparent.
Finally, the tangential (hoop) stress along the crack axis is shown plotted
in Fig. 17 for Case I for two times prior to crack extension. The relatively
small departure from zero in the starter crack region can be attributed to the
fact that the stresses plotted are the Gauss point values nearest to the crack
axis and not the values on the axis itself.
9.
DISCUSSION
The examples presented indicate that explicit time integration dynamic

transient computer codes are capa ble of modelling dynamic transient crack
propagation problems. Good agreement with results from other sources is
generally evident. However, for the stress criterion case of the DCB specimen
a small amount of crack propagation is observed when oq',ac= 10
(Fig. 2), whereas theoretically no crack propagation is possible under
such displacementcontrolled conditions. The degree of crack extension is
small in relation to that occurring in the other situations considered and is
possibly a consequence of the numerical procedure employed to achieve
steady state conditions and to then release the crack tip nodes.
For larger cJaQ ratios (i.e. smaller toughness) the deviation of points on
the crack growthtime plots from the mean velocity line was greater than
for cases of smaller ac values. This observation is in agreement with the
results of Kanninen 4 but appears to conflict with the values presented in
reference 2.
In the present study ajac and R have been assumed to be constant.
Experimental evidence exists to indicate that these quantities are crack
velocitydependent, but little published work is as yet available. The explicit
scheme employed in integration of the equations of motion make the
inclusion of such a dependence an easy task. A t any instant of time in the
analysis, the crack velocity is known. If, from experiment, R can be
188
D. R. J. OWEN
expressed a s a function of the speed of p r o p a g a t i o n , such a variation can be

readily included in the numerical c o m p u t a t i o n s .
R e g a r d i n g crack p r o p a g a t i o n in pressurised pipelines, it is difficult t o
assess the applicability of transient d y n a m i c finite element m e t h o d s at this
stage. F o r a meaningful analysis both viscoplastic a n d largedeformation
effects must be included. F o r viscoplastic material behaviour m o r e work is
clearly necessary on formulating numerical processes for fracture energy
evaluation. A t present, however, it can be said that numerical pipe fracture
studies a p p e a r to promise a m e a n s of providing information which c a n n o t
be provided by experiments.
REFERENCES
1. Kanninen. M . F . (1978). 'A critical appraisal ofsolution techniques in dynamic
fracture mechanics'. Proc. 1st Int. Conf. Num. Meth. Fracture Mechanics.
Swansea, Jan. 9 l.
2. Shmuely. M. and Peretz, D. (1976). Int. J. So/ids Structures. 12, 6779.
3. Kanninen, M. F. (1973). Int. J. Fracture. 9(1), 8392.
4. Kanninen. M. F. (1974). Int. J. Fracture. 10(3). 41530.
5. Kobayashi, A . S. and Mall. S. (1976). 'Dynamic photoelastic analysis of three
fracture specimens'. Technical Report No. 24. Department of Mechanical
Engineering, University of Washington.
6. Keegstra. P. N. R.. Head, J. L. and Turner. C. E. (1977). 'A transient finite
element analysis of unstable crack propagation in some twodimensional
geometries'. Proc. 4th Int. Conf. Fracture. Waterloo.
7. Dvorak, G. J. (1971). 'A model of brittle fracture propagation'. Eng. Fract.
Mech.. 3.
8. Malluck. J. F. and King, W. W. (1978). 'Fast fracture simulated by a finite
element analysis which accounts for cracktip energy dissipation". Proc. 1st Int.
ConJ. Num. Meth. Fracture Mechanics. Swansea Jan. 913.
9. Rydholm. G.. Fredriksson. B. and Nilsson. F. (1978). 'Numerical in
vestigations in rapid crack propagation'. Proc. 1st Int. ConJ'. Num. Meth.
Fracture Mechanics. Swansea, Jan. 913.
10. Fearnehough. G. D. (1974). 'Fracture propagation control in gas pipelines'.
Paper 1, Crack Propagation in Pipelines. International Symposium.
NewcastleuponTyne. March.
11. Maxey, W. .. Keifner. J. F., Eiber. R. J. and Duffy, A . R. (1973).
'Experimental investigation of ductile fractures in piping'. International Gas
Union Conference, Nice.
12. Fearnehough, G . D . . Jude. D.W. and Weiner. R. T. ( 1971). 'The arrest of brittle
fractures in pipelines'. Proc. Conf. Practical A ppi. Fracture Mechanics to
Pressure Vessel Technology. C33. Institution of Mechanical Engineers.
London, May.
13. Shannon. R. W. E. and Wells. A . A . (1974). 'A study of ductile crack
14.
15.
16.
17.
18.
19.
20.
21.
22.
23.
24.
25.
26.
27.
28.
29.
189
propagation in gas pressurized pipelines'. Paper 17, Crack Propagation in

Pipelines. International Symposium. Newcastle upon Tyne, March.
Alder. G. M. (1973).'The pressure loading on the walls of a propagating shear
fracture in a high pressure gas main'. Research report. Department of
Mechanical Engineering. University of Edinburgh.
Alder, G. M. (1972). 'The flow through a propagating shear fracture in a high
pressure gas main'. Research report, Department of Mechanical Engineering,
University of Edinburgh. August.
Hahn.G. T., Sarrate, M., Kanninen, M. F. and Rosenfield, A. R. (1973). Int. J.
Fracture. 9(2). 209-22.
Poynton. W. .. Shannon, R. W. E. and Fearnehough. G. D. (1974). Trans.
ASME. J. Eng. Mat. Techno!.. Paper No. 74. Mat. 10. June.
Poynton. W. A. (1974).'A theoretical analysis of shear fracture propagation in
backfilled gas pipelines'. Paper 14. Crack Propagation in Pipelines,
International Symposium. Newcastle upon Tyne. March.
Poynton. W. A. and Fearnehough. G. D. (1972).'An analysis of shear fracture
propagation in gas pipelines", in Proceedings oj an International Conference on
Dynamic Crack Propagation, Lehigh University, July (Ed. G. C. Sih).
pp. 183-96.
Kanninen. M. F.. Sampath. S. G. and Popelar, C. (1975). 'Steady state crack
propagation in pressurised pipelines without back-fill'. Research report.
Battelle Columbus Labs.. Ohio. January.
Ayres. D. J. (1975).'Elastic plastic dynamic analysis of opening area of an axial
crack in a pipe under pressure'. Paper L7/6, 3rd Int. Conf. Struct. Mech.
Reactor Techno/.. London. September.
Shantaram. D.. Owen. D. R. J. and Zienkiewicz. O. C. (1976). 'Dynamic
transient behaviour of two and three dimensional structures including
plasticity, large deformation effects and fluid interaction', //;/. J. Earthquake
Eng. Struct. Dynam.. 4, 561-78.
Zienkiewicz, O . C . and Nayak. G. C. (1971). 'A general approach to problems
of large deformation and plasticity using isoparametric elements'. Proc. 3rd
ConJ. Matrix Meth. Struct. Mech.. Wright-Patterson AFB. Ohio.
Nayak, G. C. and Zienkiewicz, O. C. (1972). 'Elasto-plastic stress analysis. A
generalisation for various constitutive relations including strain softening'. Int.
J. Num. Meth. Eng.. 5, 113-35.
Clough. R. W. ( 1973). 'Numerical integration of the equations of motion', in
N A TO Lectures, pp. 525-33, Alabama Press.
Tong. P.. Pian. T. H. H. and Bucciarelli. L. L. (1971). 'Mode shapes and
frequencies by finite element method using consistent and lumped masses".
Computers and Structures, 1, 623-38.
Rock, T. A. (1978). 'Dynamic analysis of civil engineering structures'. Ph.D.
Thesis, University of Wales, Swansea.
Hinton. E.. Rock. T. A. and Zienkiewicz. O. C. ( 1974). 'A note on mass lumping
and related processes in the finite element method'. Internal Report No.
C/R/225/74. University of Wales. Swansea.
Irons. B. M. (1976). 'The semiloof shell element', in Finite Elements f or Thin
Shells and Curved Members (Ed. R. H. Gallagher and D. G. Ashwell). Wiley,
London.
190 D . R. J. OWEN
30. Nagarajan, S. and Popov, . P. (1975). 'Nonlinear dynamic analysis of
axisymmetric shells". I nt. J. Num. Meth. Eng., 9, 535-50.
31. Perzyna, P. (1966). 'Fundamental problems in visco-plasticity'. Recent
Advances in Applied Mechanics, Vol. 9, Academic Press, New York.
32. Crisfield. M. A. (1974).'On an approximate yield criterion forthin steel shells'.
Department of the Environment. TRRL Report 658 Crowthorne.
33. Robinson, M. (1971). 'A comparison of yield surfaces for thin shells'. I nt. J.
Mech. Sci., 13, 345-54.
34. Leech, J. W. (1965). 'Stability of finite difference equations for the transient
response of a flat plate', AI AA Jl, 3(9), 1771-3.
35. Cormeau. I. C. (1975). 'Numerical stability in quasi-static elasto/viscoplasticity", I nt. J. Num. Meth. Eng., 9, 109-27.
7
Survey of Computational Methods for Transient
Fluid-Structure Problems in Reactor Safety
P. FASOLI-STELLA and A. V. JONES

lspra, Italy
1.
Centre,
INTRODUCTION
Transient fluid-structure problems in reactor safety include a vast range of

phenomena associated with possible or hypothetical malfunctions of
nuclear installations. Among these are processes which have attracted a
great deal of theoretical and experimental attention, such as:
(1)
(2)
(3)
(4)
(5)
(6)
(7)
Rapid loss of pressurised coolant from gas-cooled or water-cooled

reactors (so-called blowdown problems).
Violent thermal interactions between hot material and cold liquids
and the consequences for safety-related structures (fuelcoolant
interactions).
Fires, depressurisations and gas explosions as they affect reactor
secondary containments.
Sabotage problems.
Thermal consequences of the development of blockages in the
coolant channels of a fast reactor subassembly.
Immediate hydrodynamic and thermal effects of transient
neutronic overpower of a reactor core (whole-core accident
problems).
Containment of a fast reactor following an HCDAa 'hypothetical core disruptive accident".
This chapter will concentrate upon the last of these problems, but since
all of them involve highly transient situations with strong interactions
between fluids and structures, much of what is said regarding computational techniques is applicable mutatis mutandis to the others.
191
FIG. 1. Loop-type LMFBR. A. Large rotating shield; B. Intermediate rotating

shield; C. Small rotating shield; D. Shield drive motors: E. Control rod drives: F,
Fuel transfer channel; G, Pressure surge shielding tank; H. Basalt shield boxes; J.
Thermal shield; K. Dip plate; L, Sodium level; M. Instrumentation support
structure; N,Core instrumentation plate; O. Annular storage; P, Core; Q. Blanket;
R. Guard vessel; S, Core support plate; T. Reactor vessel; U, Sodium outlet; V.
Sodium inlet; W. Emergency heat exchanger.
TRANSIENT FLUID-STRUCTURE PROBLEMS IN REACTOR SAFETY
193
1.1. Containment of an HC DA in a Fast Reactor

Figure 1 is a schematic diagram of a liquid metal cooled fast breeder
reactor (abbreviated here to fast reactor), this particular example being of
the loop type. Controlled nuclear fission within the core produces
considerable heating of the fuel, and the fuel is prevented from melting only
by a continuous pumped flow of the coolant (usually sodium) which leaves
the core with a temperature of 500-600C. The coolant, which, together
with the core and internal piping, is enclosed in the primary containment,
passes through heat exchangers mounted either externally, as in the figure,
or internally, as in pool designs. Within the exchangers its heat is given up to
secondary coolant circuits and eventually employed to produce steam
which drives conventional turbogenerator sets. A schematic diagram of a
pool-type LMFBR appears in Fig. 2.
We now consider one possible type of accident in such a system, the
HCDA.
Although no purely mechanistic pathway leading to HCDAs has been
found, and although instrumentation and engineered safety features of
current designs have been introduced with the intention of ruling such
accidents out completely, the potential consequences are serious enough for
it to be considered worth while to analyse them in detail. The chain of
causation is generally agreed to be as follows:
(1)
(2)
(3)
(4)
Because of transient neutronic overpower or an impediment to

coolant flow, there is a temporary failure to cool the core
adequately.
Fuel melts and is relocated by gravity and hydrodynamic forces.
The result can be a neutronically more reactive configuration,
leading to further fuel melting and relocation, increased cooling
difficulties, and so on. until an uncontrollable situation exists in the
core. This phase of the accident is the proper concern of whole core
accident analysts.
Increasing heating results in fuel and coolant vaporisation, rupture
of the cladding containing the fuel and (in some as yet not clearly
defined manner) the creation of a bubble of vaporised fuel and
coolant, the so-called core bubble.
This core bubble, which has a volume of the order of 1 m 3 and an
internal pressure variously estimated as lying between 50 and 500
atmospheres, then expands against the restraints of the surrounding coolant and structures, on which it exerts the so-called
containment loading.
194
P. FASOLI-STELLA AND . V. JONES

COVER
SUBASSEMBLY
CONTROL
GRIPPER
LIFTING
MECHANISM
- S T O R A G E RACK
MECHANISM
MECHANISM
SUBASSEMBLY
MECHANISM
MECHANISMS
TRANSFER
O P E R A T I N G FLOOR
.ROTATING PLUGS
SHIELD COOL
ING DUCT
P R I M A R Y TANK
COVER
MECHANICAL
SUBASSEMBLY
S T O R A G E RACK
PUMP.
HEAT E X C H A N G E R -
PRIMARY
TANK-
REACTOR
VESSEL-
NUCLEAR INSTRU
M E N T THIMBLE
-BLAST
SHIELD
-AIR BAFFLE
FIG. 2.
Pool-type LMFBR.
The object of analyses of containment loading and response, then, is to

discover the class of HCDAs which a given fast reactor containment
(usually designed more for ordinary operation than for accident situations)
is capable of withstanding without developing breaches sufficient to allow
the release of significant quantities of fuel vapour and radiologically
hazardous fission products to the secondary containment and thence
(under the most pessimistic assumptions) to the external environment and
the general public.
195
The major burden of the containment loading is borne by the primary

tank and by the head cover, which must remain intact for escape of the
fission products to be prevented. This load is, however, transmitted both
through the coolant and through the various internal structures, and its
correct evaluation requires an adequate modelling of all the transmission
processes and of their temporal evolution. An additional point is that
within the primary vessels are provided auxiliary heat removal systems
whose function is to keep the reactor cool in the event of a failure of the
primary cooling system, and whose continued operability after the accident
is essential for the long-term safety of the reactor. Thus, loadings on such
auxiliary heat removal systems must also be evaluated as part of the
containment problem.
1.2. The Sequence of Events in LMFBR Containment Loading due to an
HCDA
The following account is intended to describe the evolution of a 'typical'
core bubble expansion, but is based on the results of model tests using lowpressure explosives. If real HCDA peak pressures should turn out to be a
few tens of atmospheres only, the description could be misleading, but for
the present we must continue to be 'conservative' and think in terms of quite
violent HCDAs which produce core bubbles with high initial pressures
which are well sustained during the early phase of the bubble expansion,
and in this case the model tests are thought to be representative.
When the core bubble is formed, shock waves propagate through and
interact strongly with the surrounding structures, and give rise to a fluid
flow: this passes upwards through the axial breeder and the above-core
structures associated with the con t rol system, the instrumentation tree, etc.;
sideways through the radial breeders, shield rods and core barrel; and
downwards through a possible axial breeder, the diagrid and the coolant
inlet plenum. Because of differences in path lengths, pressurisation
resulting from this flow tends to become established first below the diagrid
(which was previously thrust downward by the core pressure and is now
subject to an upward thrust), and then at the upper end of the core barrel;
finally, the whole region outside the core barrel becomes pressurised, with
the region of peak pressure (arising through convergence ol"pressure waves)
occurring roughly at the level of the core. The overall loading history for the
outer tank is first a downward thrust on the curved bottom and then a radial
thrust distributed along the cylindrical vertical portion. It should be noted
that this loading varies both in space and in time, as pressure waves
propagate, reflect, meet and interact. For all but the mildest HCDAs the
196
loads are sufficient to cause plastic yielding (outer tank strains ~ 5 %). For
the core barrel the loading is quite different, since this structure is first
pressurised internally, which causes plastic deformation, and then
externally, which possibly gives rise to buckling.
For most reactor designs and for moderate HCDAs there then follows an
interlude when the core bubble pressure has dropped considerably,
pressure waves have been reduced and the general level of fluid pressures is
insufficient to cause further plastic deformation of the vessel. The core
expansion work has been partitioned roughly equally into structural
deformation and kinetic energy of the fluid, and the coolant is moving
vertically both inside and outside the core barrel.
After the coolant free surface has risen a distance equal to approximately
15% of the total height of the reactor vessel (by which time the core bubble
has expanded to touch the core barreland almost fills the core barrel itself).
it impacts the vessel roof, thrusting the cover gas into a narrow annulus
where roof and vessel meet, and subjecting both to a loading with a short
rise time. The vessel deforms radially, mostly at its upper end, and pressure
waves propagate both downwards into the coolant from the roof region and
upwards from the bottom of the tank as the upper portion of the tank
contracts longitudinally during its radial expansion.
When these secondary pressure waves have become so reduced as to
cause no further deformation, the containment loading phase of the
accident is over and the deformed reactor, which is still pressurised by the
hot core bubble and the containment capability of which has possibly been
compromised, is handed on to the next phases of analysis, that of the
assessment of heat removal capabilities and of radiological consequences.
1.3. Characteristics of the Fluid-Structure Interaction Problem
It is convenient to consider these under the three headings of fluid flow
characteristics, structures, and features of the fluid-structure interaction.
Fluid Flow Characteristics
The coolant is a single-phase liquid, which interacts with another singlephase fluid, the cover gas, via well-defined sharp interfaces and with the core
bubble via an interface at which phase changes and mass transfer can take
place. The core bubble itself contains several components and material in
all three phases in general. Although considerable theoretical effort is
currently being expended to understand the evolution of the core bubble,
including its thermal and mechanical interaction with the surrounding
coolant.' in containment studies it is customary to assume no mass transfer
197
across the core bubble and uniform pressure within it. It is also customary
to assume a uniform pressure within the cover gas space (significant
pressures exist there only when the cover gas has been compressed into a
small annular volume), and under these assumptions the fluid flow is purely
single-phase with time-varying boundaries and time-dependent pressure
boundary conditions.
Because of the large scale of the system, surface tension and viscosity
effects are negligible, and significant turbulence is confined to regions small
enough for its effects to be modelled adequately by friction coefficientsfor
example, for flow through perforated structures. It appears from recent
work 2 that at least for dip plates pressure drop formulae suitable for steady
flow can also be applied to transient situations with the time scales existing
in containment loading problems.
For a significant proportion of the evolution of the accident the
compressibility of the fluid is important, as witness the account of the
loading process in the preceding section in terms of wave propagation. The
fluid also undergoes displacements of the same order as the largest length
scale in the system, culminating in roof impact.
Structures
The structural elements of the system arealmost all of the same material,
austenitic stainless steel. There is weld material at the seams of the outer
vessel and elsewhere, and irradiation-embrittled steel in the neighbourhood
of the core.
Deformations are, in general, elastic-plastic with strain-hardening and
significant strain rate effects. The principal structures are roughly classified
below:
(1)
(2)
(3)
(4)
(5)
Axisymmetric thin shells. These include the outer tank and core
barrel. Junctions between thin shells are found where the core
support skirt joins the outer tank in some designs.
Strong structures through which no flow takes place. This class has
as chief' representative the roof and its hold-down system.
Strong structures through which there is flow. This class includes
the diagrid and the dip plate, and the heat exchangers and pumps in
pool designs.
Weak structures through which no flow takes place. Examples are
heat shields and insulation.
Weak structures through which there is flow. This includes most of
the remaining structures, such as breeder zones, shield rod arrays
and the above-core structures.
198
P. FASOLI-STELLA AND A. V. JONES
Features oj the Fluid-Structure

Interaction
From the above account of the structures involved and from Fig. 1 it may
be stated that in intact loop designs the gross geometry and flow are
axisymmetric, apart from the effects of piping connections, while other
non-axisymmetric features (off-axis core bubbles, above-core structure, dip
plates) either have little effect on the overall dynamics or have a
considerable degree of radial symmetry. In pool designs the situation is
much less clear-cut, and some details, such as loadings on pumps and heat
exchangers, definitely require a full three-dimensional treatment. In both
designs, if a vessel should split o r a rotating shield in the roof lift, a new and
very non-axisymmetric phase of" the flow begins, one which clearly requires
three-dimensional analysis.
In the case we are considering, fluid and structural motions are strongly
coupled, both locally and globally. As an example, loading of a vessel at one
point produces deformation and the emission of pressure or tension waves
into the liquid at that point, while wave propagation in both fluid and vessel
can induce deformation and pressure loading elsewhere. Vessel deformation at roof impact produces a pressurisation of the liquid at the bottom
of the vessel.
We next remark that the system contains a wide variety of length scales,
ranging from the thickness of' the vessels and the hydraulic diameter of the
coolant channels in the breeder zones to the length of the reactor vessel
itself.
It also displays a broad spectrum of time scales, from that for wave
propagation in the thickness of the outer vessel to that from core bubble
initiation to roof impact. Very influential on the overall dynamics are those
situations in which liquid flows through or around deforming or weakly
restrained structures (shield rod arrays, breeder zones, dip plates, deflector
plates, insulation and vessels).
Finally, a key feature of the fluid-structure interaction is roof impact, the
phenomenon of mass fluid motion suddenly beginning to interact with a
massive deformable structure.
1.4. Popular Simplifications of the Problem
Although considerable progress has been made in the numerical analysis
of three-dimensional structural problems. 3 - 4 the analysis of threedimensional compressible fluid flow has only recently come into its own."'
while current three-dimensional coupled fluid-structure capability is
limited to a few very specialised geometries and flows. A fully threedimensional problem, such as that of containment, with its wave
199
propagation effects and large fluid displacements, is not yet manageable

because of restrictions of computer speed and storage and of logical
complexity of code organisation. For the purpose of calculation, thus, it is
necessary to simplify the problem somewhat, with engineering judgement
and knowledge of what is computationally practicable as guides. The
simplification of axisymmetry is most appealing, for the reasons set out in
the previous section. A discussion of the progress which has been made
towards analysing non-axisymmetric geometries appears in the final section
of this chapter. Inclusion of the inner and outer vessels, of the roof, of the
core support skirts (if any) and of the diagrid is essential for satisfactory
modelling, and these structures are well described by an axisymmetric
approximation. As a further approximation, the roof can be treated as a
rigid disc held in position by some sort of elastic-plastic restraint, the
response of the vessels to roof impact being calculated directly, and the
response of the real roof with its numerous inclusions and penetrations
under the calculated roof loading history being analysed separately, usually
by finite element techniques. This procedure is conservative (i.e. it
overestimates both the vessel deformations and the loading on the roof),
but as some fully coupled calculations 6 have shown, not grossly so. The
diagrid can be treated as a deforming thick disc with a resistance to flow, while
breeder zones and shield rod arrays may be considered as inertial masses
which permit some relative fluid motion. The assumption of some sort of
global 'equation of state' for such stacked structures is necessary in any case
because of the impossibility of combining calculation of the overall
containment problem with that of the flow between, deformation, and
bodily motion of" the arrays, which is a fully three-dimensional compressible
problem in highly variable geometry and with a characteristic length scale
much smaller than that of the rest of the containment problem.
Above-core structures and dip and deflector plates may be modelled as
distributed flow resistances, while it is customary to ignore heat shields and
insulation completely to a first approximation, since their energy absorbing
capability is generally small.
Thus, we end up with a simplified containment problem, the geometry of
which consists of a core bubble surrounded by inertially constraining
structures, bounded below by the diagrid, a deformable disc with flow
resistance, laterally by a thin shell (the core barrel) and above by the flow
resistances of above-core structure and a dip-plate (Fig. 3). The whole
ensemble is enclosed by the shell of the outer tank and by the rigid roof.
Although such a geometry is very much simplified, it is still too complex
for design studies, parameter surveys and, indeed, any study involving more
200
P. FA SOLISTELLA A ND A . V. JONES
AXIS OF ROTATIONAL
SYMMETRY
HOLD DOWN SYSTEM
COOLANT
ABOVE-CORE
STRUCTURE
(flow resistance) I
.OUTER VESSEL
(shell)
CORE BARREL (shell j

BREEDER ZONES AND
SHIELD RODS
(equation of state)
SUPPORT SKIRT (shell)
FIG. 3. Simplified containment problem.

than one or two heroic calculations. There are good reasons to simplify still
further by ignoring the inertial constraints near the core, treating the
diagrid as rigid, and omitting or drastically simplifying the flow resistances
above the core. Such simplifications not only aid understanding, but also
allow the validation of calculational techniques and of the codes in which
they are embedded. Since the predictions of containment codes play an
important role in the licensing procedure for commercial LMFBRs. such
validation is taken seriously by the reactor safety community. In one such
validation programme, COVA , 7 60 model experiments are being
performed, beginning with extremely simplified geometries and pro
gressively adding more and more features of real reactors, each case being
calculated by the codes to be validated. In this way the suitability and range
of validity of individual approximations, either physical or numerical, can
be assessed.
T R A N S I E N T F L U I D S T R U C T U R L PROBLEMS IN RF.A CTOR SA FETY
201
1.5. Structure of the Remainder of the Chapter

Section 1 presented the problem of fast reactor containment loading,
outlined the sequence of events and the important physical phenomena in
the containment loading process, and described the simplifications
commonly applied to make its computation practicable. Section 2 contains
the physical and computational fluid dynamics necessary to an appre
ciation of Section 3. which reviews the fluidstructure coupling techniques
commonly applied in containment codes. Section 4 reviews more recent
developments in the computation of containment loading, discusses some
numerical results and details those aspects of the problem which still await
solution.
2.
PHYSICA L A ND C O M P U T A T I O N A L FLUID DYNA MICS
This section presents those aspects of fluid dynamics most relevant to the
containment problem. It begins by discussing the hydrodynamic equations,
their Lagrangian and Eulerian formulations and the effect of compres
sibility on their mathematical character; carries on to survey the numerical
methods available for their solution ; and concludes by reviewing the special
problems posed by physical discontinuities in fluid flow.
2.1. Hydrodynamic Equations
The treatment of fluid flow involves the solution of the equations of
conservation of mass, momentum and energy.
A Newtonian viscous flow is, in general, described by the NavierStokes
equations, which in the absence of external forces and in Cartesian
coordinates, are: 8
pUj
(pup,
Cl
CX:
c pu,
Ct
CX:
+ ^~
(conservation of mass)
CU:
CU:
OX:
CX
+
CX:
OX;
(conservation of momentum)
II
p\e
jiPi)
OX,
+ OX;
(1)
(2)
pu, e +
+ j
CX:
CT
cx,
CU:
+ CX
CU,
ou.
cx,.
dx,
cx.
(conservation of energy)
(3)
p=f(p,e)
+ oj;
(equation of state)
(4)
202
P. FA SOLISTELLA A ND . V. JONES
where is the specific mass, u, is the components of'the velocity vector 0,p is
the fluid pressure, e is the internal specific energy, Fis the temperature,
and are the viscosity coefficients, and represents the coefficient of heat
conduction. The tensor notation on the i,J,k indices has been used to
indicate the summation.
In the applications which we are considering, the viscous terms and heat
conduction may be regarded as negligible compared with the other terms in
eqns. (1) and (3), which leads to the simplified Euler equations:
^ 4 ^ = 0
ct
cx,
CpU:
C
Cp
+ (pu,Uj)= ~
(5)
(6)
</
u
c /
u \
c
P\e + T) + Pl' I e +
et \
2
cx,
P=f(e,p)
~ipu)
(7)
C'.Y,
(8)
The NavierStokes and Euler equations are nonlinear partial

differential systems. The nonlinearity is due to the advection terms
'CpU,
cpu,uk
ex,
CX:
etc
to the nonlinear behaviour of the materials and to the presence of viscosity

and heat conduction, etc.
The systems of equations (5)(7). written in conservative form, may be
expressed compactly in the vector equation
fw
+ VF = 0
(9)
et
where the /; components of w are the state variables [p,pup(e + (u2/2))] and
F (a nonlinear differentiable function of w) isa vector of fluxes, of mass, of
momentum and of energy. The conservative form allows an easier
understanding of the features of'the system and plays an important role in
many numerical approaches.
In many cases system (9) is reducible to a quasilinear system of the first
order, as in many specific casesthat is, we can write:
~ + /l(w)V w = 0
ct
where A (w) = CF,Cw, is an matrix.
(10)
TRANSIENT FLUIDSTRUCTURE PROBLEMS IN REA CTOR SA FETY
203
For such quasilinear systems a fairly complete mathematical theory has

been developed. 9
The vector equation (10) is called hyperbolic if A has n real and distinct
eigenvalues. When couples of eigenvalues are real but coincident, the
system is termed parabolic, whereas in the case of complex eigenvalues it is
termed elliptic. Depending on the features of" the eigenvalues of matrix A ,
system (10) may have even a hybrid character (hyperbolicparabolic,
elliptichyperbolic, etc.).
To introduce the concept of characteristic curves, we consider the specific
case of onedimensional isentropic flow:
L , lu. p] = p, + upx + pux = 0
j
2
L2[u. p] = p(u, + uuj + c px = Oj
where c is the sound speed and c2 = Cp/Cp. p,. px, u,. ux indicate the local
derivatives of and u.
In order to recast this system in the form of eqn. (10), we have to define a
matrix A (u, p):
A =
It
f>
c2/p
It
(12)
whose eigenvalues are = it + c. Because the two eigenvalues of matrix A

are real, the system is hyperbolic. For each of these values of there exists a
nontrivial solution /J,/.2 of the system:
(u )/2 + p/J = 0
, , .1 _ r
c22/
/?.322 +, ,..
(u .)/.
=0
(13)
when = u + c,'/} 1 = c/p and when = u c, 1 / / . 2 = c/p.

This means that it is possible to define two characteristic directions given
by:
= (u + c )
,
<14>
= ( c)'/j
By differentiating u and p along these directions the equations in the so
called characteristic form are found:
U. + C/P
, c/p
= 0)
pp = 0J
Px
204
P. FASOLI-STELLA AND A V. JONES
These equations express the fact that the characteristic curves in the .v, /
plane represent motions of possible disturbances whose velocities are
dx
= u+c
>
The characteristic curves play a fundamental role in defining the

properties of one-dimensional flow, such as uniqueness of the solution,
propagation of discontinuities and separation between regions of different
type. They can also be used directly to calculate the propagation of
disturbances in one-dimensional compressible systems. 10
In two-dimensional problems the characteristics are surfaces, and the
theory becomes very complicated, although it is possible to define the bicharacteristic curves.
The time-dependent Euler equations for compressible fluids are then of
hyperbolic type, whereas it has to be noted that the Euler equations for
incompressible cases exhibit a hybrid elliptic-parabolic character, which
gives rise to quite different numerical problems.
In general, because of the non-linearity of F, non-continuous solutions of
system (9) will exist for all time, even for perfectly reasonable initial and
boundary conditions. Experience indicates the occurrence of shock waves
in such cases and the corresponding solutions of the Navier-Stokes
equations for vanishing viscosity and heat conduction exhibit regions of
very high gradients, tending to local discontinuity. Mathematically
satisfactory descriptions of such discontinuous flows are obtained by means
of the concept of generalised solutions introduced by L a x . 1 1 1 2
The vector function w is called a generalised solution of eqn. (9) if it
satisfies the integral form of this set of conservation laws:
d
di
u, d V +
F,n,dS = 0
(17)
over some test volume V with surface S.

In this case the function w may be discontinuous and is then termed a
weak solution or a solution in the sense of distributions of the conservation
laws (9). To assure the uniqueness of the weak solution, w must satisfy the
so-called entropy condition.
In the case of react or containment studies the fluid treatment involves the
205
solution of a so-called initial value problem represented by the Euler

equations in axisymmetric coordinates. The state of the fluid at the initial
time t = 0 is known in a region R = R(r, z,0) of the plane r,z, and
boundary conditions (e.g. the action of walls) are known at every time t. We
have to calculate the state of the fluid and the region R(r, z, t) occupied by
the fluid for subsequent times 0 < t < T. We deal in general with several
fluids in which case R = UR, (R, is the region of the ;th fluid) (Fig. 4). If
D,= D,(r,z,t) is the surface bounding region R then the surfaces D,
Ro
R
l_ i.
_L.
"2
Vi
\
L_
t.T
t= 0
FIG. 4. Regions of fluid.

represent the positions of the material interfaces of the fluid (contact
discontinuities). Thus we have the tasks of determining the surfaces of
discontinuity. D and of the subsequent formation or evolution of shock
waves.
2.2. Lagrangian and Eulerian Formulations
In solving a hydrodynamic problem the first question is which type of
coordinates to adopt. It is possible to choose a Lagrangian approach, in
which the reference frame moves with the elements of the fluid; an Eulerian
approach, in which the reference frame is fixed in the space; or a 'mixed'
reference frame which moves in an arbitrary fashion with respect to the
fluid.
Depending on the approach, eqns. (5)-(7) have different forms. If we
express the equations in terms of the time derivatives with respect to the
moving fluidthat is, of the Lagrangian derivative
cl
d7
cl
+ -V
(18)
206
the Euler equations assume a particularly simple form. By differentiating

the divergence terms in eqns. (5)-(7) we obtain the Lagrangian form:
dp
di
= -pV-fl
p-=-V,
de
p-r=
dt
-/>V-0
(19)
(20)
(21)
To solve the problem numerically, the region R (in our case the plane /, z) is
discretised into a rectangular mesh of cells with sides parallel to the
coordinate axes or into a set of arbitrarily shaped elements. All variables of
the fluid are calculated for subsequent times at a discrete set of mesh points.
In the Lagrangian form the mesh points follow the particle paths of the
fluid elements. The independent space variables refer to a coordinate
system fixed in the fluid (e.g. the masses or initial positions of the fluid
elements) and the mesh points (or nodes) move in a way determined by the
approximate solution of hydrodynamic equations. The Lagrangian point
of view has frequently been preferred, especially in one-dimensional
problems, because: (1) this form of equation is simpler; (2) the positions of
boundaries and of material interfaces are automatically determined in the
calculation, since we follow each bit of fluid: (3) during the calculation the
mesh following the fluid flow becomes finer in the region of greater density.
Difficulties with this approach arise in multi-dimensional problems
when, because of sliding or shear of different parts of fluid, the mesh
becomes highly distorted (Fig. 5). In this case the calculation becomes very
inaccurate and often must be terminated prematurely.
In the Eulerian formulation the independent space variables refer to a
coordinate system fixed in space, so that the fluid streams through the cells.
12
11
t.
10
Fio. 5. Lagrangian mesh evolution.
207
In principle every hydrodynamic problem can be treated in Eulerian

coordinates, even in the case where the region R is multiply connected or
turbulent flow appears. The major difficulty lies in properly determining the
paths of the surfaces approximating the boundaries of the region R and the
material interfaces. These surfaces must move with the fluid, and,
consequently, they create irregular time-dependent boundary zones in the
fixed Eulerian mesh in which it is often difficult to define the fluid variables
accurately or to decide on suitable approximations to the continuity
equations. In addition, the advective terms appearing in the Eulerian
formulation (spatial derivatives of vector B) need a careful numerical
treatment to avoid instabilities and severe errors in the results. 13
There have been several attempts to combine the Eulerian and
Lagrangian
approaches to obtain
the advantages of both
formulations. 10 ' 35
2.3. Compressibility
The hydrodynamic equations are applicable to a very broad range of
fluids. In the safety calculations which we are considering the coolant must
be treated as compressible during the wave propagation phase, but it can be
considered as quasi-incompressible in the second stage of the flow before
roof impact.
In the compressible case the situation is described by the Euler equations
and the specific energy of the fluid consists of two components:
E = e + \u2
(22)
14
namely the specific internal energy and the specific kinetic energy. In
compressible fluids an interchange can occur between the internal energy
and the kinetic energy, so that in some region of the fluid most of the local
energy can be internal energy, whereas elsewhere it is the energy of flow
which is dominant. On the other hand, in many cases of interest the energy
of flow is small: the flow can then not compress the fluid significantly, no
appreciable interchange of energy takes place and we say that the fluid
behaves incompressibly.
We can consider the comparison between compressible and incompressible fluids from a different point of view. In the absence of viscosity and
heat conduction the hydrodynamic equations are hyperbolic and contain
two characteristic speeds, the particle's velocity, 0", and the sound speed, c.
We can distinguish between classes of problems, depending on the ratio (the
Mach number) of the flow speed to the sound speed
M = M/c
(23)
208
When M < 1, we have a subsonic flow; when M > 1, we have a supersonic

flow. If M is very small, so that the flow speed is very small compared with
the sound speed, the fluid can be considered as incompressible, so that
dp
~ =0
(24)
d/
Using the continuity of mass, this condition for incompressibility is
equivalent to;
V-B = 0
(25)
This assumption simplifies the form of the equations considerably but has
the effect of defining an infinite sound speed.
In the incompressible formulation we no longer need an energy equation,
because the other equations completely define the fluid velocity and the
pressure. Nevertheless, as already mentioned, the incompressible equations
have a mathematical behaviour different from that of the general equations
and require their own numerical techniques. For this reason a variety of
specialised numerical methods has been developed to calculate incompressible flows accurately and economically. 1 3 1 4
In recent years implicit techniques have been developed which are so
designed as to be able to treat both compressible and incompressible
phases. 15
In the following paragraphs we shall concentrate on the approximate
numerical solution of the compressible hydrodynamic equations and on
those implicit algorithms able to treat both compressible and incompressible flows.
2.4. Finite Difference Solution of the Compressible Hydrodynamic
Equations
General Remarks
The finite difference (FD) technique has been chosen by most workers for
hydrodynamic problems. The first reason for the popularity of FD was the
simplicity of this approach: the difference equations could be obtained
directly by approximating the differential operators in a local sense, using
Taylor series expansions at the nodes of the mesh.
Very many FD methods have been tried for the compressible
hydrodynamic equations, both because of the diversity of the physical
phenomena described and because of the non-linearity of the problems:
each problem or class of problems may require special techniques for
its detailed and efficient solution and for the imposition of boundary
209
conditions. These techniques include the use of Lagrangian, Eulerian or

hybrid meshes, conservative and nonconservative formulations, explicit
or implicit time integration and the manner in which the shock or contact
discontinuities are treated.
This variety in solution techniques is found also in numerical schemes for
containment calculations.
The set of" equations which has been defined in Section 2.1 constitutes an
initial value problem and may also be represented by the timedependent
equations: 1 4
dw
= Lw
(26)
d/
where w is the column vector w = (p, pu pe) and L is a spatial differential
operator. FD methods replace the spatial operator L by some difference
operator L* operating on w, where w is defined only on a discrete mesh.
Hence, the righthand side of (eqn. 26) is replaced by some time average
w"+ w = r.A tL*"+1w"+ ' + (1 i;)A tL*"w"
(27)
and 0 < ; < 1. In choosing an FD technique some important properties of

the approximation have to be ascertained and. in particular, its consistency,
convergence and stability.
Roughly speaking, consistency tells us about the error of approximation
of differential operators by a difference operator operating on the true
solution. We say a scheme is consistent when the difference between L* and
L, defined as the truncation error, goes to zero as the time step / and the
mesh size / are progressively reduced.
A scheme is convergent when the solution of the difference equation
approaches the true solution as mesh spacing goes to zero.
Stability is a very important property of the difference equations. A
numerical method is stable if a small error at any step produces a smaller
cumulative error at succeeding steps. If this property does not hold for a
numerical method, an error occurring at any stage grows without bound.
As an example, consider the case of a simple firstorder ordinary differential
equation and suppose that an error occurs at step At. We are interested
in the amplification, g, of this error at step (n + \) A t:
"+'=<;"
(28)
where g is termed the amplification factor and depends on the particular
210
difference scheme. We require for stability

|c"+1l<|c"l
(29)
\gQ"\ < Ic"l =>

|g|<I
(30)
(31)
For any difference scheme condition (31 ) must be satisfied if the scheme is to
be stable.
Unfortunately, the investigation of such properties of FD schemes is
possible only when the differential equations are linear except for some
special cases. Possible tools of analysis are linearisation and heuristic
procedures. 16
If in eqn. (27) is zero, the method is explicit, since iv" +1 is defined
explicitly by w". In this case the computation of w"+i is very simple for
compressible fluids, but the magnitude of the time step / is limited by the
Courant-Lewy-Friedrichs 1 7 stability criterion. If c is non-zero, the method
is implicit and system (27) leads to
5+iu+i =
-p u
(32)
which is a matrix equation. Only in special cases, such as in onedimensional schemes, can S be inverted. In general, multi-dimensional
problems S has a complex form and only relatively slow approximate
methods (Gauss-Seidel; successive overrelaxation and other iterative
techniques) are available to solve this system.
In compressible hydrodynamics FD methods are the most popular.
Lagrangian Methods
In Lagrangian coordinates the compressible hydrodynamic equations
assume a particularly simple form, which allows one to overcome the
difficult problem of the treatment of advective terms (19)(21). For this
reason, at least in one-dimensional problems, the Lagrangian formulation
may be regarded as the basic approach. In multi-dimensional problems,
however, shear flow in addition to compressible motion is to be described : a
Lagrangian mesh can be difficult to define and can quickly become
unacceptably distorted. Nevertheless, several quite successful attempts
have been made to extend the application of the Lagrangian formulation to
two-dimensional cases also.
In FD two-dimensional axisymmetric codes using quadrilateral meshes
(Fig. 6) the only essential difference between the various schemes lies in the
21 1
FIG. 6.
Lagrangian difference mesh.
approximation to the gradient of the pressure appearing in the momentum

equation. 18
The component s of this quality at point /,/'of the mesh may be calculated
as a function of the fluid variables at points 1,2, 3 and 4:
P_
Cr
Cp_
C
-=[tl
2,
Pi)(=2 4) (Pi P 4 X 1 J ]
-^-[(Pi
2,
P3)(r2 ''4) (Pi
(33)
'
- U)]
or as a function of the fluid variables at points 5, 6, 7 and 8:

1
Cr
Cp _
C
-2-2[^
- '
-*-
--)- - ( ~)(5
- -")]
(34)
-Pi)Oc-
' 8 ) - -(Pe-
-Pa)(r5
- ''s)]
where , and 2 are the areas of corresponding quadrilaterals. In these

schemes the thermodynamic variables (, and e) are defined at the centres
of the cells, the geometric variables (positions r and 2, velocities ur and u,,
accelerations ar and a.) at the nodes. Then, in expressions (33) and (34) the
pressures p,p2, ^Ps,Pb, , etc., are defined as arithmetic averages of
pressures in neighbouring cells.
These schemes are explicit in time because the gradient components, and
hence, the nodal accelerations ar and a., at time (n + \)A t are calculated
using the values of variables at time nA t. The acceleration components
allow the calculation of the velocity components:
'AFI
(35)
212
The new velocities give the nodal positions, which, in turn, give the new
volumes of the cells and their densities. At this stage it is possible to solve the
internal energy equation and to update the pressures by means of the
equation of state. The time loop is then complete and may be repeated foi
further time steps. The Courant condition imposed on the time steps tc
ensure stability of the Lagrangian mesh is:
/ < min (AlI c)
(36)
where A l is the smallest dimension of a cell, c is the sound speed in that eel'
and the minimum is taken over all cells.
As mentioned previously, a severe limitation of Lagrangian schemes is
mesh distortion, which often produces growing errors in the results after
certain time. Many attempts have been made to apply remapping 01
rezoning procedures with the aim of reestablishing a regular mesh during
the calculation. 19 We shall return to this topic later.
Eulerian Methods
The most general FD approaches for the solution of multidimensiona
problems in compressible fluid dynamics use Eulerian meshes. Owing to thc
difficulty in treating advective terms appearing in the Eulerian formulatior
(l)(3) and to the great variety of features displayed by fluid dynamic:
problems, many different algorithms have been proposed and applied. Th<
conservative form (eqn. 9), is generally preferred because the conservatior
laws of mass, momentum and energy are then automatically satisfied
Under particular conditions such as exist in high Mach number flow, the
use of a nonconservative equation for the internal energy can be
advantageous in limiting inaccuracies associated with large kinetic
energies. In several codes both formulations for the energy equation art
provided.
In Eulerian schemes pressures, densities and energies are always definec
at the centre of the cells, but the velocity components may be defined at the
cell corners, in the middle of the cell sides or at the centre of the cells.
The most significant feature of any Eulerian scheme is the approximatior
adopted for the advective terms. If the conservative vector form (9) is
adopted, the divergence operator on the righthand side is replaced by the
sum of fluxes through each cell surface: 14
2
F.A t/A V
(37;
213
where the subscript refers to the 2N adjacent points in space

dimensions.
In the firstorder Lax scheme we have:
2
+l
w"
\w[
> FÂ t/A V
j=l
(38)
i=l
13
Expanding in Taylor series, as in Hirfs stability analysis. Lax" s scheme

is seen to introduce an effective diffusion coefficient which can give rise to
inaccuracies in the results. In many cases second or higherorder accuracy
schemes must be used for the solution. Necessarily, in any secondorder
accurate method two storage locations are required for each space point on
the mesh. These may be employed to store two adjacent time levels in the
calculation (twostep procedures) and the time derivative may be time
centred. In the 'leapfrog" scheme (Fig. 7)
1
step (1)
iv" ' = "
F"Aa t/A V
r=J
IN
step (2)
>
,,_y^ +Fl,A
u " ++ 2 =
= uu " )
+[
(39)
i/
A V
1=1
In step(l) auxiliary variables are determined at auxiliary points (crosses)

which are then used to define fluxes F"+ ' at the main time step. The first step
is to be considered as a provisional step. This class of twostep methods has
attained considerable importance in numerical studies and it is the most
popular for solving compressible fluids (LaxWendrolf, RubinBurnstein,
etc. 13 ).
Another class of wellknown methods includes the particleincell
(PIC) 20 and fluidincell (FLIC) 2 1 methods.
The genesis of PIC is different from that of most other methods, because
an attempt is made to model the physical process through a finiteparticle
approach. It utilises Lagrangian fluid particles to transport mass,
momentum and energy through an Eulerian mesh of cells. This technique,
which combines Lagrangian and Eulerian advantages, makes great
demands on computer memory and calculation time.
A continuum method which evolved from the PIC code is FLIC, a two
step scheme. In the first step provisional values wr and it. are calculated
using only the contribution of the pressure gradients and of the external
214
. FASOLI-STELLA A N D A. V . JONES
FN
FW
FS
1>
FE
ii
EVEN TIME (main

mesh)
X ODD TIME (auxiliary mesh)
FIG. 7.
Leap-frog mesh.
forces, if present. The provisional energy (nonconservative formulation)

or (conservative formulation) is calculated only from the pressure term of
the equation plus external forces. In the second step only the contributions
of advection terms are calculated. The mass, momentum and energy fluxes
across each cell interface are calculated by donor cell differencing (upwind
method ' 3 ) based on provisional values of velocities r , Ctz and energy . or .
This final advective contribution must be added to the provisional (tilde)
values.
FLICtype methods are the basis of several FD codes in Eulerian or
mixed coordinates, treating reactor containment problems, as OIL. 2 2
TOIL. 2 3 H E L P 2 4 and PISCES2DELK. 25 as well as the FE code
EURDYN/1M. 2 "
For all these Eulerian explicit schemes the stability Courant
FriedrichsLewy 17 condition must be satisfied:
/ < min [A l/(\u\ + c)]
(40)
where A l is the smallest dimension of a cell and \u\ + c is the sum of the
sound speed and the flow speed in this cell.
215
We conclude our review of Eulerian FD techniques by outlining a

completely different method devised by Harlow and Amsden 1 5 to treat all
flow speedsthat is, flows in which the Mach number may in principle vary
from zero to infinity. A s we mentioned before, this feature is particularly
interesting for the safety problems with which we are concerned.
This method, named ICE (Implicit Continuousfluid Eulerian), reduces
essentially to the MA C method 2 7 for incompressible flow as the sound
speed becomes infinite and to an implicit variant of the usual Eulerian
techniques as the sound speed becomes small.
Starting from the equations in conservative form, in cylindrical co
ordinates with axial symmetry (to obtain plane coordinates we put r = 1),
the numerical finite difference equations for the mesh of Fig. 8 are
P
^~TLl
"" ;>/,,> 2, (PO? + + ( l .,2,/ i + ( .,2,]
rA
[(PUr)"o
2 ) . / ( l 2) (Pr)"+(1 2 ) . / ; +(1,2)1
T:[(/'":)"H,2|W1',
Ar
V')^
(pur)i:l\2).j(pî"+o2,J_
At
')MI>
(P.-)"jf'|l 2, -
Al
(.-)",-+11,2 _
,-
~ Az{P'->
(41)
Pi+iJ
'
V)
,.
2,]
2> (F:)" J + ,2,1
..
A r(P'J
(P'.j - P'+ l.j) +
_ -
'+(1 DJ
.
PiJ+)
+ ^ ^ ( P l i P"j+i)
+ s".j + (,2,
(42)
where R"+l 2)J and 5" + ( , 2) contain nonadvanced time factors; the
weighting functions and 0 with magnitudes between 0 and 1 denote the
relative levels of timecentering of corresponding terms; and is a hybrid
function, approximating p"+ ' and defined by
Pu = Pi + UPI.V PIJ)
< 43 >
216
P. FA SOLISTELLA
A ND . V. JONES
j + 1/2
i+1/2
j -1/2
FIG.
where c" = (Cp/Cp)'' is closely related to the square of the speed of sound.
Eliminating by algebraic substitution (pMr)"+(', 2 ) J , (pwz)"J+,i 2) in equation
(41) and substituting p"* '' in eqn. (41), by means of eqn. (42), an implicit
equation in pu is obtained:
Pi.] =
where the
iPil.j
+ A 2pi.j
1 + iP+
\.j + A A P.J*
+ ,
(44)
,2, and A . factors are functions of

1
,( !
1
BLJ = + 202 [ + -
c,j
\Ar_-
Equation (44) becomes the usual approximation to the elliptic Poisson

equation c,; -> co. In some circumstances a direct solution method can be
used; for others it may be necessary to employ a relaxation technique. The
determination of for every cell allows the calculation of p"+ [ by means of
eqn. (42) and the calculation of new values of veloci ties from eqn. (41). The
energy equation can then be solved explicitly.
In the ICE method there are, of course, restrictions upon the time step
(although they are less onerous than the Courant condition), the chief of
which is the so-called Euler condition At < min(A.v/|w|).
The reactor containment codes SEURBNUK-2 2 8 and ICECO 2 9 employ
this technique.
2.5. Othe r Me thods of Inte gration
We shall devote only a few words to the other methods employed to solve
the equations of motion of compressible fluids.
In one-dimensional problems the method of characteristics can give
accurate results in the description of shock waves: the equations are put in
217
the characteristic form and the discretisation is made along these lines (Fig.
9). This method becomes very cumbersome in multidimensional cases and
applications to twodimensional situations are few. 10
Semianalytical techniques are used in particular when properties of the
flow allow simplification of the original system of partial derivative
equations (Fourier analysis, integral equations, methods of lines, etc.).
FIG. 9. Characteristics method discretisation.

However, the finite element technique seems to be a method with power
comparable to that ofthe finite differences for this class of problems. FE has
attracted interest in fluid dynamics only in the last few years, so that FE
experience is less than that for FD in this field, and one cannot yet come to a
final decision regarding the relative advantages and limitations of these
techniques for fluid dynamics problems.
In the field of application the continuum is subdivided into a set of
isoparametric triangular or quadrilateral elements identified by the
coordinates of their vertices or nodes (Fig. 10). Within a finite element (e)
the fluid velocity is approximated by continuous functions. A linear
function of the nodal values is generally adopted in compressible
problems: 31
'
(45)
where i/'0' is the approximation of the true solution, the 'shape functions'
N{kci are linear or bilinear functions ofthe spatial coordinates in the element
(e) and u.c) are the values of u at the element nodes.
Under this hypothesis the pressure, density and internal energy are
218
constant on each element and are directly calculated by solving the

conservation equations of mass and energy and the equation of state for
each finite element (e). To calculate the velocity components the Galerkin
method is applied to the momentum equations and to the boundary
condition. The FE method may be applied with Lagrangian, Eulerian.
mixed or hybrid coordinates. As in FD, the corresponding algorithms have
the same types of difficulties: distortions and inaccuracies ofthe long-term
|
U
FIG. 10. Quadrilateral and triangular finite elements.

results in Lagrangian schemes and a need for sophisticated treatment ofthe
advective terms in Eulerian or mixed coordinates.
FE and FD methods originate from substantially different mathematical
approaches, but the discrete equations are very similar for particular cases
and have common features. It may, however, be noted that for the FE
schemes it is more difficult than in FD to define the order of approximation
and to make a stability analysis.
There are great differences between FD and FE techniques in the
definition ofthe mesh and in the logical structure ofthe program: 3 2 in the
FD programs the meshes are implicitly numbered by an ij system because
of the regularity of the zones, whereas in FE the elements are explicitly
defined. In FD methods the velocities ofthe nodes are expressed directly in
terms of the pressure gradients of the neighbouring cells, while the
boundary cells require special treatment.
In FE force contributions exerted by each element are calculated for all
nodes and then these contributions are assembled to obtain the nodal
velocities. The advantages of this aspect of FE are substantial with
problems having different types of meshes connected together. However,
these differences are only partially inherent to the corresponding
approaches, and one sees signs of an exchange of ideas between the two
219
procedures: for example, the FD code YA QUP ,:> employs a nodal force
technique as in FE, whereas some FE programs have incorporated /,/
systems to number fluid elements.
In conclusion, FE fluid dynamics compressible codes give results, having,
in general, the same degree of accuracy as FD codes. Major programming
effort is required but iscompensated by advantages in the treatment of fluid
structure interactions. We shall come back to this point later.
2.6. Treatment of Discontinuities
One ofthe most striking features of compressible fluid flow is the presence
of shock waves. A shock wave is produced in a fluid when a succession of
compression waves, each propagating faster than its predecessor, pile up.
determining an abrupt transition of the field variables.
In the solution of inviscid compressible equations shock waves are seen as
discontinuities of the solutionthat is, surfaces across which the fluid
variables and their derivatives are discontinuous. In nature the shock wave
has a finite thickness, usually very small, associated with the mean free path
ofthe particles ofthe fluid. Shock waves are entropyproducing, so that
over the thickness ofthe shock the kinetic energy ofthe short wavelengths is
transferred into internal energy.
Mathematically the solution of the inviscid equations in the case of' a
shock discontinuity is guaranteed by RankineHugoniot relations relating
the fluid variables across the shock. 9
In the numerical solution, rather than follow the shock discontinuity and
attempt to decide where RankineHugoniot conditions have to be applied
(shock fitting technique 30 ), several mechanisms have been tried in the
discrete equations to take shocks and their motion into account
automatically. The difficulty is that the thickness, s , of a shock wave is
much smaller than the mesh step mean length, /: in the absence of spatial
dissipative mechanisms, oscillations develop behind the shock wave
because it is not possible to represent correctly the physical process by
which the kinetic energy of the wave is degraded into internal energy by
molecular collisions.
The idea of introducing an artificial viscosity term to transform the
energy of the largemesh oscillations into internal energy of the fluid was
first proposed by Richtmeyer and Morton. 3 0 This has the effect of
broadening the thickness of the shock to at least several zones of the mesh
and of smearing the oscillations.
In general, it is possible to define artificial viscosity terms with the
satisfaction ofthe Rankine Hugoniot conditions on either side ofthe shock
220
without degrading the results over the remainder ofthe flow field. One of
the most common forms of the artificial viscosity is
(rtQpA(VQ)2
J0
ifV-fl<0
ifV-B>0
(46)
This term is to be added to the hydrostatic pressure ofthe cell; A is the cell
area; aQ is a constant whose value controls the spread ofthe shock and is
determined empirically (a standard value is 2-56).
Frequently, and particularly in the treatment of solids and liquids, a socalled linear pseudoviscosity is required: 33
r L cp/f 1 2 |V-B|
q
^\0
tfV-B<0
ifV-u>0
(47)
In liquids and solids the linear pseudoviscosity is employed in association

with the quadratic pseudoviscosity to reduce spurious oscillations; aL must
be small to avoid excessive diffusion of shock wave front (ocL =* 5 IO" 2 ).
When the artificial viscosity terms are to be included in explicit schemes, the
stability time step may be reduced by the effect of these diffusion terms. 34
To be consistent, it is obviously essential to introduce the artificial
viscosity terms in both the momentum and the energy equations. Instead of
adding explicit artificial viscosity terms, a damping mechanism may be
added implicitly by choosing an appropriate form of the difference
equations. In particular, in the Eulerian schemes the approximation of
advective terms can introduce non-zero coefficients of second-order space
derivatives producing a net diffusive effect on the shock discontinuities.
These methods may still require additional explicit artificial viscosity to
stabilise strong shock calculations.
The methods using implicit artificial damping are reviewed in detail by
Roache. ' 3 We mention here in passing the upwinding difference schemes
that is, all those first-order schemes in which for stability and accuracy each
of the advected properties p, pu, pe is differenced according to the sign of
the advection velocity on each cell side. The PIC, FLIC and YAQUI
techniques have this feature and introduce an effective 'viscosity' through
the truncation errors of the one-sided differences.
Implicit dissipative mechanisms are present also in the Lax-Wendroff
method and in all the two-step methods through the truncation errors. For
the FE algorithms it seems to be difficult to establish the nature of implicit
pseudoviscosity introduced by the algorithms, and at present the only check
is the quality of the results.
22 1
Another type of discontinuity, the contact surface or material interface

between two different fluids, can exist in a fluid.9 In contrast with a shock,
fluid does not flow through a contact surface. Mathematically the
conditions to be satisfied at the contact surface are the continuity of the
pressure and of the normal component of the velocity, whereas the density
and internal energy are discontinuous. Concerning the numerical
treatment, in the Lagrangian approach there are no problems, because
the material interface is identified by a mesh line during the calculation. In
the Eulerian approach the interface treatment becomes more difficult,
because at each time step the new interface position must be identified in the
fixed mesh and there are a number of mixed cells containing different fluids
and so requiring special treatment.
Frequently, in two-dimensional codes, such as T O I L 2 3 and H E L P , 2 4
both velocity components are assumed continuous across the interface to
simplify the multi-fluid cell problem. It is worth noting that in the
Lagrangian approach, if thecontact surface is represented by a Lagrangian
mesh line without sliding, both components of the velocity are implicitly
assumed continuous.
At each time step the interface in a Eulerian mesh is identified by means of
massless marker particles which permit the calculation of the different
partial volumes in the mixed cells.
In the mixed coordinate codes the interfaces can be associated to
Lagrangian nodes of the mesh.
3.
STRUCTURES AND FLUID-STRUCTURE INTERACTION
3.1. Structures
In view of the remarks concerning the customary simplifications of the
fast reactor containment problem in the first section of this chapter, we shall
specialise immediately to structures which are figures of revolution about
the vertical axis of symmetry of the reactor. For thick structures such as the
diagrid, one may make use of the fact that their displacements are always
small, and calculate them by Lagrangian techniques, either finite
difference33 or finite element. 36 The equations to be solved are similar to
eqn. (33) but with pressure replaced by the full stress tensor and more
complex equations of state including elastic-plastic behaviour. 37
Whichever type of modelling is employed, since motion parallel to, rather
than through, the structure is small, the use of slide-line techniques is not
necessary.
222
The other important structures which are figures of revolution are the
inner and outer vessels or, in different terminology, the core barrel and the
primary envelope. Both these structures may be considered as thin shells for
analytical purposes, since in all commercial reactor designs and in most
model experiments the ratio of thickness to radius curvature is
approximately 1:50. We may to the approximation employed here consider
them to be symmetrically loaded, and the chief theoretical complications
arise from the fact that they are not, in general, of the same thickness
throughout, that they may suffer quite large elasticplastic deformation
under popular loading assumptions, with strains up to 10%, and that
considerable fluid flow tangential to the shells takes place, necessitating the
numerical treatment of sliding (for which see later in this section, and also
Section 4) in Lagrangian coordinates and of irregular cells with moving
walls in Eulerian coordinates.
In the early stages of development of containment codes it was customary
to model the reactor vessels as continua, subdividing the vessel thickness
into about five Lagrangian cells. This computational device was the
simplest possible, allowed considerable freedom in modelling the
elasticplastic behaviour ofthe vessel material, and could easily be inserted
into the Lagrangian containment codes containing slideline logic, which
then represented the state of the art. Its sole disadvantage was the
consequent very low value of the stability time step for these explicit codes,
set by the Courant condition in the shell material:
At
< Ah
(48)
where /t is the thickness of a shell subdivision. In numbers, for a reactor

accident requiring the calculation of"perhaps 50 ms real time to roof"impact.
the stability time step could be 5 //s or less. This severe cost penalty can be
somewhat reduced by substepping the structural calculation (i.e.
performing several structural steps per fluid step), but choice of this path
greatly complicates the treatment ofthe fluidstructure interaction, and it
has not generally been followed. Code developers have preferred instead to
use for the vessels one ofthe several existing versions of thin shell theory,
which ignores propagation of waves through the thickness ofthe shell, the
process which had such a limiting effect on the time step in the previous
approach.
3.2. Thin Shell Theory
In view of the level of this course, the treatment of thin shell theory
FIG.
223
Axisymmetric shell coordinates and forces.
presented in this section will be brief, and will serve mostly to define the
notation for the account of fluid-structure coupling which follows.
Taking the usual path of development, 3 K J 9 we assume that the thickness
ofthe shell is small compared with its principal radii of curvature, that shell
deformations are small (again compared with these radii), that stresses
normal to the thin dimension are negligible and that material lines originally
normal to the shell reference surface (usually the shell mid-surface) remain
normal to it and unstrained. This last assumption is sometimes called the
hairbrush hypothesis. Finally, in accordance with our assumptions about
axisymmetric loading, we assume that the shell is originally a figure of
revolution and retains axial symmetry during deformation. A useful
reference to the somewhat simplified theory which obtains in this case is
reference 40.
Under these assumptions we may write the equations for the dynamic
equilibrium of an infinitesimal element ofthe shell. Figure 11 shows such an
element, and the various forces and couples acting upon it. The shell
224

(rN^de^dsde^rNf,)
--daf, (,
- + ^
rM+de + d s d 8 ^ ( r M ^ )
Pj;rd9ds
(rQ)d8+
ds
(rQ)dsde
Ng d 8 d s
P 0 rdSds
Mgdds cos
Forces acting on a shell element.

coordinates are taken to be s, the arclength along a meridian from some
fixed datum plane; 0, the angular displacement from a given meridional
plane; and , the complement of the angle formed with the axis of
symmetry by the tangent plane. An additional positional parameter is r, the
distance from the axis to the point in question. We also define the unit
vectors s. and to lie in the directions of increasing 5 (along the meridian),
of increasing 0 (along the circumference) and of increasing (normal to the
tangent plane), respectively, so that (,,) forms a righthanded set. The
forces acting on the element of the shell are then TV^s, and ), where
NJ> is the force per unit length of element edge in the (, ) plane and Nt, Q
are similarly defined.
Because of possible bending moments in the shell, we must add couples
per unit length of element edge Mes and , as shown in the diagram. By
considering the equilibrium of the element about (see Fig. 12) we may
express Q in terms of the couples M0 and and write
rQ=(rM
ds
M(,cos
(49)
225
If the circumferential and meridional curvatures, 0, , are defined by

= sin /r;
= ~^
(50)
we may derive expressions for the resultants ofthe forces on the element per
unit facial area, P and P,\
Pn = N^+NuXl)--(rQ)
v
r d.v
1d
cos d)
, = -^(')-0~^+
(51)
(52)
where Pn is the stress in the direction and P, that in the s direction (normal
and tangential stresses).
If P0 and P, are the fluid pressures outside and inside the shell,
respectively, and we assume that both act normal to the shell (i.e. that
viscous drag on the shell is negligible), we may write down the equations of
motion of the shell element:
hPsan = P0 - , + P n ;
A p A = P,
(53)
where p5 is the local density ofthe shell material and an, a, are the normal
and tangential accelerations. In the global r-z coordinate system we then
have
d2r
-J-T = -a n sin<p + a, cos ;
d/ -
d2z
- = an cos a. sin
dt
(54)
for the radial and axial accelerations of the element.

To close the system of shell equations it is necessary to express
N0, , Mu, in terms ofthe shell stressesand to relate the shell stressesto
the deformations by some constitutive law. If a coordinate is introduced
which is zero in the shell midplane and which increases in the direction of ,
and if the meridional and hoop stresses at any point in the shell (s, ) are
(5, ) and a0(s, ), respectively, then the forces and couples on an element
are
Ch 2
KW =
aa(s, )
= 0,}
h, 2
M.(s) =
(55)
(., ) d i
- ft/2
The strains are assumed to vary linearly through the thickness of the
226
shell. The calculation ofthe stresses and strains is made in an incremental

fashion: stress increments are related to strain increments and to current
stresses by the shell material constitutive law. If, for conciseness, temporary
two-component vectors are formed ofthe stresses, strains and curvatures
by = (,,,)1, a = (0,), etc.. the incremental formulation may be
expressed symbolically
= (,)
(56)
while the previous statement concerning the assumed strain distribution

through the shell thickness is written
() = -
(57)
This completes the equations of motion for the shell.

Stress-strain Relationship for the Shell Material
Employing the incremental formulation mentioned above, incremental
strains , are split into elastic and plastic parts:
dt:3 = + <5
= ,
(58)
while the assumption of zero stress through the shell thickness allows the
strain in that direction to be calculated immediately and the new local shell
thickness found. For elastic-perfectly plastic materials with yield function
y(a) the Prager theory of 'plastic potential' 37 gives expressions for the
incremental plastic strain components:
=^-
= ,
(59)
The most popular yield function in containment codes is that ofthe von
Mises criterion:
= 2 + 2 - 0 - 2 < 0
(60)
in which is the yield stress, which is combined with the Reuss flow rule to
obtain the (elastic)' stresses 0, at any time.
For plastic-hardening materials (and the stainless steel of containment,
vessels is plastic-hardening) several different models have been proposed,
including isotropic hardening, kinematic hardening4-"' and the mechanical
sublayer model. 41 This last model, 42 which is now employed in several FD
and FE structural codes. 4 3 , 4 4 has generally been preferred for containment
calculations because of its relative success in simulating material behaviour
in load reversal situations.
227
3.3. Discretisation of the Shell Equations

Because of the occurrence of highorder spatial derivatives in the shell
equations, the finite difference representations of' these equations
commonly use a staggered mesh with different quantities defined at
alternative nodes. A s a Lagrangian example we may cite the REXCOH
code. 4b The treatment of thin shells in this code (in contrast to the much
more general formulation in its successor REXCOHEP 4 7 ) assumes that
/fluid 1
shell
/ fluid
j+i -.
P
j -___^ /
j-i J
/ /
M /W"~~~
i-1
/
/
+1
fc.
FIG. 13. Discretisation of shell in REXCO-H.
shells are 'almost vertical' or 'almost horizontal', and breaks down for
curved vessels such as the outer tank in Fig. 3. It is, however, adequate for
the analysis ofthe simpler containment experiments and is worth describing
both on this account and to give the flavour of more complex treatments.
Figure 13 shows the discretisation and physical variable allocation for a
portion of a shell in contact with fluid on both sides. Note that shell nodes
and fluid nodes are aligned, and that the shell is 'almost vertical'. The
equations of motion for the shell nodes (ij) are somewhat analogous to eqn.
(34). The radial acceleration is given by
M
i/ij=
-(Pij-,--
"F ('i-
j ~ "i\
+Pi)(-lJ+l
I
2/( i] + 1
ii 1 +
-O-i
Zf-1-
'+
'+
sw)
\j)
(61)
where M is the instantaneous mass of the node (a weighted average of the

masses of fluid and shell in the surrounding cells) and P, is the pressure in
the cell having as corners the nodes (ij), (i + \j), (ij + 1 ) and (i + [j + 1 ).
228
The shell strength is represented by the equivalent pressures P2 and PJ,

where
and
P2 = (^),v + (i/2)/( r .v + (i/2)Sin ;j .) + P2i

Ps = {)-2-<2,^.)
+ P23
Pu = ( ^ " ' - ' - - ~()j+11 2)''u+ ( i / 2)Cos^y]
(62)
(63)
(64)
where / is the length of the segment with end points (i,j {-), (i,j + 2).
For the 'almost vertical' wall the radial acceleration of shell node and
fluid node are assumed equal, while their vertical accelerations are allowed
to differ. Fluid may thus 'slide' along the wall, and the treatment is valid so
long as the wall is not much inclined to the vertical.
The vertical fluid acceleration is given by an expression like eqn. (61) but
taking no account of the presence of the shell:
M/j^fluw = (P,j - P,- u _ ,)(/- ij+ , - rij_l + r,_ - r+ ,_,)
- (P,-ij-
P.j-i)(rii+>
" ' " o - i +ri+ij-r,-ij)
(65)
while the shell vertical acceleration is calculated by a finite difference

representation of a simplified version of eqn. (54):
'"(^Xheii = W i j + u 2>' + ( 2) sin (/>,; - (Mu)i]-{il2)ru-a
2) sin </>,_,_,
(66)
where m is the mass of the shell segment.

The modelling of the shell dynamics given above is unsatisfactory in that
the shell is required to lie more or less along the mesh linesthat is. the
'local coordinate directions" (s, ) must not differ seriously from (r, z). There
is a corresponding restriction on the direction of the fluid flow close to the
shell. More recent developments have aimed at removing these restrictions
by allowing the shell to assume any angle to the mesh and by imposing on
the flow only the physically reasonable condition of continuity of the
normal velocity between fluid and shell at the interface without restriction
on the tangential velocities.
Figure 14 shows the shell discretisation and forces in a typical modern
FD shell formulation. The shell is here modelled as a sequence of
interconnected frustra of cones, with force variables being allocated to the
midpoints of the sections (crosses) and angle and position variables to
the nodes (dots). The shaded portion can be considered as a 'shell element'.
The FD representation of the shell equations must be so chosen that the
resulting difference equations are stable and convergent (see Section 2 for
229
definitions), and in practice they must also include a certain degree of

numerical damping of undesirable parasitic shell vibration modes. The
most easily excited of such modes is that in which successive shell segments
move inwards and outwards alternately, with a frequency which depends on
the fineness of the discretisation. This mode is the shell equivalent of the
well-known two-dimensional Lagrangian problem of'hourglassing'. The
numerical damping should quench such short-wavelength modes but must
not affect the modes of longer wavelength which represent the true physical
*-p*-i
/X/^M--^
(N^^fN^-i
FIG. 14. Shell discretisation and forcesarbitrary orientation of shells.

motions of the shell. The selection of a suitable FD shell scheme thus
requires both nice judgement and a willingness to test and assess a number
of plausible alternatives. One scheme which seems to have found
widespread acceptance is that of Witmer and Balmer. 4 8 4 9 In this
formulation the internal forces acting on a shell element are resolved in the
radial and axial directions instead of in the normal and tangential
directions, as in eqns. (51) and (52).
With the notation of this section the FD forms of the radial and axial
shell pressures are
(Pr-Az-rA0)k
= (//V^cos^,,
2)
- (rosing.,,
- (rN^cos),,_ {1
2>][rk(sk+
2)
i
sk_l)/2]~
+ [(rQs'm\
+ ( 2|
(ha0/r\
(67)
(P:AzrA0)k
= (/A ^sin) k + ( , / 2 ) )_(
2)-
-(rcos</.) t _ ( W 2 ) ][r k (5 k + 1 -sk^)/2]-
[0'cos(/;) t + ( ,
2|
(68)
230
(where sk is the total length ofall the segments from some reference circle to
node k) which are related to the normal and tangential components of eqns.
(51) and (52) by
Pn = P. cos Pr sin
, = P. sin + F r c o s
(69)
For the calculation of the forces , Nu, Q at the half-points one may
calculate stresses at all the points, both crosses and dots. This ensures that
() + (, , takes account of changes in the length ofthe segment, but entails
the calculation of stresses at points where they are not really required in
order to ensure spatial centring of the difference equations. Warham 5 0 has
obtained good results with a method in which the + (1 2) required by
Witmer's method are replaced by ak values. The loss of centring is
compensated by a halving of the effort expended in calculating stresses, and
numerical results differ little from those obtained with the fully centred
scheme. With this modification the difference equations become
(PAzrA0)k = (\[(),
+ {1 2)
+ (^-
- [CF. + U 2, 0Qko
(P,AzrA0)k = (),
+ 2,
+ [\
Qkn
2,
2) ]/2
4-
(^/^
2,]kC^+, ^ ) / 2 '
2) ]/2
(70)
(A a o cos0) 4
)/2]-
(rNJk_{ll2][rk(sk+l
(71)
The shell curvature (y^)k is estimated by

(\ = (fc+t
(72)
-)/2
while the shell force quantities are calculated from

(^)t
Ql = Qkk-i-, ( 1
+ l.2,
= K)A+/^.)/2
(73)
',%+-(\
2)
'k + l l
- (cos -
)k + ( , 2, [(hM0)k
+ (hMu)k
, ]/:
(74)
The primes indicate that '. Q' are non-centred quantities.

This completes a rather detailed description of a common FD
formulation of the thin shell equation as used in such containment codes as
REXCO-HEP, 4 7 ASTARTE 5 1 and SEURBNUK. 2 8 In discussing FE
formulations we shall be brief, since they will be more familiar to those
working in the field of structural mechanics.
Finite element discretisations of shell structures are usually based on
231
specially designed shell elements. The curvatures, angles and other

variables required for the determination of the nodal forces are calculated
automatically by expressing them in terms of the nodal positions and
displacements and using displacement functions which are linear in the
axial direction and cubic or of higher order in the radial direction. To
maintain accuracy when large rotations of an element with respect to its
original orientation occur, a system of local 'convected coordinates' is often
invoked. 53 The accuracy of the numerical realisation of the local
coordinate (,,) theory of Section 3.2, when applied to cases of large
rotations, is somewhat controversial, 57 but current opinion is that the
incremental calculation of most important variables should conserve
accuracy under large rotations if a sufficiently fine discretisation is
employed. The more general problem of what degree of discretisation is
necessary for both shell and fluid to achieve some specified accuracy in
calculated shell deformations in any particular case has not yet been
resolved theoretically for either FD or FE, and numerical experience
coupled with comparison with experience 55,56 is currently the only
available guide in this respect.
3.4. Fluid-Structure Coupling
As has been remarked previously, when Lagrangian methods are used for
the fluid, the fluid nodes must be permitted to slide along the structure,
while remaining in contact with it to ensure continuity of normal velocity
between fluid and structure. With Eulerian techniques for the fluid the
movement of the (Lagrangian) structural mesh over the fixed fluid mesh
produces many configurations in which each structural mesh or element
overlaps and partially covers several Euler meshes. The details of both cases
are discussed at the end of this section. If mixed coordinates are employed
for the fluid, the problem of coupling is solved in principle for those
portions of the structure which are always submerged, since the 'arbitrary'
velocities ofthe fluid mesh points can be made to coincide with those ofthe
shell nodes, while the fluid flows through the meshes in contact with the
structure in the same way as it flows through other meshes. Unfortunately,
there may exist portions of the structure which only come into contact with
the fluid some time after the calculation begins (e.g. the upper portion ofthe
outer vessel and the roof in Fig. 3) and some sort of sliding treatment
becomes necessary, as for Lagrangian methods. 2 6
Various degrees of coupling are possible between the two calculational
systems ofthe structure and the fluid, which may be roughly classified as
strong and weak.
232
In strong coupling, when the fluid is treated as Lagrangian, the forces

acting on the structural nodes and on the fluid surface nodes are determined
simultaneously, and both sets of nodes are moved simultaneously, so that
the normal component of velocity is always continuous across the
boundary. In the case where the fluid is treated in an Eulerian manner, the
fluid pressure in boundary cells and the flow into those cells must be such as
always to correspond to the instantaneous motion of the structure.
In weak coupling the fluid pressures are used to drive the structure to a
new configuration with new velocities. The fluid is then assumed to take up a
position which brings it into contact with the new structural position and to
have a continuous normal component of velocity over the step. The new
distribution of fluid nodes or of cell masses and momenta in Lagrangian
and Eulerian cases, respectively, leads to a new pressure field in
the fluid. This new pressure field is employed to drive the structure again
and the cycle is complete. In the case of weak coupling it is possible to use
smaller time steps for the component with the more restrictive limitation on
the time step (generally the structure), so that a finite number of substeps are
taken by one component during a single step of the other. Collocation or
matching between the two components takes place only at each full step,
and old or estimated values of fluid pressures or velocities are used to drive
the substepping component. If the fluid calculation is Eulerian, weak
coupling operates in this manner: the present structural velocities result in
changing volumes ofthe fluid cells bordering the structure, which, as far as
the fluid is concerned, have the same effect as influxes or effluxes of fluid
through the walls of those cells. In an implicit Eulerian technique of the
ICE 1 5 type the new fluid pressures may be found by use of a version of eqn.
(44) for the border cells in which the A contain mass and momentum
contributions from these fictitious fluxes. This pressure field is employed to
drive the structure as described above to close the weak coupling cycle. It is
again possible to substep the structural calculation by using for the
fictitious fluxes input to the fluid pressure calculation their mean values over
the fluid time step.
In general, then, we may visualise a matrix of possible methods made up
of all the combinations of explicit-implicit, Lagrange-Euler for the fluid,
explicit-implicit for the structure and strong-weak for the coupling
employed between them. Not all entries in the matrix represent viable
calculational procedures for logical and practical reasons, as is explained in
the review of Belytschko. 32 In practice, containment codes have tended to
use explicit methods for the structure and either Lagrangian explicit 47 5 ' or
Eulerian implicit 28 29 methods for the fluid, with weak coupling in either
233
case. An exception is the FE ALE code EURDYN-1M, 2 6 where the

combination is again explicit-explicit but the coupling is strong at positions
where no sliding treatment is necessary. A detailed discussion of this code is
the subject of the chapter by Dr Dona.
After these general remarks concerning the possible types of coupling
between fluids and structures, we now specialise to the case of weak
coupling between liquids and thin shells and give some explanation of how
this is managed in containment codes. Following the sequence of the
previous section, we begin with REXCO-H as being typical of early
developments in this field.46 where the fluid is treated in a Lagrangian
manner, the fluid calculation is explicit Lagrangian and no substepping is
employed for the shell. We then go on to discuss more sophisticated explicit
Lagrangian codes 4 7 5 1 , 5 6 and conclude this section with an exposition of
the coupling mechanism in implicit Eulerian codes. 2 8 , 2 9
As already explained in Section 3.3, eqns. (61), (65), (66), in REXCO-H a
single radial acceleration is calculated for 'nearly vertical' walls, while
vertical accelerations for shell and fluid are calculated separately. In
consequence, when the shell and fluid nodes are moved to their positions,
their radial coordinates will coincide but, because of differences in axial
coordinates, the fluid node will not generally be in contact with a shell
segment. In the spirit ofthe other approximations made for'nearly vertical"
walls, the code ensures continuity of radial rather than of normal velocities
by moving the fluid nodes radially onto the shell. The shell is then rezoned
to ensure that the shell nodes coincide with the new positions of the fluid
nodes, for which process new values of shell element thicknesses, velocities
and strains are computed by averaging procedures. The shell and fluid
nodes then coincide at cell points, the configuration is again as in Fig. 6 and
the calculational cycle is closed. Note that with this treatment of the
coupling problem the shell is not handled in a Lagrangian manner but is
rezoned each cycle. The accurate following of material histories has been
sacrificed for the convenience of having a simple interface between the fluid
and the structure at each cycle.
Lagrangian containment codes of more recent development 47,5 ' ' 5 6 make
use of slide line logic to couple fluid and shell. Section 4 explains the
concepts of" master and slave lines in a general context, and we may
concentrate here on their application to the particular problem of
fluid-structure coupling.
The slide line is so arranged as to coincide with the wetted shell surface,
and the fluid nodes are attached to the 'slave' side of the slide line in
positions in general not coincident with those of the shell nodes (see Fig.
234
15). In the simplest treatment the total fluid forces acting on a shell segment
AB are determined by forminga vector sum ofthe products of pressure and
directed area for the intersections of AB with fluid cells, in this case with
cells XY W and YZV. The shell internal forces are calculated from the shell
stresses and the combined forces used to advance the shell in space (eqns.
54) by one complete fluid time step, substepping the shell motion if this is
considered desirable. Ina purely explicit fluid calculation, the shell internal
SLAVE
MASTER
^y s l i d e l i n e
. s h e l l nodes
internal f l u i d
nodes
Interface f l u i d nodes
FIG. 15. Slide line treatment of fluid shell coupling in Lagrangian codes.
pressures together with the pressures in surrounding cells are used to move
the fluid nodes by one complete time step, a vector sum again being used to
obtain the total force on a cell face. In the case of cell Y VZ this sum includes
contributions from shell segments AB and BC. At the end ofthe fluid time
step nodes which were originally on the shell surface will no longer be so in
general and they are moved so as to bring them into contact with it again in
a direction normal to the shell surface. The fluid pressures may now be
calculated from the new cell volumes and the whole process repeated.
A coupling mechanism such as that described above is conceptually
simple, allows fully Lagrangian treatment of the shell and ensures
continuity of normal velocities in the long term, since nodes originally in
contact with the shell remain in contact. It is, however, rather complex from
the point of view of computer logic, and the arbitrary displacement of fluid
nodes to lie on the slide line at the interface produces perturbations which
may show up as pressure oscillations at the shell (see example of Section 4).
Some more gradual matching of shell and fluid normal velocities is clearly
desirable, but it has proved difficult to achieve a satisfactory scheme from
this point of view without an explosion in logic requirements.
235
To conclude this section, we describe the essentials of the coupling

between shell and fluid when the motion of the latter is calculated by an
implicit Eulerian technique of the ICE 1 5 type. Figure 16 shows the
intersection of a Lagrangian shell with the rectangular Eulerian mesh, and
the positions at which radial velocities (u) and axial velocities (v) are stored.
Notice that at mid-points outside the shell these velocities are not defined,
since no fluid is present.
SHELL
FIG. 16. Eulerian fluid cells and Lagrangian shell nodes.

To explain the treatment of shell motion, it is convenient to take first the
case of a shell with a prescribed motion. The condition of the continuity of
the normal velocity at the shell leads to consideration ofthe conditions of
continuity of mass and momentum within the boundary cells which have
the shell as Lagrangian boundary. One such cell is shaded in Fig. 16. The
continuity equations allow the calculation ofthe velocity field within the cell
and, hence, at the shell. Equating shell and fluid normal velocities produces
a relationship between the fluid momenta at the cell mid-sides and, hence,
between the pressures in the boundary cells and in the surrounding cells.
This relationship is a Poisson equation similar to eqn. (44) but with
modified coefficients A. Its solution for boundary cells may therefore be
integrated with that for the entire fluid region.
To complete the exposition, we now generalise to the case where the shell
no longer has prescribed velocities but moves under hydrodynamic and
internal forces. In this case, in accordance with the general lines ofthe ICE
method, the fluid pressures which contribute to the shell acceleration are
236
taken to be the timeadvanced values. The mean normal velocity ofthe shell
over the time step includes contributions from these pressures as well as
contributions from the shell internal forces. The two contributions are
separated, and matching of normal velocities at the shell as in the previous
paragraph again leads to a Poisson equation but with the A coefficients
taking account of the acceleration imparted by the fluid to the shell and with
the shell internal forces included in the source term. Substepping is possible
with this technique, the A in this case being again modified to take into
account the change in shell configuration over the fluid time step. 2 8
The logical problems with this technique arise from the fact that in some
cells not all of the required information is available. In some cells with a
Lagrangian boundary only one or two ofthe velocity field components will
be defined, the remaining midside points ofthe Eulerian cell being outside
the fluid. It is therefore necessary to extrapolate from surrounding cells,
taking care in the logic that these do in fact have their full complement of
defined midside velocitiesthat is, that they are not themselves boundary
cells. The logic for such extrapolation procedures can become exceedingly
involved when all possible configurations of shell and fluid are considered,
and this logical complexity forms the basic drawback to Eulerian codes as
applied to fluidstructure problems.
These descriptions of fluidstructure coupling techniques conclude this
section.
4.
CURRENT STA TUS A ND FUTURE TRENDS
4.1. Experience with Containment Codes

In the previous sections we have outlined the models and the numerical
methods for the treatment of compressible fluids, of structures and ofthe
fluidstructure interaction. In this subsection we shall turn our attention to
some existing reactor containment codes and to the quality of their
production. Table I is a list of the fluid structural codes used or under
development at the Joint Research Centre (JRC) in fast reactor
containment studies.
REXCOEUR 6 5 is a JRC development ofthe now outmoded REXCOH
code 4 6 written by A rgonne National Laboratory. It is the simplest of the
twodimensional Lagrangian fluidstructural codes which have been used
to study containment models. A STA RTE. 5 1 from A WRE (UK), and
ARES, 51 ' from Interatom. are more sophisticated codes designed for the
same purpose. All these codes employ explicit Lagrangian schemes like eqn.
(33) or eqn. (34) t treat fluids, and, as already mentioned in Section 3.
H
PC
>
Z
H
r
r
TABLE I
7
y.
Code
Fluid
mesh
Flow type
Discretisation
type
Slide line
Rezoning
Structure
continuum
equations
Thin shell
treatment
H
7:
C
H
r
REXCOEUR
ASTARTE
ARES
SURBOUM
SEURBNUK.S
EURDYN/IM
Lagrangian
Lagrangian
Lagrangian
Eulerian
Eulerian
mixed
compress.
compress.
compress.
incomp.
compress.
compress.
II)
ID
FD
ID
FD
FE
limited
yes
yes
no
no
limited
no
no
yes
no
no
continuous
no
yes
yes
no
no
yes
yes
yes
yes
yes
yes
yes
m
-i
C
B)
s.
50
rr
>
n
H
C
je
>
n
m
H
<
K)
U)
238
various approximations of thin shell theory in Lagrangian coordinates to

treat the shell structures. Because of the Lagrangian approach, the fluid
mesh becomes distorted as the calculation proceeds, leading to inaccurate
results and severe restrictions on the time step to assure stability of
calculations. To try to improve the quality of the solution, damping
mechanisms such as mesh rgularisation 57 or rotational pseudoviscosity 58
have been introduced in the REXCO-EUR, ASTARTE and ARES codes.
Mesh rgularisation is a mechanism to keep each node in the centre of the
fluid zone defined by its eight neighbouring nodes. To this end a combined
elastic and viscous force is applied to each node. Rotational pseudoviscosity aims at reducing the well-known 'hour-glass' distortions ofthe mesh.
In this case additional components proportional to the relative rate of
rotation of opposite sides of a cell are applied to the nodes.
Considerable skill is needed in applying these techniques so that mesh
distortions become less troublesome, while the overall dynamics are not
significantly affected.
Rezoning techniques seem a more suitable tool to extend the validity of
Lagrangian calculations, but there are problems in their application,
because it is difficult to define general criteria for the automatic
construction of a new regular mesh in each problem. At the same time, it is
difficult to ensure satisfactory conservation of both momentum (a linear
function of the velocity) and the total energy (a quadratic function of the
velocity).
Many criteria have been applied in Lagrangian programs to remap the
mesh. There are two main possibilities. The first is to move the mesh points
at fixed times by considering only the geometric properties of zones,
such as angles, side lengths, areas, etc.. and trying to give a more
regular shape to each zone (Fig. 17). The other is to build a new
mesh taking into account the actual motion of the fluid, so that the
rezoned mesh is adapted to the type of flow being calculated (Fig. 18).
A great difficulty lies in the application of rezoning techniques near
material interfaces, because in order to preserve one of the advantages
of the Lagrangian approach, that of the accurate following of interfaces, it is, in general, preferred to associate the same Lagrangian nodes
with the interface during the entire calculation.
The success of the application of rezoning Lagrangian nodes is. in
general, limited, but rezoning techniques are still under development.
Indeed, as we shall see, the use of mixed coordinates involves the definition
at each time step ofthe displacements ofthe mesh nodes, and this procedure
may be considered as a continuous rezoning of the mesh.
239
1 ) Method of e q u a l areas
x 0 is moved to x 0 0 in o r d e r to make t h e
areas of A as equal as possible
2 ) C e n t r o i d of c e n t r o i d s method
FIG.
17. Rezoning criteria.
33
Slide lines have been introduced into Lagrangian codes with the aim of
decoupling the tangential notion of fluid and structure meshes or of two
portions ofthe fluid. In some particular cases they may even be useful in the
treatment of fluid distortions. In FD, ED codes the slide lines are generated
as the joins of pairs of contiguous points of constant i (or j). One line is
specified as the master, the other as the slave line (Fig. 19). During a
calculation both surfaces of a slide interface are advanced, assuming that
the normal acceleration is continuous across the interface but that the
tangential accelerations may differ on opposite sides of the interface, so
allowing a discontinuity in the tangential velocity to develop. The master
material provides a geometrical boundary to the slave material, while in
turn the slave material exerts a pressure force on the master material. The
Rezoning
based on v e l o c i t i e s
FIG. 18. Rezoning based on nodal velocity.
240
important point is that the parameters ofthe slave material are associated
only with the slave material and that material providing the fixed boundary
(master) is treated as if the boundary were a moving surface on which
pressure forces act.
Slide lines can be used in the fluid when large transverse motion along / or
j lines is expected at predetermined positions. In such cases it is also
possible, in principle, to use a different mesh size on the two sides.
Master
line
1 SLave
ine
1
1
Master
Stave
points
points
I
I
1
FIG. 19. Slide line.
Unfortunately, experience shows that any mesh distortion tending to
'wrinkle'a slide line blocks all tangential motion at the line and so defeats its
object.
In the Lagrangian codes which we are considering, the shock
discontinuities are treated by means of a combination of quadratic and
linear pseudoviscosities (eqns. 45 and 46). At least in the first phase of our
containment studies, these mechanisms are found to give a good
representation of the wave propagation.
Figures 20 and 21 present specimen results of a calculation of COVA
experiment IT9 made with the code ASTARTE. Figure 20 shows the
Lagrangian mesh just after roof impact. The vessel deformation is clearly
distinguishable, and incipient gross mesh distortions may be seen in the
vicinity of the charge. Figure 21 displays experimental and calculated
pressure and impulses for an Eulerian wall station at the point marked A in
Figure 20. It is clear that, while the perturbing effects ofthe fluid-structure
coupling mechanism mentioned in Section 3 have produced very spiky
pressure predictions which are difficult to compare with experiment, the
predicted impulse shows quite good agreement, especially at later times.
Figures 22 and 23 show the mesh configuration of time = 5 ms without
rezoning and with rezoning (centroid of centroids method) in the COVA
241
FIG. 20. ASTARTE mesh configuration.

IT5 calculations (rigid vessel, LDE charge). This result has been obtained
with the code ARES. As it is possible to see, the application of rezoning has
had a limited effect on the mesh distortions. In the rezoned mesh the
tangling of mesh lines has disappeared but at the same time a certain nonorthogonality has spread over most of the mesh.
The limitation of Lagrangian codes in the description ofthe second phase
of the containment problems (inertial phase and roof impact), especially
when internal structures are present, has necessitated the development of
more powerful and sophisticated codes, such as pure Eulerian or mixed FD
or FE codes.
At present SEURBNUK-2, developed jointly by UKAEA and JRC, is
one of the most advanced codes in the field of reactor containment.
SEURBNUK-2 is an Eulerian two-dimensional axisymmetric FD code,
based on the ICE technique for the calculation of the (compressible)
coolant motion. Some configurations which the code can tackle are
IJ
4-
I J
Pai
1.20
Impulse
(lO^Pa.s)
2.00
100
2.40r
Pressure
(W7
1.60
EXPERIMENT_
/ ^
JUSTARTE
0.60
r'
*~
>
>
V.
-
0.20
0.60
1.20
ASTANTE
0 60
SEXP
\
0.00
L Ll-l-.9
0.00
12 '
10
time (ms)
Fie;. 21. ASTARTE predictions and experimental results for pressure and impulse
TRANSIENT FLUID -STRUCTURE PROBLEMS IN REACTOR SAFETY
243
=jrffi
>\^-^
||j|
jky
-~~~
j
...
PIG. 22. ARES mesh configuration
without rezoning.
FIG. 23. ARES mesh configuration

with rezoning.
presented in Fig. 24. The internal and external structures are treated in
Lagrangian coordinates by FD and approximated by thin shell theory (see
Section 3) Perforated plates are treated as porous rigid structures.
SEURBNUK-2 is a development of the incompressible Eulerian code
SURBOUM, which was developed jointly by Belgonucleaire and
UKAEA 5 9 and is based on the MAC 2 7 technique.
244
Thin or thick
tank
High p ressure gas

i gid
diagrid
Cover gas
Thin tank-.
\
\
\\
A\V
)
c)
di
e)
f)
:>
^
h)
isl5
\
s
3 KV
k)
i)
S3
ji
DE3
3
)
l)
FIG. 24.
mi
SEURBNUK-2 configurations.
245
As we mentioned before, one of the most difficult problems in Eulerian

fluid dynamics is the treatment of boundaries and material interfaces with
reference to the fixed mesh. Massless marker particles indicate the position
of free surfaces at each time step. At the end of the time cycle, the new
positions of the markers are concluded by use of velocities of the mesh
pointsand the weighted areas method. 27 In the present version of this code
there is no treatment of two fluid cells: the integration domain R(r, z, t) is
I''jH
I
I ' I I I | l
"j^xW |
I lx I
"Ht"
Internal c e l l s
Boundary cells
I i | iBj
^__1J_L_LJ
FIG. 25. Cell labelling in Eulerian codes.
the coolant region (Fig. 25) and the bubble-coolant interface represents an
external boundary on which the gas bubble exerts a pressure force. The
pressure values in the boundary cells are obtained by means of
interpolations, and the uniform pressure of the gas bubble is a function
(through the equation of state) of the bubble volume.
The ICE scheme introduces an implicit diffusive mechanism: in the
containment problems of the COVA series we have not needed to introduce
an explicit quadratic or linear pseudoviscosity.
In Figures 26 and 27 are presented some results of an IT9 COVA
calculation with SEURBNUK for comparison. Figure 26 shows the
pressure and velocity distributions calculated for experiment IT9 well after
roof impact (7 ms). The outward motion of the deformed vessel and the
pressurised regions close to the roof and close to the curved bottom of the
vessel are clearly indicated. Figure 27 presents calculated and experimental
pressures and impulses for the same gauge station as for Fig. 21 (with which
this figure may be compared). The damping and smoothing effect of the
numerical diffusion present in the SEURBNUK code may be seen in the
pressure records, while good agreement is observed for the overall impulse.
P. FASOL I-STEL L A AND . V. JONES
246
..:::::::
I"
"J
_,.
t ; : : : : ; : : :
:"
___.---.
" : _--,
-H
^^-=^
::
: ::
;::::::: 2::^::::
:::::::-i
: 11 inti
\\
"
:::::::;:
--- -
r tt
-
i -
S
r
:r
::
1I
.Ij . . . . . . - . : ; : : ; ; : ; :
..c:-:
_ :
^::
S _
:_._:
~<
"s::;;::
s : " ;
;
"^1
S;:::::::J-
ff -ax/s
FIG. 26. SEURBNUK: pressure and velocity distributions.
EURDYN-1M 2 6 is an FE mixed coordinates fluid-structural code under
development at JRC for the treatment of both containment and
subassembly dynamics problems. It holds great promise of being able to
overcome many difficulties of Lagrangian and Eulerian schemes, to treat
complicated structures and to simplify the fluid-structure coupling
problem. The numerical schemes employed in this code and the first
significant tests form the subject of the chapter by Dr Doea.
4.2. Recent Developments
The previous sections have described the principal features of the
fluid-structure problem in fast reactor containment, and have given some
account of the codes at present in use in this field. In this final section we
Pressure
(MPa)
Impulse
(KPa.s)
20
20
16 -
16
73
>
y.
Fi
r.
12
.
7>
-
70
C
ra
r
7=
>
H
C
c/:
>
g
16
time
FIG. 27.
(ms)
SEURBNUK: pressure and impulse predictions and experimental results.
<
IJ
t-
248
seek to explain the directions taken by recent research work inspired by the
limitations ofthe various techniques currently available.
As mentioned previously, the codes most frequently used in containment
problems are either axisymmetric Lagrangian or axisymmetric Eulerian.
and each type of code has its own built-in limitations. Lagrangian codes
have proved very suitable for the calculation of the early stages of the core
bubble expansion, before much fluid motion takes place, since the
calculation ofthe propagation and reflection of'shock waves is their forte.
They are also well adapted to the detailed analysis of wave propagation
effects in subsections of the whole containment geometry, such as the
volume enclosed by the diagrid, the underside of the core support skirt and
the curved bottom ofthe outer vessel. On the other hand, when relative
motion is large, their accuracy declines and the calculation time required
increases rapidly because of mesh distortions limiting the time step.
Effective termination of the period of validity of Lagrange codes can
occur distressingly early when the core bubble is quite close to the
perforated diagrid, to give but one example. It is also difficult to insert the
resistance effect of nearly fixed perforated structures into Lagrangian
codes, so that both the direct modelling of perforated structures and their
substitution by distributed resistances come up against code limitations at
an early stage.
Eulerian codes are ideally suited to the calculation of the overall
progression of the fluid flow from initial core expansion to roof impact and
after. Their limitations appear most clearly when flow through complicated
geometries or deformable structures must be calculated. The insertion of
distributed resistances in Eulerian codes is quite straightforward, but
separate experiments may be necessary in order to determine what type of
resistance law should be used. In addition, when coupling with complicated
structures must be modelled, the computer logic required can become
extremely heavy.
In accordance with the practical requirements of cheap computing for
long real times of logically simpler but accurate fluid-structure coupling,
and of modelling of large fluid displacements at all flow speeds from
supersonic to far subsonic (almost incompressible), current efforts are
being devoted to developing more general techniques with both Lagrangian
and Eulerian features. This class of technique has two main representatives
in the shape of hybrid coordinates and mixed coordinates. In hybrid
coordinates the integration field is discretised by a fixed Eulerian mesh, but
selected portions ofthe field may be handled by means of a Lagrangian
mesh which is superimposed on the Eulerian mesh and fully coupled to it.
249
Structures worthy of such a Lagrangian description would include the core

bubble and the various deformable structures. In this technique the main
logical and computational problem is that of the calculation offluxesinthe
'mixed' cells where Euler mesh meets Lagrange mesh. The problem seems to
be soluble, however, and the chief representative of this type of code,
PISCES-2DELK. 2 i is able to tackle problems of considerable complexity.
The other general technique, that of mixed coordinates, has much in
common with Lagrangian methods with rezoning, of which it forms the
logical extension. As described in Section 2, the formulation of the
hydrodynamic equations in mixed coordinates allows the cell boundaries
and nodes to move in some arbitrary way, and calculates fluxes between
cells on this basis. In rezoned codes it is assumed that the nodes move with
the fluid, and when the rezoner is invoked, a new mesh is superimposed and
material variables are remapped from the old mesh onto the new. With
mixed coordinates one may allow some nodes to move in a Lagrangian way.
others to remain fixed and yet others to move in some convenient manner
not immediately related to the fluid flow. Problems of fluid-structure
coupling become very simple in principle, since fluid nodes at the structure
can remain attached to it throughout the calculation without imposing any
restriction on the fluid tangential velocity. Practical experience with
EURDYN-1M, 2 6 a mixed coordinate code, is described by Dr Dona.
With mixed coordinates and purely explicit techniques (as in the ALE 6 0
method) one cannot easily treat all flow speeds, and there is now a tendency
to build into the algorithm an implicit step in the calculational cycle which
lifts the restriction of the Courant condition when flow speeds are low.
Algorithms of the YACQUI 3 5 type result, but their application to
containment problems is only just beginning.
Besides these modern trends in numerical techniques, there is also a
tendency to insert more physics into the calculation of containment
problems. In this context we may mention the F E 3 6 modelling of complex
thick internal structures, the modelling of perforated plates 8 and the much
more ambitious modelling of heat transfer, mass transfer and phase change
processes ' in such a way as to integrate the calculation of the energy source
(the charge or core bubble) and the response of the coolant and its
containment.
Turning now to three-dimensional problems, the limitations of
customary techniques mentioned above are found to apply with redoubled
force. In most practical geometries Lagrangian codes soon find themselves
in the regime of large displacements for which they are ill-suited, while the
logic problems experienced in the treatment of interfaces and (Lagrangian)
250
structures in Eulerian codes are severely aggravated. Both techniques are

subject to practical computer limitations on storage and time requirements.
At installations with virtual storage facilities the treatment of' very large
problems is possible in principle, but because of the relatively long time
required to access stored quantities, time restrictions can become very
irksome.
Consequently, three-dimensional fluid analyses have mainly been done
for relatively simple geometries where not too many meshes are required for
adequate description. It is worth noting that most three-dimensional fluid
methods are formulated in terms of finite differences (BAAL, 5 the version
of SOLA in the review of Hirt 61 ), while structural analyses have almost
invariably been cast in terms of finite elements. 62,63 The problem of the
coupling of these two formulations in three dimensions and the application
of the result to practical containment problems is very much in its infancy.
Hence, little can be said about it and other related three-dimensional
developments at present except to remark that for really effective threedimensional work it may be necessary to invest in some sort of parallel
processing supercomputer, and to reassess numerical algorithms in terms of
their potential for vectorisation. 64
REFERENCES
1. Bell.C. R.. Bleiweis, P. B. and Boudreau. J. E. (1976).'Analyses of LMFBR
core disruption and accident phenomena using the SI M M ER-1 code'. Int.
Meeting Fast Reactor Safety and Related Physics. Chicago. Illinois. October.
2. Liebeq. G. and Smolderen, J. (1977). 'Etude numrique d'un coulement
travers un orfice bidimensionel parois minces en regime instationnaire'. Test
Report 129. Von Karman Institute.
3. D esalvo. G.J. and Swanson, J. A. (1975). A NSYS Engineering Analysis System
Users' Manual. Swanson Analysis Systems. Elizabeth. Pa.
4. Marchetas. A. H. and Belytschko. T. B. (1974). 'Nonlinear finite-element
formulation for transient analysis of three-dimensional thin structures'. ANL
8104.
5. Pracht, W.E. and Brackbill, J. U. (1976).'BAAL: a code for calculating threedimensional fluid flows at all speeds with an Eulerian-Lagrangian computing
mesh". LA-6342.
6. Schoeberle, D. F.. Chang. Y. W.. Grildys. J. and Bratis. J. C. (1976). 'Nonlinear
dvnamic analysis of reactor head covers". ANL-8101.
7. Holtbecker. H.. Hoskin. . E.. Rees. N. J. M.. Tattersall. R. . and
Verzelletti. G. (1976). 'An experimental programme to validate wave
propagation and fluid flow codes for explosion containment analysis'.
International Meeting on Fast Reactor Safety and Related Physics. Chicago,
October.
25 1
8. Smolderen, J. (1971). 'Numerical methods in fluid dynamics. Introductory

remarks". Lecture Series 34, Von Karman Institute.
9. Courant, R. and Friedrichs. K. O. (1948). Supersonic Flow and Shock Waves,
Interscience, New York.
10. Methods in Computational Physics, Vol. 3 (Ed. B. Alder). A cademic Press,
New York, 1964.
11. Lax. P. D. (1954). "Weak solutions of"nonlinear hyperbolic equations and their
numerical computation'. C o mm. Pure A ppi. Math.. 1, 135.
12. Lax. P. D. (1973). "Hyperbolic systems of" conservation laws and the
mathematical theory of shock waves", Reg. Conf. Series in A ppi. Math.. 11'.
SIAM.
13. Roache. P. (1972). Computational Fluid Dynamics, Hermosa Publishers,
Albuquerque, New Mexico.
14. Potter. D. (1973). Computational Physics. Wiley. New York.
15. Harlow. F. H. and A msden. A . A . (1971). 'A numerical fluid dynamics
calculation method for all flow speeds', J. Comp. Phys., 8, 2.
16. Hirt, C. W. (1968).'Heuristic stability theory for finitedifference equations', J.
Comp. Phys.. 2, 33955.
17. Courant. R.. Friedrichs. K. O. and Lewy. H. (1928). Math. A nn.. 100, 32.
18. Chang, Y. W. ( 1977).'Structural dynamics in LMFBR containment analysis. A
brief survey of computational methods and codes', SM1RT 4. San Francisco,
California. A ugust.
19. Browne, P. L. (1966). 'Rezone: a proposal for accomplishing rezoning in two
dimensional Lagrangian hydrodynamics problems'. LA 3455.
20. Harlow. F. H. (1963). 'The particleincell computing method for numerical
solution of problems in fluid dynamics', Proceedings of Symposium In A pplied
Mathematics, Vol. 15.
21. Gentry. R. .. Martin. R. E. and Daly. B. J. ( 1966). 'An Eulerian differencing
method for unsteady compressible flow problems'. J. Comp. Phys., 1.
22. Johnson. W. E. (1965). 'OIL. A continuous twodimensional Eulerian
hydrodynamic code'. General A tomic GA A M5580.
23. Johnson. W. E. (1967). 'TOIL. A twomaterial version of the OILcode',
General Dynamics GA MD8073.
24. Hageman. L. J. and Walsh. J. M. (1971). 'HELP. A multimaterial Eulerian
program for compressible fluid and elasticplastic flows in two space
dimensions and time'. A D726459.
25. Hancock. S. L. (1976). 'Finite difference equations for PISCES2DELK, a
coupled Euler Lagrange continuum mechanics computer program", TCAM 76
2, Physics International Co., San Leandro, California.
26. Dona, J., FasoliStella, P. and Giuliani, S. (1977). 'Lagrangian and Eulerian
finite element techniques for transient fluidstructure interaction problems',
SMIRT4, San Francisco, California, A ugust.
27. Welch, J. E., Harlow, F. H Shannon. J. P. and Daly, B. J. (1966). 'The MA C
method. A computing technique for solving viscous, incompressible, transient
fluidflow problems involving free surfaces'. LA 3425.
28. Cameron. L G . . Hankin, B.C., Warham, A. G . P., Benuzzi, A. and Yerkess, A.
(1977). 'The computer code SEURBNUK2 for fast reactor explosion
containment safety studies', SMIRT4, San Francisco, California, A ugust.
252
29. Wang, C. Y. (1975). 'ICECOAn implicit Eulerian method for calculating

fluid transient in fast reactor containment', ANL 75-81.
30. Richtmeyer, R. D . and Morton, K. W. (1967). Difference Methods for I nitialValue Problems, Interscience, New York.
31. Finite Elements in Fluids (Ed. R. H. Gallagher, J. T. Oden, C.Taylor and O. C.
Zienkiewicz), Vols. 1, 2, Wiley, New York, 1975.
32. Belytschko, T. (1977). 'Analysis of fluid-structure systems', CONFABRE, San
Francisco, California, August; Nuclear Engineering and Design, 42.
33. Wilkins, M. L. (1969). 'Calculation of elastic-plastic flow', UCRL-7322Rev. 1.
34. Burstein, S. Z. and Trkei!, E. L. (1977). 'Eulerian computations in domains
with moving boundaries', 5th Conference on Numerical Methods in Fluid
Dynamics.
35. Amsden, A. and Hirt, C. W. (1973). 'YAQUl: An Arbitrary LagrangianEulerian computer program for fluid-flow at all speeds', LA-5100.
36. Belytschko, T. and Kennedy, J. M. (1975). "Response of core grid support
structures in LMFBR to nuclear excursions'. Proc. ELCALAP
Seminar,
Berlin.
37. Prager, W. (1959). An I ntroduction to Plasticity, Addison-Wesley. Reading.
Mass.
38. Flgge. W. (1967). Stresses in Shells. Springer-Verlag. Berlin.
39. Krauss. H. (1967). Thin Elastic Shells. Wiley, New York.
40. Reissner. E. (1941).'On axisymmetric deformations of thin shells of revolution'.
Proc. Symposium in Appi. Maths., 63.
41. D uwez, P. (1935). 'On the plasticity of crystals', Phys. Rev., 47, 494.
42. Hunsaker. B., Vaughan. D . K.. Stricklin. J. A. and Haisler. W. E. (1973). 'A
comparison of current work-hardening models used in the analysis of plastic
deformation'. TEES 2927-73-3, Texas ASM University. October.
43. Atluri. S., Witmer, . .. Leech. J. W. and Morino. L. (1971). 'PETROS 3:
Finite difference method and program for the calculation of large
elastic-plastic dynamically induced deformations of multilayer variable
thickness shells'. MIT. ASRL TR 152-2 (BRL CR 60). November.
44. Key. S. W. (1974). 'A finite element procedure for the large deformation
dynamic response of axisymmetric solids'. Comp. Meth. Appi. Mech. Eng.. 4,
195-218.
45. Argyris. J. H. and Chan, A. S. L. (1972). "Static and dynamic elastoplastic
analysis by the method of finite elements in space and time'. International
Symposium on Foundations of Plasticity. Warsaw. August.
46. Chang, Y. W.. Grildys. J. and Fistedis. S. . (1971). 'D ynamic response of
reactor containment to high energy excursions'. SMIRT 1. Berlin. October.
47. Chang. Y. W. and Grildys. J. (1975). 'REXCO-HEP. A two-dimensional
computer code for calculating the primary system response of fast reactors".
ANL-75-I9. June.
48. Witmer. . .. Balmer. H.A.. Leech. J. W. and Pian. T. H. H. ( 1963).'Large
dynamic deformations of' beams, rings, plates and shells", AI AA Jl. 1(8),
1848-57.
49. Balmer. H. A. (1965). "Improved computer programs D EPROSS 1. 2 and 3 to
calculate the dynamic elastic-plastic two-dimensional response of impulsively
loaded beams, rings, plates and shells of revolution', ASRLTR 128-3. August.
253
50. Warham. A. G. P. (1977). Private communication.

51. Cowler, M. S., Hoskin, N. E. and Rowlinson, A. G. (1972). 'Calculation ofthe
dynamic response of reactor containment systems to full core explosions'.
Comp. Phys. Comm.. 9.
52. Belytschko, T. and Hsieh. B. J. (1974). "Nonlinear transient analysis of shells
and solids of revolution by convected elements", AIAA Jl, 12(8), 1031-5.
53. Belytschko. T. (1975). In Nonlinear Analysis Descriptions and Numerical
Stability in Computer Programs in Shock and Vibration (Ed. Pilkey and
Pilkey), Shock and Vibration Information Center. Washington, DC.
54. Greenspon, J. E. (1967). Fluid Solid Interaction, ASME. New York.
55. Szilard. R. (1973). Hydrodynamically Loaded Shells, University Press of
Hawaii. Honolulu.
56. Doerbecker. K. (1973). 'ARES. Ein zweidimensionales Computer Program'.
SMIRT 2, Berlin, September.
57. Butler, T. D. (1970). 'LINC method extensions'. Second Conference on
Numerical Method in Fluid Dynamics. Berkeley. California.
58. Cherry. J. T. et al. (1970). 'Two dimensional stress-induced adiabatic flow'.
UCRL 50987.
59. Namias. M. and Lamotte, H. (1972). 'Description of the SURBOUM
Program', Belgo Nuclaire 308.39/177/No. 40.
60. Harlow. H. and Amsden. A. A. (1971). 'A numerical fluid-dynamics
calculations method for all flow speeds'. J. Comp. Phys.. 8, 197.
61. Hirt.C. W.. Ramshaw. J. D. and Stein, L. R. (1978). 'Numerical simulation of
three-dimensional flow past bluff bodies'. Comp. Math. Appi. Mech. Eng.. 14,
93-124.
62. Dona. J.. Giuliani. S. and Halleux. J. P. (1976). 'Theoretical aspects ofthe
Eurdyn computer programs'. EUR 5473e.
63. Bathe. K. J., Wilson, E. L. and Iding, R. H. (1974). 'NONSAP'. Rep.
UC SESM 74-3. University of Berkeley. California.
64. Patterson. G. S. (1978). 'Prospects for computational fluid mechanics'. Ann.
Rev. Fluid Mech.. 10, 289-300.
65. Benuzzi, A. and Fasoli-Stella. P. (1977). 'REXCO-EUR code. User's manual'.
EUR 5813 EN.
8
Finite Element Analysis of Transient Dynamic
Fluid-Structure Interaction
J. DONEA

Italy
1.
INTRODUCTION
The general topic of dynamic fluid-structure interaction is indeed a

particularly broad subject in that it simultaneously brings together all the
aspects associated with both structural dynamics and fluid dynamics. Each
of these two areas are complex by themselves; however, when considered
together, the situation becomes even more complex. In fact, the dynamic
interaction (or coupling) between the fluid and solid response can be viewed
as a feedback loop of the type illustrated in Fig. 1 : the structure surface
loading is not known a priori but depends on the interface pressures in the
fluid; the fluid response is in turn a function of the structure's surface
motion.
It is the scope of this chapter to show how the finite element method
(FEM) can be employed as a numerical tool to solve coupled fluid-structure
problems under transient dynamic conditions. Three distinct tasks are
involved in applying the FEM to such problems:
(1)
(2)
(3)
The development of a fluid analysis algorithm.

The development of a structural analysis algorithm.
Coupling the fluid and structure algorithms.
While efficient finite element procedures are available for the non-linear
transient response of complex structural components, it is only in very
recent years that the potential of the FEM as a discretisation technique for
problems in fluid dynamics has been thoroughly investigated. A special
emphasis will therefore be placed in these notes on the FEM treatment of
the hydrodynamic part of the problem, as well as on the coupling between
the fluid and structure algorithms.
255
256
J. DOEA
DISPLACEMENT
AND STRESS
FIELDS
EXPLICIT FORCES
A
STRUCTURAL RESPONSE
V
MOTION RESPONSE
AT
INTERFACE
INTERFACE
PRESSURES
I
FLUID RESPONSE
1
PRESSURE AND
VELOCITY FIELDS
FIG. 1. Feedback loop in fluid-structure problems.

The hydrodynamics formulation will be presented in Section 2. It is
applicable to both compressible and incompressible flows and is based on
an arbitrary Lagrangian Eulerian description. Section 3 briefly discusses a
particularly efficient algorithm for structural analysis in the presence of
geometric and material non-linearities. The coupling between fluid and
structure algorithms is then introduced. Finally, selected applications are
presented in Section 4.
FLUID ANALYSIS ALGORITHM

In determining a method for the finite element analysis of transient fluid
dynamics problems the major choices confronting the analyst are:
(1)
The choice of mesh description: Lagrangian. Eulerian. mixed

Eulerian-Lagrangian or arbitrary Lagrangian Eulerian.
TRANSIENT DYNAMIC FLUID-STRUCTURE INTERACTION
(2)
(3)
(4)
257
The choice of an algorithm appropriate to the flow regime of

interestthat is, whether to use a compressible technique or an
incompressible technique.
The choice of time integrationthat is, whether to use an explicit or
an implicit method.
The choice of appropriate finite elementsthat is, whether to use a
fine mesh of simple elements or a coarser mesh of higher-order
elements.
For the transient fluid-structure interaction analysis discussed here, we

make the following basic choices:
(1)
(2)
(3)
(4)
A rbitrary Lagrangian Eulerian mesh description, as it succeeds to a

certain extent in combining the best features of both the
Lagrangian and Eulerian approaches.
Compressible and incompressible fluid algorithm: if the fluid of
interest is liquid, the effects of compressibility are minor as far as its
flow is concerned. However, important wave propagation effects
may be expected in problems where the loading is of impulsive
nature.
Explicit time integration, as being appropriate for responses of
wave propagation type and enabling an easy treatment of material
non-linearities.
Simple finite elements, as they are particularly suited to describe
shocks and strong gradients, while higher-order elements are better
adapted to problems characterised by smooth response.
The above basic choices will be discussed in detail in the next paragraphs.
2.1. The Arbitrary Lagrangian Eulerian (ALE) Formulation
Mesh Description, Material, Spatial and Mixed Coordinates
Two basic viewpoints are generally considered in discretising a fluid by a
finite difference or finite element method. The first is Lagrangian, in which
the mesh of grid points is embedded in the fluid and moves with it; the
second, known as Eulerian, treats the mesh as a fixed reference frame
through which the fluid moves.
The Lagrangian description fixes attention on specific particles of the
continuum, whereas the Eulerian description concerns itself with a
particular region ofthe space occupied by the continuum. As shown in Fig.
2, a representative particle of the continuum occupies a point P 0 in the
258
J. DOEA
/
/Rit)
\
\
**
*
U
~ ~ ~ * * ^ * ^
J><^.
j<
j^**"
s ,^**"
^,^
, > < * ) <.
s / ^ - * "
j/^
^*
C (t)
. / **
**
J
^ /
C(0) = R(0)
C = C ontinuum
R = Reference frame
FIG. 2. Kinematics in the ALE description.

initial configuration of the continuum at time / = 0 and has the position
vector
* = (ax.a2,ai)
(1)
The coordinates a, are called the material coordinates. In the deformed
configuration the particle originally at P 0 is located at the point and has
the position vector
x = (.v1..v2..v3)
(2)
The coordinates (.v,. x2,x3) which give the current position of the particle
are called spatial coordinates. The vector joining the points P 0 and is the
displacement vector. This vector may be expressed as
=
(3)
In the Lagrangian description the motion (or flow) of the continuum is

expressed in terms of material coordinates by equations of the form
*/=/(0
(4)
These equations may be interpreted as a mapping of the initial

configuration into the current configuration. On the other hand, in the
Eulerian descriplion. the motion of the continuum is defined by means of
the inverse of eqns. (4) in terms of the spatial coordinates as
a, = F,(x t)
(5)
The Eulerian description may thus be viewed as one which provides a

tracing to its original position of the particle that now occupies the location
TRANSIENT DYNA MIC FLUIDSTRUCTURE INTERA CTION
259
(.,,.,, .v3). The necessary and sufficient condition for the inverse functions
(5) to exist is that the Jacobian determinant
J = \cx.\da\
(6)
should not vanish.

Since eqns. (4) and (5) are the inverses of" one another, any physical
property of the continuum that is expressed with respect to a specific
particle (Lagrangian description) may also be expressed with respect to the
particular location in space occupied by the particle (Eulerian description).
For example, if the Lagrangian description of the density is given by
= P(a r)
(7)
the Eulerian description is obtained by replacing a, in this equation by the

function given in eqn. (5). Thus, the Eulerian description of the density is
p = p[F,(xt),t]
= p*(xt)
(8)
where the symbol p* is used to emphasise that the functional form of the
Eulerian description is not necessarily the same as the Lagrangian form.
Both the Lagrangian and the Eulerian approaches present advantages
and drawbacks. A clear delineation of fluidfluid and fluidsolid interfaces
and wellresolved details of the flow are afforded by the Lagrangian
approach, but it is limited by its inability to cope easily with strong
distortions, which often characterise flows of interest. By contrast, in the
Eulerian formulation strong distortions can be handled with relative ease,
but generally at the expense of precise interface definition and resolution of
detail.
Because ofthe shortcomings of purely Lagrangian and purely Eulerian
approaches, techniques have been developed that succeed to a certain
extent in combining the best features of both the Lagrangian and Eulerian
approaches. One such technique is the A rbitrary Lagrangian Eulerian
(ALE) method. A LE treatments have been developed by Noh, 1 Trulio. 2
and Hirt, A msden and Cook 3 in finite difference formats. Dona, Fasoli
Stella and Giuliani 4 and Belytschko and Kennedy 5 have reported finite
element ALE methods. The ALE formulation treats the mesh as a reference
frame which may be moving with an arbitrary velocity w in the laboratory
system. Depending on the value of the velocity w, three basic viewpoints
may be listed:
(1 )
w = : The reference frame is fixed in space and this corresponds to

the Eulerian viewpoint in which the motion is described in terms of
260
(2)
(3)
J. DOEA
spatial coordinates. A particle is identified by the position (fixed in

space) it occupies at time /.
w = : Where is the particle velocity in the laboratory system
that is, from eqn. (3), V = X = u'. Here, the reference frame
moves in space at the same velocity as the particles and this
corresponds to the Lagrangian viewpoint in which the motion is
described in terms of material coordinates. A particle is identified
by its position vector a at time / = 0 in the initial configuration.
w : The reference frame moves in space at a velocity w which
is different both from the particle velocity and from zero. Such a
reference frame is called A rbitrary Lagrangian Eulerian, and any
point of it is identified by its instantaneous position vector (see
Fig. 2). The variables (, t) are called the mixed variables. It should
be emphasised that the position vector is a priori arbitrary and
consequently independent of the motion of the particles. In the
Arbitrary Lagrangian Eulerian description a particle is still
identified through its material coordinates, a in the initial
configuration of the continuum; however, this identification
process is indirect and takes place through the mixed position
vector, X, which is linked to the material variables (a, t) by the law
of motion of the reference frame. This motion is expressed through
equations of the form
Ci=J*(al.a2,ai,t)
(9)
The ALE description may thus be viewed as a mapping ofthe initial

configuration of the continuum into the current configuration of
the reference frame. By analogy with eqn. (6) one defines the
Jacobian
J=\cL,\caj\
(10)
for the transformation between mixed and material coordinates.

Kinematics in the ALE Description
Since the ALE description does not refer directly to the particles in motion,
it will be useful to describe the flow of a continuum only if it can be linked to
either of the classical Lagrangian or Eulerian descriptions. How this can be
accomplished will be briefly illustrated in this paragraph. References 6 and 7
should be consulted for a more comprehensive treatment.
Let us first consider the determination of the time rate of change of a
physical property g associated with a particle characterised by its initial
261
position vector, a. Such a time rate of change is called material derivative

and will be denoted D/D/.
By analogy with eqn. (8), let g(a, t), g*(x,t) and g**(X,t) be the
Lagrangian, Eulerian and mixed representations of the quantity g:
If g is given in Lagrangian representation, its material derivative is simply
given by
Dg _ ig(a, Q
Di
t
;ii)
where | indicates that the material coordinates are held constant in taking
the derivative.
If g is given in Eulerian form, its material derivative contains two terms,
since the instantaneous position vector ofthe particle varies with time:
Dg_cg*(x,Q
Dt
t t
cg*(,
t)
(12)
'V
CX:
where , are the components ofthe particle velocity. The first term in eqn.
(12) is the spatial derivativethat is, with the point fixed in position. The
second term represents the contribution of movement and is called the
convective term.
Finally, if g is given in mixed representation, one may write, according to
eqn. (9),
g**(X. t) = g**(J*(H, t), t) = g(, )
(13)
By analogy with eqn. (12), the material derivative again contains two
terms and is given by
Dg =
D/
cg**(X,t)
cl
Bg**(C,t) V:
(14)
<W
Using eqn. (13), one may write

g(, /)
eg**( J)
tit
f I
Jg**(X,t) ,
<t
'w
(15)
Thus, noting that cCJct = ii,, we may rewrite the material derivative (14) in
the form
D g ^ cg(,/)|
Dr ~
et
+ O,
'F)
g**(X,t)
,
(16)
262
J. DONA
Expression (16) contains the partial derivatives of g** with respect to the
mixed coordinates ,; these can be related via eqn. (13) to the partial
derivatives of g with respect to the material coordinates a,. To this aim use
must be made of the mixed Jacobian (10) to write
(17)
Substitution of eqns. (17) into eqn. (16) enables one to express the material
derivative in mixed representation in terms of material variables only.
The Jacobian J provides a mathematical link between the current volume
element d V (function of the mixed variables) in the reference frame and the
volume element d F 0 (function of the material coordinates) in the initial
configuration:
dV=JdVQ
(18)
From relation ( 17) and making use of the derivation rule of a determinant,
one may easily show that the following identity holds:
~ = Jdivw
ct
By use of this identity, eqn. (15) can be rewritten as
eg
eg** , ,. , .,
ct
(19)
, 1 c J
( div (?**w) g**
cl
J cl
or equivalently as
~ (Jg) = j(CX^ + div (g**w)

et
\ ct
Now, considering the definitions
(20)
g*(x.t)=g*{J(z,t).t)
(21)
g**(X.t)=g**(f*(*,t).t)
we note that in the particular case where / = /*. the mixed representation,
g**. of the physical quantity g becomes identical with its Eulerian
representation, g*. Under this condition, we have
cg**(f*(.t),t)
(
cg*(J(ii.t).t)
( t
ii'
=0
(22)
263
so that eqn. (20) appears to be the fundamental relationship enabling us to

'translate'' any law expressed in terms oj spatial derivatives into an equivalent
Jaw expressed in mixed variables. This will be illustrated below for the basic
conservation laws of mass, momentum and energy.
Basic Conservation Laws in Differential Form
Using Cartesian tensor notation and neglecting viscous effects, we
express the conservation equations of mass, momentum and energy in
spatial (Eulerian) coordinates as
^ +^
8
^ + ^(pv,vj)
c t
=0
(23)
CXj
Ct
= pg,-^-
cj
( , = 1,2,3)
epe
c
c
- - + (pvj-e) = pvjgj - (pvj)
Ct
(24)
CXj
CXj
(25)
CXJ
In these expressions is the density,/) the pressure, the fluid velocity and g
the gravity. The total specific energy, e, is defined by
e = li- 2 +
(26)
where /' is the specific internal energy. For compressible fluids, the pressure
is defined by an equation of state of the form = p(p, i).
In an incompressible fluid the pressure is an intrinsic and independent
variable of the motion and is not related to any thermodynamic equation of
state; it is an implicit variable which instantaneously'adjusts itself' in such a
way that the incompressibility constraint, div v = 0, remains satisfied.
Introducing the Eulerian forms (23)-(25) of' the conservation laws into
the right-hand side of eqn. (20), we obtain the local form of these laws ina
moving reference frame:
^(pJ)
Ct
C
+ J^-(p(vJ-wJ))
=0
(27)
CXj
- (pv~J) + J ~ (pv,(vj - wj)) = J\pg,

ct
oxj
V
J-\
ex, J
(28)
-- (peJ) + J~(pe(vj
- Wj)) = J[pvjgj - 4~ (pvj))
(29)
J
J
ct
CXj
\
eXj
' J
In the above relations, which are of basic nature for the ALE formulation.
264
J. DOEA
the same symbols are used to represent both the Lagrangian and the mixed
functional forms of the physical quantities. Moreover, the mixed position
vector is written instead of X It should thus be reminded that the symbol
/t indicates a time derivative with the material coordinates being held
fixed. It is easily verified that the particular cases of purely Lagrangian and
purely Eulerian descriptions are contained in the above equations. Taking,
for example, the continuity equation (27), we note that
(1)
If w = v, it reduces to
TtM
(2)
= m(pJ) = 0
(30)
which is the typical Lagrangian form of the continuity equation.

If $ = o, eqn. (19) implies that cJ/ct = 0 and eqn. (15) states that
cp{, i)/ct = cp(x, l)/ct. In these conditions, eqn. (27) reduces to
the usual Eulerian form, eqn. (23).
Basic Conservation Laws in Integral Form

The statements of conservation of mass, momentum and energy
contained in eqns. (27)(29) are often more conveniently expressed
when integrated over a control volume V. This volume is bounded by a
surface S and may be moving with an arbitrary velocity w.
To express the basic conservation laws in such an integral form, let us first
consider the time rate of change of the integral
G(t)=
g**(X,t)dV
(31)
Jl'(l)
ofthe property g expressed in mixed coordinates. Such a time rate of change

is given by
dG(f) _ d
dt
g**dF
~Tt
1 (I
~g
c J)
J^< t + g
( ,dV0
j \ ^ + gdiv*}dV0
1 0
ct
where g(a, /) = g**(/*(a, /), /) (see eqn. 13).

Now, in view of eqn. (15), we may rewrite eqn. (32) in the form
<iG(t)
dt
> 0
eg*
t
+ w gradg** I g**divvv >JdV0
(33)
265
or, using eqn. (18) and the divergence theorem.

dGQ)
dt
I'm
Ct
dV + O
g**wdS
(34)
Mn
where n denotes the outward normal to the surface 5. The integral form of
the conservation laws for mass, momentum and energy are readily obtained
by replacing g** by p, pv and pe, respectively, in eqn. (34). By use of eqns.
(23)-(25) and applying the divergence theorem, the following conservation
statements are obtained:
dM _ d
dQ
d^
d7~d7
(mass)
pdV = C) p(w - v) n d S
~dT~d7
p v d F = ()
gradpdF
p$\v V)-dS
F(i)
Mo
V(l)
pgdF
Jl(l)
(momentum)
d_ d
d7"d7
pedV = O
no
(35)
Mii
I IM
pe(vi-$-dS
- O p\dS+\
su)
Mn
(36)
pv-gdF
Jfin
(total energy)
(37)
One may also write an integral expression governing the time rate of change
of the internal energy contained in the volume V(t). Such an expression is
obtained in the form
d/ _ d_
d7~d7
pidV =
V(i)
pdivvdF
p/'(w - ) d S
no
JS(r)
(internal energy)
(38)
When w = v, eqns. (35)(38) are Lagrangian and the time rate of change,
d/dt, coincides with the material derivative, D/Di. In fact, the material
derivative of C7(/) in eqn. (31) is given by eqn. (34), provided that the surface
S(t) displaces with the particle velocity V. Thus,
DG(t)
Dt
eg" dV + O g**v
} v u ) ct
MO
dS
(39)
g**(v w ) d 5
(40)
or, by use of eqn. (34),

DGO)
D/
dG(t)
+
dt
This implies that d/dt = D/Di when w = v.
266
J. DOEA
When w = 0, eqns. (35)-(38) are Eulerian and the time rate of" change,
d/dt, now represents the spatial derivative as defined in eqn. (12). In fact,
when w = 0, eqn. (34) reduces to
dGQ)
dt
eg**
-4 dF
(41)
ct
To achieve the discretisation of two-dimensional fluid domains, we shall

employ 4-noded quadrilateral elements with rectilinear sides and bilinear
local velocity field, as well as 3-noded triangular elements with linear
velocity field. The density and the specific internal energy (and consequently
the pressure) will be assumed uniform over each of these elements. It follow s
that direct use can be made ofthe integral forms (35) and (38) of mass and
internal energy conservation to update the element density and specific
internal energy, provided that V(i) represents the current volume of the
element. The situation is quite different as far as conservation of
momentum is concerned.
Since the velocity field in each element depends on six or eight parameters
(the nodal values of the velocity), the integral statement (36) is clearly
unable to furnish enough equations to update the element momentum. This
situation is overcome by formulating a variational statement associated
with the differential form (28) of the momentum equation.
Variational Form of Momentum Equation: The Principle of Virtual Power
The principle of virtual power6 is the basis for finite element models in
fluid mechanics. Its role corresponds to that of the principle of virtual
displacements in solid mechanics. This principle is derived by operating on
the differential form (28) of momentum conservation and on the associated
stress boundary relation
T,= -njCjj,
= S,
(42)
where T, represents the components of prescribed boundary loads (force

per unit area), iij are the direction cosines of the outward normal to the
boundary, c is the Kronecker delta and the fluid pressure.
In order to isolate the acceleration term in the differential form (28) of
momentum conservation, we rewrite this equation as
pJ-^
Ct
+ v,~(pj)
Ct
+ JipitOj
CXj
- Wj)) = j(pg, - -)
\
cx,J
(43)
and make use ofthe continuity requirement, eqn. (27). and ofthe property
div (B) = a div B 4- B grad a
(44)
TRANSIENT DYNA MIC F L U I D S T R U C T U R E
INTERA CTION
267
to reduce eqn. (43) to the simple form

CI
cp
ex
~ = p(w v) grad v + pg
ct
(45)
which will be used together with the boundary relation, eqn. (42), to
formulate the principle of" virtual power. To this aim, let us multiply eqn.
(45) by an arbitrary admissible variation, of the fluid velocity and
integrate it over the volume V(t). Similarly, we multiply eqn. (42) by v and
integrate it along the boundary surface S(t). We then add the two integral
equations so obtained; the result is
, zr d V = j
v,pg,dV
,(\ ) grad v d V +
Jl'IO
Jl'IO
ep
v,^JdV
+ O
v,(T,S,)dS
(46)
CX:
VO)
We now integrate by parts the pressure term in eqn. (46) and apply the
divergence theorem; this yields the principle of virtual power:
cv,
, dV
et
,( ) grad , d F 4
viPg,dV
VU)
( ;) d V +
m
, , dS
(47)
SU)
Variational Form of Continuity and Energy Equations

The integral expressions (35), (37) or (38) and (47) form the basis for A LE
finite element models in fluid dynamics, provided that the density as well as
the internal energy may be assumed uniform in each element. This is the
case for the simple elements for which detailed equations will be given
below.
However, prior to deriving the governing equations for such simple
elements, we shall briefly indicate the procedure to be followed if one wishes
to employ higherorder elements (e.g. 8node serendipity or 9node
biquadratic elements). In such elements the density and the internal energy
are not elementwise constant and the corresponding conservation laws
must be expressed in an appropriate variational form, in complete analogy
with what has been done in establishing the principle of virtual power.
Taking, for example, the continuity equation (27) and making use of
eqns. (19) and (44), one may write
cp
't
(w v) grad div
(48)
268
J. DOEA
Multiplying this equation by an arbitrary function and integrating over

V(t), we obtain
Cfl
^dV
1 (I)
\pp\\\dV
(\\ ) g r a d p d F
ct
(49)
no
Equation (49) is the weak variational form of the continuity equation which
should be used instead of eqn. (35) when dealing with variable density
elements.
In a similar fashion, the internal energy equation
j(piJ)
+ ^(PiOj
Ct
Wj)) =
Jp~
CXj
(50)
CXj
may be manipulated to give

di
+ ( iv) grad i = div
t
(51)
Multiplying this equation by the same arbitrary function and integrating

over the volume, V(t), we obtain
C' .
-
=
l IO
et
\\
( ) grad dV
(52)
VU)
IO
Equation (2.52) is the weak variational form ofthe internal energy equation
which replaces the integral statement in eqn. (38) when higher-order
elements are employed. In practical applications and /will be defined over
each element by
p =
1"JPJ]
(53)
where NjS are appropriate shape functions and the weighting function
would be = )V.
2.2. Computational Schemes and Governing Equations for Constantpressure Elements
In order to illustrate the application of ALE finite element procedures in
fluid dynamics, computational schemes based on explicit time integration
will be presented. A particular scheme valid for both compressible and
incompressible flows will be described in detail. The governing finite
element equations for 4-node quadrilaterals and 3-node triangles will then
be discussed.
269
Computational Schemes
As shown in the previous sections, the following conservation laws form
the basis of numerical solutions based on constant pressure elements:
dM
dt
d/_ d
d7"d7
d r
=
p d F = O p(w v ) d S
dt ,
s
(mass)
pdivvdF
pidV = p/(w v) n d S
(internal energy)
d_
dT ~ d7
pedV=
O pe(w v j d S
OpvdSf
,( ) grad , d F
Jr
(6,)
ex,
(55)
pgvdF
(total energy)
,dV
et
(54)
(56)
v,pg,dV
+ O ovTaS
j
(principle of virtual power)
(57)
The numerical solution of the above equations will be accomplished by

discretising the flow domain into quadrilateral and/or triangular finite
elements with straight line boundaries. Density, internal energy and
pressure are assumed uniform within each element. The fluid velocity is
approximated locally by
v=-XA/ J (.v,.v)v J
(58)
where N are appropriate shape functions and v, denotes the velocity vector
at node j . The coordinates (x.y) as well as the mesh velocity, w, are also
interpolated over the element by the above shape functions.
The finite element equations for mass and energy are obtained by
evaluating the integrals in eqns. (54) and (55) or (56) over each element,
using the local fluid and mesh velocities.
The global nodal equilibrium equations expressing momentum conservation are obtained by assembly of the element contributions. This is
done by applying the principle of virtual power, eqn. (57). in which the
velocity variation is defined as
v = Nj(x, y)dVj
(Galerkin process)
(59)
270
J. DOEA
This yields the following matrix system which approximates the momentum
equation:
cv
[M]<^}={F
l\
( I
+ \Fb] + >F\ + [F\
(60)
In these equations [M] is the global mass matrix, consisting of element

contributions of the type
N,pNjdV
Mj =
(61)
\F{\ indicates the global nodal loads induced by transport of momentum

components, with at the element level for node /.
N,p(vi ) grad vxV
(62)
N,p(w ) grad r dl
=
.
F h l denotes the global nodal loads due to body forces g:
Fl
N,pgxdV
n.i =
N,pgvdV
(63)
F p | are the global nodal loads due to the fluid pressure r.

CN:
p^rdV
),i
,
(.V
cN,
pr^dV
p.I
(64)
CY
[F] accounts for externally applied loads T:
Ff=
NTdS
(65)
' = o
NTdS
The main difficulty in solving the conservation equations for mass,

energy and momentum is due to the transport (or convective) terms. This is
271
particularly evident in the discrete momentum equations (60), which

describe a nonlinear system of ordinary differential equations.
Faced with the prospect of solving large nonlinear algebraic systems in
connection with implicit time integration techniques, it is highly attractive
to exchange the stability advantages of implicit methods for the
computational simplicity of explicit methods. In fact, the A LE finite
formulations reported in the literature are all based on explicit time
integration with lumped (diagonal) mass matrices.
Belytschko and Kennedy 5 employ an explicit central difference procedure
in which the velocities are updated by
v(t + {At) = v(t {A t) + A tv(t)
(66)
where v(t) results from eqns. (60), in which the righthand side is evaluated
at time t. This necessitates a forward extrapolation of the velocities of the
type
v(t) = v(i \A i) + \Atv(t A t)
(67)
so that the transport terms are for time t like the rest ofthe terms. The nodal
displacements, as well as the element mass and specific energy, are updated
by
A(t + A t) = A (t) + A tA (t +A I)
(68)
where a forward extrapolation of density and specific energy is needed for

the evaluation of A at time / f {A t.
Dona. et al.4 report an explicit scheme based on the Newmark/j
method. The nodal displacements are updated by
(1 + ) = () + Al(w(t)
+ y w"(t) J
(69)
Mass and energy are advanced by

A(t + At) = A(t) + AtA(t + iAt)
(70)
and, finally, the velocities are updated by

v(l + At) = v(t) + y [0) + v0 + ?)]
(71)
Here, again, suitable extrapolations are used for the evaluation of the
transport terms.
An interesting computational scheme based on explicit time integration
and valid for both compressible and incompressible flows can be obtained
272
J. DONA
following the ideas developed by Hirt et al.3 and Stein et al.9 in a finite
difference context. In deriving such a scheme, one notes that the convective
terms in eqns. (54)(57) vanish in the Lagrangian formulationthat is,
when w = v. It is this reduction that leads one to first solve the Lagrangian
integral equations in the first step of the calculational cycle, phase I, and
then to add the convective contributions as a separate step in the last phase,
phase III. The middle step, making up phase II, adds an implicit pressure
calculation that permits solutions to be obtained at all flow speeds.
In describing the above three phases of the calculational cycle, we shall
closely follow Stein et a/.9 and translate into finite element terms their finite
difference formulation.
Phase I: Explicit Lagrangian calculation. This step calculates the
Lagrangian nodal velocities resulting from the previous cycle's pressure and
body forces. From eqn. (60) with w = in eqn. (62) and using lumped
masses.
At
(72)
M[{Fil + {fil]
The tilde superscript indicates an intermediate time for this phase of cycle
+ \, and the superscript /; indicates previous cycle values.
The specific internal energy, i. is advanced to an intermediate time from
the integral equation (55) with w = v:
Ai
M".
p"div
f" +
dV
(73)
In this equation. A /L. is the element mass and a meanvelocity has been used,
as suggested by Pracht,' in order to conserve total energy exactly in the
finite element formulation.
Phase II: Implicit Lagrangian calculation. The basic objective of the
implicit Lagrangian phase is to obtain the new velocities that have been
accelerated by pressure forces at the advanced time level. This phase
eliminates the usual numerical stability condition that limits sound waves to
travel no further than one element per time step. This is accomplished by
iterating the tilde quantities from phase I so as to provide an advancedtime
set of" pressures for use in the momentum equations. Since the time
advanced pressures depend on the new densities and velocities, which, in
turn, are functions of the new pressures, these pressures are defined
implicitly and must be solved by iteration.
TRANSIENT D YNAMIC FLUID -STRUCTURE INTERACTION
273
Let the superscript L denote advanced Lagrangian time values, and let
the superscript n denote values at the beginning of a cycle.
The relation for density is the Lagrangian mass conservation relation:
At
(pL - p") + p'-divv = 0
(74)
The relation for pressure is

(75)
J(PLJ)
and velocity is advanced by

At
M'
\F(p'-)\
(76)
where pL = pL p".
Equations (74)-(76) are solved iteratively by use of a Newton-Raphsonlike scheme.
Following the convergence of the iteration, a final adjustment of the
'Lagrangian phase' internal energy is made by
At
M"
''
+ '
dV
(77)
As the last step in the implicit Lagrangian phase a quantity representing an

intermediate total energy is calculated for each element of the mesh:
(78)
zkOjYdV
Phase III: Rezone or convective-flux calculation. Phase I and II were

assumed Lagrangianthat is, the contributions from the convective terms
in eqns. (54)-(57) were not evaluated. In phase III we now add the
contributions from these terms to the time-advanced level L values.
We assume that the mesh velocities w have in some way been designated.
In the applications given in Section 4 an automatic mesh displacement
prescription algorithm is employed. Its description will be given in a
forthcoming paper. ' ' The change in volume of the element from time level
L to + 1 is

At O (w
-fL)dS
(79)
274
J. DONA
From eqn. (54) the updated element mass is

Mn+i
= Mn + / 0pHy
_ ^
Bd5
(80)
while the new density is given by

M'^l/V+l
p"+1 =
(81)
From eqn. (56) the contribution from the convective term to the total
specific energy yields
e+
^K
M'y
_AL_ O W ( W VL) dS
(82)
M"
From eqns. (60) and (62) the contribution from the convective term to the
velocity components at node / is
, " + 1 = F
J.I
J'
At
AT
N,pL(w \L)gradedV
(83)
From these timeadvanced velocities we can calculate the specific kinetic

energy; hence, the specific internal energy is
lar 1 ) 2 d F
(84)
Finally, the new pressure is calculated from the equation of state

p"+1 =/(p"+1,/"+1)
(85)
This completes the r teps contained in a cycle ofthe solution technique.

Details of Governing Equations for Constantpressure Elements
We shall now specialise the finite element equations given above to the
case of an axisymmetric quadrilateral element defined, as shown in Fig. 3.
by nodes i.j. k. I. The corresponding equations for a triangular element with
three nodes are derived in a similar manner.
Mass and energy equations. In a typical quadrilateral element with nodes
i.j.k.l ordered in anticlockwise order (Fig. 3), the fluid velocity is
approximated by
r_
= XA UC.,,W r } .
Iv. m\
(m =
i.j.k.l)
(86)
275
FIG. 3. 4Node isoparametric element.

where Nm are the shape functions expressed in terms of normalised
coordinates (, ) and (vrm, v.m) are the radial and axial velocity components
at node m. Referring to Fig. 3, the element shape functions are: 1 2
N, = i(\ 0 ( 1 );
Nk = i(i + 0 ( 1 + i ) ;
Nj = i(\ + 0 ( 1 I J )
N, = l(l 0 ( 1 + ?)
(87)
Assuming that the coordinates (r,z) and the mesh velocity w are also
interpolated over the element by the above shape functions, specific forms
are easily derived for the various terms in the conservation statements of
mass and energy.
The transport term in eqn. (54) for mass conservation has the following
explicit form for side ij of the axisymmetric quadrilateral element:
p(w v) dS = ^
{l(Wr 0(2,,. + Tj) + (ni v{)(r, + 2r)](Zj z,)
+ [<y= i0(2r ; + r.) + (wi vi)(r, + 2r.)](r,. r)\

(88)
276
J. DONA
For the mass equation to be stable, the density, pu, is evaluated as a

weighted average of the densities in the elements e and ec on either side of
the boundary segment ij:
Pu = i[V 0L,j)P' + (l + ; > c c ]
(89)
The coefficient a,; is chosen proportional to the velocity flux through side ij
and evaluated as suggested in reference 13.
As full donorelement weighting is too diffusive for most circumstances,
generally 0 < |,| < 1. Once the element mass has been updated, the
element contribution to the global diagonal mass matrix can be evaluated.
To this aim, the nodal masses M, are computed from the element mass M"
by
Nprdrdz
M,s=a.,M";
a =
(90)
rdrdz
Theweights 3^ are evaluated by use of a 2 2 Gaussian quadrature. Energy

transport in eqns. (55) and (56) is also expressed in the form given in eqns.
(88) and (89).
Momentum equations. The various nodal loads in momentum equation
(60) are evaluated at the element level and then assembled into the global
nodal load vectors.
The nodal loads (62) accounting for transport of momentum
components are evaluated by use of a 2 2 Gaussian quadrature. To
stabilise the momentum equations, it is well to introduce some selective
upwinding into the transport term. To this aim, Belytschko and Kennedy 5
replace eqn. (62) by
Ki = (1 + V>)Fu
(9 O
where
= O s i g n
cN
C
f(vrwr)
cr
cN
4(v:w:)}dV
cz
oc = 1 gives a full donor approximation;

ex = 0 corresponds to a centred approximation.
Because a greater proportion of donorelement weighting is required to
stabilise the mass equation than the momentum equations. ' 3 it is well to use
a different (smaller) in the momentum equations.
277
The nodal loads (63) induced by body acceleration, g, are evaluated at the
element level as
(92)
i:Hfc}
where M, is the nodal mass defined by eqn. (90).

Applying the Green-Gauss theorem and neglecting the virtual power due
to the discontinuities in pressure between elements, the nodal loads (64)
induced by the element pressure, p, are found to be, at node /',
F; , = () pN,rdz;
F:, = - O pN,rdr
Jit
(93)
Jit
These expressions are easily analytically integrated with the following

result:
F;.,. = P- [(r, - z,)(2r, + Tj) + (z, - z,)(2r, + r,)]
fp, = K'', - rj)(2r, + Tj) + (r, - r,)(2r, + r,)]
Artificial viscosity. An artificial viscosity is required to permit shocks to
form in the mesh. The artificial viscosity, q, used here is composed, as
suggested by Wilkins et al.,XA of a quadratic term and a linear term in the
divergence of fluid velocity. The quadratic portion is identical to the von
Neumann q for calculating shocks. ' 5 The linear term provides damping for
oscillations that occur behind the shock with the q method of calculations.
The following viscous pressure is thus added to the pressure obtained from
the equation of state at each time step before computing nodal forces:
q = p[C\l2 (div v)2 C,/adivv]
q=0
for div v < 0)

for divv > 0J
In this expression, / = ^JA, where A is the area of the element, a is the

dilatational wave speed and C'A, CL are numerical coefficients. According to
Wilkins et al.,1* CA % 2, CL = 0-8 for Lagrangian calculations.
A second type of viscosity which is needed is antihour-glass viscosity. The
hour-glass mode is shown in Fig. 4. It can be seen that this mode of
deformation ofthe quadrilateral element causes no volume change, and,
hence, no pressure. Thus, any excitation which induces this mode of
deformation is unresisted and elements can become quite distorted in these
278
J. DOEA
FIG. 4. Hourglass type of mesh instability.

modes. A finite element form of' Maenchen and Sack ' 6 rotational antihour
glass viscosity has been proposed by Belytschko and Kennedy. 5 Their work
should be consulted for the details of the formulation.
STRUCTURAL
ELEMENTS A ND COUPLING
FLUID A N D STRUCTURE
BETWEEN
3.1. Structural Elements

Because of the strong history dependence ofthe constitutive equations
and the complex shapes of solids, a Lagrangian mesh must be used in the
structural analysis. Despite its severe stability limitations, explicit time
integration is preferred, for no significant additional costs are introduced
by nonlinearities. The semidiscretised structural equations can be written
in the form
['
[M]l({Ftxi\
{Fm})
(96)
where { x | lists all nodal acceleration components, [M] represents the

global diagonal mass matrix, F,j are the externally applied nodal loads
and \Finl\ are the internal nodal forces.
279
At the element level the internal nodal forces are given in terms of the
stresses by
!F,,r =
[B]{a\dV
(97)
Here [B] is the strain-nodal displacement matrix, {} the stress matrix and
F e the volume of the element.
In modelling the structure we use either continuum (triangular) elements
or structural (rectilinear beam, conical shell) elements based on the
Kirchhoff hypothesis. Non-linearities due to large linear and angular
displacements are treated by a corotational technique developed by
Belytschko and Hsieh. 1 ' This technique consists in formulating the
equations of motion in terms ofconvected coordinates that rotate but do
not deform with the elements. Such a procedure linearises the straindisplacement relationships (provided that the strains can be assumed small)
and simplifies the nodal force-stress relations (97) within the elements'
convected coordinates.
Elastic-plastic relations of the incremental type are used for material
description in connection with the von Mises criterion of yielding. 18
Isotopie and kinematic hardening models are easily implemented. Several
structural materials show a significant dependence ofthe stress-strain curve
on the strain rate. Halleux 19 recently developed a pointwise mathematical
model for treating this effect.
3.2. Fluid-Structure Coupling
Finite element algorithms having been used for analysing both the
hydrodynamic and the structural parts of the coupled dynamic problem, it
appears that the feedback loop which illustrates fluid-structure interaction
in Fig. 1 is only apparent. In fact, the equations of motion in fluid and solid
elements are all expressed in terms of nodal loads and are of exactly the
same form. The equations of motion governing the coupled system may
thus be solved simultaneously, so that the explicit feedback between fluid
and solid response disappears.
However, suitable conditions must be prescribed along the fluid-solid
interface expressing that relative sliding of the fluid and solid is permitted.
In order to illustrate the treatment along a fluid-solid interface, let us
consider a structural member embedded in a fluid which is allowed to slide
along both faces ofthe structure. As shown in Fig. 5, we place two nodes at
each point ofthe interface: one fluid node and one structural node. Since
the fluid is treated in the ALE formulation, the movement of the fluid mesh
280
J. DOEA
I
-ti-
^- s
C
2
Ir
"~^
Structural nodes
Fluid nodes
S = Structural element
F= Fluid element
FIG. 5. Sliding interfaces.

may be chosen completely independent of the movement of the fluid itself.
In particular, we may constrain the fluid nodes to remain contiguous to the
structural nodes, so that all nodes on the sliding interfaces remain
permanently aligned.
The following conditions must be prescribed at each point of the
interface:
(1)
(2)
(3)
The grid velocity of the fluid coincides with the velocity of the solid.
The normal velocity of the fluid coincides with the normal velocity
of the solid.
The tangential velocity of the fluid is unconstrained.
In order to specify the above conditions, a local coordinate system (/. n) is

set up for each point of the interface (Fig. 5), so that / is the tangent to the
sliding interface and is 90 clockwise from /. Whenever acorner occurs in
the interface, / is the average of the two tangent directions.
Referring to the node numbering in Fig. 5, the condition under point (1)
implies that
w, = v,
and
v.
(98)
where w indicates the grid velocity ofthe fluid and the velocity ofthe solid.
These conditions are imposed directly at each time step and serve, together
281
with similar conditions at Lagrangian fluid nodes, as 'boundary conditions'

for the automatic fluid mesh displacement algorithm.
The conditions under points (2) and (3) are enforced through the nodal
forces. The condition of common normal velocity implies that
v"3 = v'l
and
" = v"2
(99)
If the solid element is based on the Kirchhoff hypothesis, we have the

further equality
v'l = v"2
( 100)
so that
V"3
= ", = v"2 = v"4
(loi)
Hence, the four nodes must have a common normal acceleration A " which
is obtained from the equation of motion
(M 4 M(l + M(2)A "
= F"
(102)
where M s is the mass contributed by the structural elements adjacent to

nodes 1,2, while Mn and Mf2 designate the masses contributed by the fluid
elements on either side ofthe interface. The normal load F" is obtained by
transformation into the (t, n) system ofthe assembled internal nodal forces
(F*, P) contributed by the relevant fluid and solid elements:
P' = (FJ + Ff1 + F o c o s a + (Ps + F}\ + F] 2 )sina
(103)
Here a is the angle between the and directions, as shown in Fig. 5.

According to condition (3), the tangential velocities are unconstrained
and result from the equations of motion
M\ = Msv'2 = Ps
M,xv'2 = F,)
>
M,2v\=F\2\
(solid nodes)
.
(fluid nodes)
(104)
where the tangential loads are obtained by transformation, into the (t,n)
system of the assembled internal nodal forces (F*, P):
F = F"* sin 4 Focosa
F'n = Ff sina + F], cosa
(105)
F2 = F,2 sin a + P{2 cos a

The only case where the A LE finite element formulation does not allow
the simple treatment of sliding just described isin the vicinity of a fluid free
282
J. DOEA
surface which must obviously be Lagrangian. As a consequence, fluid and

solid nodes cannot be kept aligned near the free surface and special
algorithms, such as those developed by Wilkins 20 for Lagrangian finite
difference meshes, must be employed to treat that portion of the sliding
interface.
4.
APPLICATIONS
The computer code EURDYN-1M developed at JRC lspra makes use of

the finite element methodology described in this chapter to evaluate the
response of coupled fluid-structure configurations to transient dynamic
loading.
Two examples of application of EURDYN-1M to fast reactor safety
problems will be briefly illustrated here.
4.1. Hexagonal Tube under Internal Pressure Pulse
A long thin hexagonal tube filled with water is subjected to an internal
triangular pressure pulse with axial symmetry. The peak pressure is 150 bar
and the rise/decay times are of 200 ms. A situation of plane strain is
assumed, so that, owing to symmetry, the analysis may be limited to a 30
sector of the cross-section of the tube-water system. This sector has been
divided into 47 elements as indicated in Fig. 6. The hexagonal tube is
FIG. 6. Symmetry sector of water-filled hexagonal tube.
283
represented by 8 rectilinear beam elements; the tube material is

elastic-plastic and described by a trilinear stress-strain law. The water is
assumed inviscid and compressible and has been modelled by constantpressure triangular elements. Sliding of water along the tube wall is
permitted and treated as indicated in Section 3.2. The three externally
loaded fluid nodes are purely Lagrangian, while the other fluid nodes are
treated as Arbitrary Lagrangian Eulerian and are automatically moved by
the computer program.
Figure 7 (a)-(c) enables us to follow the time history of
(a)
(b)
(c)
The displacement ofthe externally loaded node on the horizontal

line of symmetry.
The water pressure in the element in contact with the tube on the
horizontal line of symmetry.
The longitudinal strain at the external surface of the tube on the
4.2. Response of a Primary Containment Model

A thin cylindrical vessel with hemispherical bottom is nearly completely
oso
wo
ISO
2.00
2 50
TIME is
100
3 50
i 00
i 50
5 00
;o'4j
FIG. 7(a). Hexagonal tube, displacement of loaded node on horizontal line of

symmetry.
284
J. DOEA
150
FIG. 7(b).
350
Hexagonal tube, water pressure in the element in contact with the tube
on horizontal line of symmetry.
050
FIG. 7(C).
200
250
300
TIMEIS IO'1)
100
150
200
250
300
350
(00
i 50
500
Hexagonal tube, longitudinal strain at external surface ofthe tube on

285
filled with water and impulsively loaded by the detonation of an explosive

charge located on the vessel axis. The top ofthe vessel is clamped on a rigid
cover.
The finite element model for this problem is shown in Fig. 8 ; it consists of
23 conical shell elements and 278 fluid elements. The vessel material is
elastic-plastic and described by a trilinear stress-strain law. The water is
FIG. 8.
Primary containment model: initial finite element mesh.
inviscid. compressible and modelled by constant-pressure quadrilateral

elements. The charge is a low-density explosive of the type described by
Cameron, Hoskin and Lancefield, 21 and it is modelled by both triangular
and quadrilateral elements. The interface between the explosive charge and
the water as well as the water free surface are taken as purely Lagrangian.
The rest ofthe hydrodynamic domain is treated in the ALE formulation,
and the corresponding nodes are automatically moved by the program so as
to maintain the computational mesh as regular as possible. Sliding of water
along the shell is treated as indicated in Section 3.2, except near the free
""I
- 1
u-iJ-_LL
>m~uL
<CKVM-
>sxSt
s
lJ
FIG. 9. Primary containment model, (a) Deformed mesh at / = 0-5 ms; (b) deformed mesh at/ = 10 ms;(c) deformed mesh at t
= l-5ms; (d) deformed mesh at = 2-0ms; (e) deformed mesh at = 4-0ms.
>
S
y
H
O
<
>
2
r
5d
H
C
je
rr
H
rr.
7:
>
H
C
2
FIG. 9contd.
rO
oc
288
J. DOEA
0 00
0 250.50 0751001251 50 1 75 2 00 2 252 502753.003 253 503.754 0 0

Time ( s x l O
FIG. 10(a). Containment model, water pressure in the element in contact with the
shell on axis of symmetry.
surface, where a special procedure is employed to deal with nodes which
are not aligned.
Figures 9(a)-(e) show the deformed configuration of the computational
mesh at times t = 0-5, 10, 1-5, 2 0 and 4 0 ms. An important increase of the
charge volume, the progressive impact of water on the containment cover
and the successive deformation ofthe vessel are clearly seen in the figures.
One also notes that the automatic mesh displacement prescription
algorithm performs quite well.
Figure 10 enables us to follow the time history of
(a)
(b)
The water pressure in the element in contact with the shell on the
axis of symmetry (negative pressures have been cut off, since water
cannot sustain tension).
The meridional strain at the external surface ofthe shell on the axis
of symmetry.
289
0.25 0.500.751.001.251 501.70 2.002 25 2 502.753.003.253 503.754.00

Time ( s x 1 0 - 3 )
FIG. 10b. Containment model, meridional strain at external surface ofthe shell on
axis of symmetry.
ACKNOWLEDGEMENTS
The author wishes to thank Mrs P. Fasoli-Stella and Messrs S.Giuliani,
J. P. Halleux and A.V.Jones for their contributions in developing the
EURDYN-1M code. Appreciation is extended to M r G . vanGoethem for
useful discussions.
REFERENCES
1. Noh, W. F. (1964). In Methods in Computational Physics, Vol. 3 (Ed. B. Alder
et ai), p. 117, Academic Press, New York.
2. Trulio, J. G. (1966). Air Force Weapons Laboratory, AFWL-Tr-66-19, June.
3. Hirt, C. W., Amsden, . . and Cook, J. L. (1974). J. Comp. Phys., 14, 227.
290
J. DONA
4. Dona, J.. Fasoli-Stella, P. and G iuliani. S. (1977). Paper Bl/2, 4th SMIRT
Conference, San Francisco. California.
5. Belytschko, T. and Kennedy. J. M. (1978). Nucl. Eng. Design, 49, 17-38.
6. Truesdell, C. and Toupin. R. (1960). 'The classical field theories', in
Encyclopedia of Physics, Vol. III/l, [Springer- Verlag, Berlin.
7. van G oethem, G .. Doctoral Thesis. University of Louvain-la-Neuve. Belgium.
to be submitted.
8. Connor, J. J. and Brebbia. C. A. (1977). Finite Element Techniques For Fluid
Flow, Newnes-Butterworths, London.
9. Stein. L. R., G entry, R. A. and Hirt. C. W. (1977). Comp. Meth. Appi. Mech.
Eng.. I l , 57-74.
10. Pracht. W. E. (1975). J. Comp. Phys.. 17, 132-59.
11. Dona, J.. Fasoli-Stella. P.. G iuliani. S., Halleux. J. P. and Jones. A. V. 'An
Arbitrary Lagrangian Eulerian finite element method for transient dynamic
fluid structure interaction problems', paper submitted for presentation during
SM1RT-5 Conference. Berlin. August 13-17. 1979.
12. Zienkiewicz, O. C. (1971). The Finite Element Method in Engineering Science,
McGraw-Hill. London.
13. Amsdcn. A. A. and Hirt. C. W. (1973). Los Alamos Scientific Laboratory
Report LA-5100.
14. Wilkins, M. L.. Blum. R. E., Cronshagen. E. and Grantham. P. ( 1975). UCRL51574 Rev. 1. Lawrence Livermore Laboratory, May 30.
15. von Neumann. J. and Richtmyer. R. D. (1950). J. Appi. Phys.. 21, 232.
16. Maenchen.G .and Sack. S. (1964). In Methods in Computational Physics. Vol. 3
(Ed. B. Alder er al.), p. 181. Academic Press, New York.
17. Belytschko. T. and Hsieh, B. J. (1973). //;/. / . Num. Meth. Eng.. 1, 255-71.
18. Dona. J.. Giuliani. S. and Halleux. J. P. (1976). Nucl. Eng. Design,37,95-\ 14.
19. Albertini, C. and Halleux. J. P. (1978). 'Material behaviour and modelling in
transient dynamic situations". lspra Course on Advanced Structural Dynamics.
October 9-13.
20. Wilkins. M. ( 1964). In Methods in Computational Physics. Vol. 3 (Ed. B. Alder
ci al.), pp.211-63. Academic Press. New York.
21. Cameron. 1. G.. Hoskin, N. E. and Lancefield. M. J. (1977). Paper 2.. 4th
SMIRT Conference, San Francisco. California.
Coupled Fluid-Structural Dynamics in Blowdown

Suppression Systems: Numerical Schemes and
Applications
R. KRIEG
Kernforschungszentrum
Reaktorentwicklimg,
1.
Karlsruhe GmbH, Institut fr

Karlsruhe. West Germany
THE PROBLEM
In a postulated failure ofthe primary system of a nuclear reactor (i.e. in a

postulated blowdown) the escaping coolant must be collected within the
reactor containment. In order to prevent excessive containment pressures,
boiling water reactors usually include so-called pressure suppression
systems, where the primary steam will be condensed in cases of blowdown.
As an example, Fig. 1 shows the pressure suppression system of the
Brunsbiittel reactor, where the condensation chamber with its water pool is
integrated into the spherical containment. The outer wall of the
condensation chamber is part of the spherical containment shell. In case of
a blowdown, steam is blown through condensation tubes into the water
pool. Under certain conditions the condensation processat the tube ends is
highly transient. For instance, steam bubbles may be formed which grow
relatively slowly and then collapse in times of some 10 ms. Owing to these
phenomena, pressure transients in the water pool and oscillations of the
walls of the condensation chamber are generated which might affect the
integrity of the containment.
A thorough discussion of these problems and the application of several
engineering models to describe the transient condensation processes have
been published by Class, 1 who presents time histories for the bubble
collapse.
This chapter will concentrate on the dynamic response of the system. I n
particular, the vibrations ofthe spherical containment shell, which is one of
the most important safety barriers of nuclear reactors, will be investigated.
291
292
R. KRIEG
27 m
13 m
Spherical Steel Containment

2 cm Wall Thickness
7.5 m
FIG. 1.
60 section of the Brunsbttel containment with pressure suppression

system.
FLUID-STRUCTURAL DYNAMICS IN BLOWDOWN SUPPRESSION SYSTEMS
293
However, the solution of this structural dynamics problem requires

knowledge of the fluid dynamic loading. At the same time, the fluid
dynamics of the water pool largely depends on the transient deformation of
the part of the spherical shell wetted by the fluid and thus representing a
boundary part of the fluid. Hence, there is a coupled problem in fluid
dynamics and structural dynamics. More precisely, there will also be a
minor feedback to the condensation processes. However, since corresponding data are lacking, this coupling will be neglected. One of the time
histories by Class for the bubble collapse will be used as input instead.
To describe the fluid dynamics involved in the transient process, a
singularity method is used. It is applicable especially to highly transient
internal flow problems with any three-dimensional geometry including
walls wetted on both sides. The method is based primarily on a rectangular
dipole element (panel) for incompressible flow developed recently. The
elements form either the fluid boundary or an appropriate enveloping
surface. The results are used to obtain a linear relationship between the
pressures and the normal accelerations at the fluid boundary. In other
words, the three-dimensional fluid dynamics has been reduced to a twodimensional problem, with the unknown quantities occurring only at the
fluid boundary. This plus the closed-form character ofthe relationships is
an attractive prerequisite for handling problems in coupled fluid-structural
dynamics. In the interface region the unknown quantities at the fluid
boundary are identical with corresponding unknown quantities from
structural dynamics. Thus, these unknown quantities may be substituted
and equations of motion for the coupled problem are obtained which are
very similar to the equations of motion for structural dynamics without
coupling. Only the stiffness and mass matrices undergo certain modifications and become asymmetric, but the matrix dimensions remain the
same. (Additional transformation yields the so-called added mass matrix or
virtual mass matrix.) Consequently, for further treatment of the coupled
problems the same methods may be used as are known from dealing with
pure structural dynamics. In this chapter the eigenfrequencies and the
corresponding mode shapes (eigenvectors) are determined first; then the
problem is solved by modal superposition.
Since the fluid dynamic and the structural dynamic conditions are
satisfied in this chapter at the same time, the method is sometimes referred
to as 'simultaneous coupling'. Similar methods, however, with applications
to the completely different field of space vehicle design, were used by
Guyan. Ujihara and Welch, 2 Khabbaz 3 and Chung and Rush. 4 For the
fluid dynamics part also Khabbaz takes advantage of a singularity method
294
R. KRIEG
which is similar to that used in this chapter. Basic differences concern the
types of singularities (sources and dipoles, respectively) and their
arrangement (panels). A similar method of solution for fluid-structural
coupling in a blowdown suppression system was proposed by Schweiger
and Mayr. 5 To describe the inertial effect of the fluid, they also use a
singularity method, but modelling ofthe geometry is not carried through in
detail. A large number of contributions to the dynamics of pressure
suppression systems were presented at Sessions B5 and J6 of the 4th
International Conference on Structural Mechanics in Reactor Technology
held in San Francisco in August 1977. There was one paper where another
singularity method had been used. 6 More recently, too, Di Maggio. Bleich
and McCormick have dealt with the fluid-structure coupling aspects of
pressure suppression systems, but under axisymmetric conditions. 7 Class
tackles the same problem. 8 Finally, some preliminary results have been
published by the author. 9 , 1 0 A more detailed paper on the treatment of
coupled problems in fluid-structural dynamics, which also includes
additional references, is under preparation. 11
2.
DESCRIPTION OF THE STRUCTURAL DYNAMICS BY

MEANS OF STIFFNESS AND MASS MATRICES
In a first step we shall discuss the structural dynamics problem without

referring to coupled fluid dynamics.
The flexible structure may be approximately represented by a finite
number of points, each of them with one or more degrees of freedom (a
maximum of six degrees of freedom per point). All these degrees of freedom
may be numbered consecutively, n= 1.2
N; the corresponding
displacements may be the components ofthe vector s; or the corresponding
loads may be the components of the vector q. Then, with s as the second
time derivative of s, the equation of motion reads
As + Bs = q
(1)
where A and B are the stiffness and mass matrices. Both may be found by
standard methods of structural mechanics, such as finite element methods.
Usually. A and B are symmetric. With appropriate initial conditions, eqn.
( 1 ) provides unique solutions for structural displacements, s. if the loads, q,
are known.
For problems in coupled fluid-structural dynamics, however, parts ofthe
loads are due to the pressure in the fluid wetting the flexible structure. These
FLUIDSTRUCTURAL DYNA MICS IN BLOWDOWN SUPPRESSION SYSTEMS
295
loads cannot be given explicitly, but are results of fluid dynamics and
depend on the unknown structural displacements, which, in the wetted
region, are identical with boundary displacement of the fluid. In other
words, fluid dynamics provides a feedback of the unknown structural
displacements, s, on the loads, q.
Therefore, in a second step we shall distinguish between those structural
displacements and loads which at the same time describe the fluid dynamic
behaviour in parts ofthe fluid boundary and those structural displacements
and loads which are not directly related to the fluid. This will be done as
follows.
The degrees of freedom normal to the fluid structure interface and
describing those structural parts which are wetted by the fluid may be
numbered separately n = 1,2,
Nl. The corresponding displacements
may be the components of the vector s 1 . The corresponding loads per unit
area of the wetted structural parts may be the components of the vector q i .
This is designated structural region 1 (the coupling region).
All other degrees of freedom may be numbered n = 1 , 2 , . . . , '2. The
corresponding displacements may be the components of the vector s 2 . The
corresponding loads may be the components of the vector q 2 . This is
designated structural region 2.
Figure 2 characterises a coupled problem in fluidstructural dynamics,
where the structure consists of a thin shell. In this case vectors s1 and q1
describe the normal displacements and loads of the wetted shell region (Fig.
3a). The vectors s 2 and q 2 describe the tangential displacements and loads in
the wetted region as well as all displacements and loads in the unwetted
region (Fig. 3b).
Withs 1 a n d s 2 as the second time derivatives of s 1 a n d s 2 , the equations of
motion read:
A n s ' + A 12s2 + B n s ' + B 1 2 s 2 = q ' i
A 2 V + A 22s2 + B 2 l s ' + B 2 2 s 2 = q 2 j
Now. A 1 1 , . . . , B 2 2 are submatrices ofthe stiffness matrix. A, and the mass
matrix. B, of the whole structure. The integers Nl and A 72 denote the matrix
dimensions.
1
A=
\Ail\Ai2}\
N]
21 r " 2 2"
2;
A ! A
I
!J/V
296
FIG. 2. Coupled problem in fluid shell dynamics.

In eqns. (2) the loads q1 are due to the fluid dynamics and depend on the
structural displacements s 1 , which, at the same time, are displacements of
the fluid boundary. The loads q 2 will be assumed to be given.
3.
DESCRIPTION O F FLUID DYNAMICS BY A SINGULARITY

METHOD
In order to solve the coupled problem, the unknown load vector q1 in

structural dynamics equation (2) has to be replaced by an appropriate
relationship from fluid dynamics. To facilitate the solution process, the
only other unknown quantities in this relationship besides the vector q1
should be the structural displacement vector, s 1 , introduced before (or its
time derivative, s 1 ). This means that the three-dimensional fluid dynamics
problem must be reduced to a boundary problem. For this purpose, a
singularity method, also called boundary integral equation (BIE) method,
or panel method, will be used.
The whole boundary of the fluid region or an enveloping surface will be
formed approximately by plane rectangular panels with uniformly
distributed fluid dipoles, where the dipole directions are normal to the
c
y
j:
-1
H
7=
>
r
<
>
:
r
C
c/i
--3
-rr
:/:
v:
C
FIG. 3. (a) Structural degrees of freedom coupled with fluid dynamics, (b) Structural degrees of freedom independent of
fluid dynamics.
298
R. KRIEG
panels.t With fluid viscosity and compressibility neglected,J closedform

solutions for the flow fields a r o u n d the dipole p a n e l s c a n be o b t a i n e d . 1 K 1 2
F u r t h e r m o r e , introducing the total pressure (which is the sum of the static
and d y n a m i c pressure) and the fluid acceleration (which is the time
derivative of the velocity with respect to a given direction) as the u n k n o w n
quantities o f t h e flow field leads to solutions which are even linear functions
of the intensity of the uniform dipole distribution over the panel.
Consequently, the total pressure and the acceleration due to all of the
panels forming the fluid b o u n d a r y can be obtained by superposition of the
c o n t r i b u t i o n s due to particular panels.
In order to d e m o n s t r a t e this, let us define some field points with a given
reference direction in each of them and n u m b e r them consecutively: m
= 1,2,
M. T h e accelerations related to these directions m a y be the
c o m p o n e n t s o f t h e vector a. The corresponding total pressures m a y be the
c o m p o n e n t s of the vector p. N o w the principle of superposition yields
relations of the following type
CX = a +
DX = +
(3)
where, in the simplest case, each c o m p o n e n t of X describes the intensity of

just one dipole panel. Each element o f t h e matrices C and D then describes
the influence o f t h a t particular panel on the acceleration or pressure in one
ofthe field points specified. T h e values ofthe elements can be obtained from
the closedform solutions for the flow fields a r o u n d the dipole panels. This
t A dipole can be generated by two sources having the same intensity but opposite
signs, with a distance approaching zero. The direction ofthe distance becomes the
direction of the dipole.
The application of dipole panels provides some advantages in comparison with
source panels widely used in aerodynamics:
(1) A cross dipole panels a pressure step occurs which allows for simulation of
pressure differences across walls.
(2) A t the common edge of two dipole panels with different orientations no
leakage peak occurs.
(3) The coefficient matrices describing the superposition of the flow fields due to
the particular dipole panels have dominant diagonals, which facilitates the
numerical treatment.
As a disadvantage ofthe dipole panel against the source panel, the solution for the
flow field is somewhat more complex.
% Owing to the highly transient flow field with relatively small velocities, the only
essential effect of neglecting the fluid viscosity is vanishing fluid damping of the
system. A ccording to GUer, 15 neglecting the fluid compressibility is tolerable.
299
requires that the locations of each field point and the given direction in this
point must be described in local coordinate systems attached to each ofthe
panels. Usually these calculations require considerable effort. In more
complex cases a single component of X may describe the intensities of
several dipole panels, thus reducing the number of unknown quantities. On
the other hand, the dipole intensity of one panel may depend on several
components of X, thus reducing pressure discontinuity at the panel edges.
The weights of these components for a particular panel must begiven. The
same weights are necessary to determine the matrices C and D. The
components of the vectors and describe the influence of given fluid
sources, if present.
Now the reduction ofthe three-dimensional flow problem to a boundary
problem is easy. We specify appropriate boundary points as given field
points. The required reference directions in the field points are normal
vectors ofthe boundary. If the number of boundary points is the same as the
number of components of the vector X (in simple cases, the same as the
number of panels), the unknown vector X can be eliminated in eqn. (3),
leaving a linear relationship between the normal accelerations and the
pressures in the boundary points. There are no other unknown quantities of
the fluid field.
In the following, the above procedure will be implemented in detail.
However, analogous to structural dynamics, we will distinguish between
those regions of the fluid boundary which are formed by the flexible
structure and regions with prescribed normal accelerations and pressures,
respectively.
Points representing boundary regions wetting the flexible structure may
be numbered consecutively m = 1, 2 , . . . , M1. The corresponding normal
accelerations may be the components ofthe vector a 1 . The corresponding
pressures may be the components of the vector p 1 . This is designated fluid
boundary region 1 (the coupling region).
Points representing boundary regions with prescribed normal accele
rations may be numbered consecutively m = 1,2,..., M2. The given
normal accelerations may be the components of the vector a 2 . The
corresponding pressures may be the components of the vector p 2 . This is
designated fluid boundary region 2.
Finally, points representing boundary regions with prescribed pressures
may be numbered consecutively m = 1,2, ...,M3. The corresponding
normal accelerations may be the components of the vector a 3 . The given
pressures may be the components of the vector p 3 . This is designated fluid
boundary region 3.
300
R. KRIEG
Boundary
Region 3
a3
Boundary Region 2
a
Boundary
1
(given!
Region I
FIG. 4. Boundary regions ofthe fluid: coupled with shell (1). given acceleration
(2), free fluid surface (3).
A special case of fluid boundary region 2 is rigid walls, where the normal
accelerations a 2 vanish. A special case of fluid boundary region 3 is free fluid
surfaces, where the static pressures are constant. With usually low fluid
velocities in this boundary region, the dynamic pressure can be neglected
and, instead ofthe static pressure, the total pressure p 3 can be assumed to be
constant.
Again, the coupled problem of Fig. 2 is used as an example. The three
different regions ofthe fluid boundary are shown in Fig. 4. Region 1 wets the
flexible shell, region 2 is rigid or the normal acceleration is prescribed, and
region 3 is a free fluid surface.
With the accelerations and pressures introduced above, eqns. (3) split up
into
C'X = a' + y '
D ] X = p ! +'
C 2 X = a 2 4-y 2
>
(4)
D3X = p 3 + 3
As in eqn. (3), matrices C 1 , D 1 , C 2 , D 3 describe the influence of the dipole
panels (with intensities represented by X) on the accelerations and pressures
301
ofthe fluid boundary. Vectors 1 ,\2,3

describe the influence of given
fluid sources.
In eqns. (4) the third and fourth equations meet the boundary conditions
in regions 2 and 3. Corresponding equations with the pressure p 2 and the
acceleration a 3 are not necessary for the solution procedure.
In order to eliminate vector X, we introduce matrix F:
'
"
F =
D1
[M 1
C2
M2
D3
}/ 3
As mentioned above, the number ôf components of vector X must be the

same as the number Ml + M2 + M3 of specified boundary points:
= M1 + M2 + M3
Furthermore, as additional considerations show, det (F), in general, does
not disappear. Therefore, F can be inverted:
F* = ( F ) " 1
Then, with the splitting of F* into matrices D* 1 , C* 2 , D* 3
= F*
the following relation is obtained:

X = D* 1 (p 1 + 1) + C* 2 (a 2 + 2) + D* 3 (p 3 + 3)
(5)
With this result, vector X can be eliminated in the first equation of (4). A fter
some rearrangements one obtains
a1 ( C ^ D * 1 ^ 1 = y 1 + C'MD*1^1 + C* 2 (a 2 + y2) + D* 3 (p 3 + 3)]
(6)
Equation (6) is the required relationship from the fluid dynamics. The
302
R. KRIEG
unknown vectors a 1 and p 1 describe the normal accelerations and pressures

at those parts of the fluid boundary which wet the flexible structure. All
other quantities are known. Accelerations and pressures inside the fluid
region do not occur, i.e. the three-dimensional fluid dynamics problem has
been reduced to a boundary problem.
4.
C O U P L I N G BETWEEN FLUID DYNAMICS AND

STRUCTURAL DYNAMICS
Before the unknown load vector q1 is substituted in structural dynamics

equation (2) by using fluid dynamics equation (6). the equilibrium and
compatibility conditions at the fluid-structure interface must be
formulated.
We require that the degrees of freedom describing the wetted region of
the structure relate to geometric points which are identical with the points
representing the adjacent fluid boundary. (In cases where the structure is
modelled by thin shells, the distance of half the shell thickness between the
structural points located on the middle surface of the shell and the
corresponding fluid boundary points will be neglected.) In other words.
points m = 1,2
M1 must be identical with points n = 1.2
N1.
which means:
M1 = A'1
Provided that the positive normal accelerations of the fluid boundary are
directed inside the fluid region, the equilibrium and compatibility
conditions at the fluid-structural interface read:
p1 = - q 1
and
a1 = s 1
(7)
Here, analogous to free fluid surfaces, it has been assumed that the dynamic
pressure at the fluid-structure interface (coupling region) can be neglected
in comparison with characteristic pressures. Consequently, the static
pressure acting on the flexible structure could be replaced by the total
pressure p 1 .
Now in fluid dynamics equation (6) vectors a 1 and p 1 can be replaced by
the structural accelerations s and the loads q'. Then, with the resulting
relation, the unknown loads q1 in the structural dynamics equations (2) can
be eliminated.
303
Finally, the following equations of motion for the coupled problem are
obtained:
(C 1 D* 1 A 1 1 )s 1 r(C 1 D* 1 A 1 2 )s 2 4(C 1 D* 1 B u rE)s 1 +(C 1 D*'B 1 2 )s 2 "]
= y 1 4 C ^ D * 1 ^ 1 4 C* 2 (a 2 4 y2) + D* 3 (p 3 + 3)]
> (8)
A 2 1 s' 4A 2 2 s 2 + B 2 1 s 1 + B 2 2 s 2 = q 2
J
E is the identity matrix.
Equations (8) are of the same type as the structural dynamics equations
and the matrices governing the coupled problem have the same dimensions
as the stiffness and mass matrices for the structural dynamics. Only the
symmetry of the matrices disappears for coupled problems. Together with
appropriate initial conditions eqns. (8) yield unique solutions for the
structural displacements s1 and s 2 . Now the other unknown quantities
describing the coupled problem may easily be obtained. By use of solutions
s1 and s 2 , the first of eqns. (2) yields the load q1 of the wetted structural
parts. Then, with eqns. (7) and (5). the vector X can be calculated, which
allows for the determination of pressure and accelerations at any fluid point
by means of eqns. (3).
For further treatment of the problem, eqns. (8) may be rewritten as
Gs + Hs = r
(9)
with
s", :Y'
N2
Y2
.V'
^DÂ11
^DÂ12;}^1
G =
"A"
21
""
"""22"""jjyv2
N'
2
I ciD*iBn
2
+ E
; c'D^B'^jA"
2
!~y' + C ' t D * 1 ^ + C* 2 (a 2 + y2) 4 D* 3 (p 3 + 3)] 1} A1

'2
304
R. KRIEG
In the literature, problems in coupled fluid-structural dynamics are

sometimes replaced by problems in structural dynamics alone. In this case
the inertial effect ofthe fluid is simulated by introduction of so-called added
masses, which fictitiously increase the structural masses. Usually the added
masses are determined by simple approximative formulae. In this way the
effort involved in solving some coupled problems can be greatly reduced,
but at the expense of accuracy.
On the basisof eqns. (8), the'precise values' of the added mass matrix can
be found. Left-hand multiplication with (C 1 D* ' ) " 1 of the first equation in
(8) yields
A n s l + A 1 2 s 2 + [B 11 -f ( C ' D * 1 ) - 1 ^ 1 + B I 2 s 2 =
Comparison with the first equation in (2) shows that coupling with fluid
dynamics can be described by introduction of the added mass matrix
(^D*1)"1
Added masses are defined only for those degrees of freedom which describe
structural parts in contact with the surrounding fluid. The transfer of
prescribed forces or accelerations from the fluid to the structure as
described by the right-hand sides of eqns. (8) is not included in the added
mass concept.
5.
SOLUTION OF T H E EQUATIONS OF MOTION FOR THE

C O U P L E D PROBLEM
The equations of motion (9) are a coupled system of second-order

differential equations with respect to time. The number of equations agrees
with the number of unknown quantities. Several methods are available for
solution. The method of modal superposition will be used here.
A first step of calculation is
H ~ ' inverse of H
r = H1G
p = H 'r
Multiplying eqn. (9) by H _ 1 from the left-hand side yields
r s - F S = p'
(10)
305
A second step of calculation:

A
diagonal matrix of the eigenvalues of (the eigenvalues are the

elements of the diagonal matrix A );
matrix of the eigenvectors of (the eigenvectors are the columns

of the matrix );
inverse of ;
="1'.
Now, replacing the unknown vector s by the vector
s = Orj
and multiplying eqn. (10) by "
(11)
from the lefthand side yields
Aa + a =
(12)
Replacing the diagonal matrix A and the vectors and by their

components /., and p causes the matrix equation ( 12) to be split up into
one secondorder differential equation for each unknown quantity :
+ = p;
n=l,2,...,Nl
+ 2
(13)
In general, the righthand sides p are functions of time. They stem from
given accelerations a 2 , given fluid boundary pressures p 3 , given structural
loads q 2 and given fluid sources ( y F F y 2 , ^ 3 ) , which are also functions of
time.
Two basic cases will be investigated in some detail:
(1) Initially the system is at rest, i.e. a(t) and (t) disappear for / = 0.
The righthand side is pn = cosQr. is the frequency of excitation and
may be given.
Then the solution of eqn. (13) yields
(cosQi costu()
for
(14)
for =
with the eigenfrequency
t In structural dynamics without coupling the matrix is usually symmetric. In this

case the transposed may be used instead of the inverse " '.
306
R. KRIEG
If, on the righthand side, cos ; is replaced by sin it, the solution of eqn.
(13) reads
1
/ ~
sin /
sin
for
(15)
sinjt /cos,,/
for =
From eqns. (14) and (15) it follows that, in general, the solution is aperiodic.
Exceptions are cases where the /,, ratio assumes a rational number. For
excitations with a frequency , which agrees with the eigenfrequency OJ,
there is resonance, i.e. the amplitudes increase monotonically (at least for
long times).
(2) Systems conditions are known for the time T, i.e. a(F)and rjn(7")are
known.
The righthand side p disappears. Then the solution for / > F of eqn.
( 13) is
= A cos 1 + B sin ,,/
( 16)
with
B
= ( ) sin F + () cosa;,,
C
() cos OJ ,,() sinojF
With these basic solutions the response of the system due to excitations
occurring only within one time interval can be investigated. Let this time
interval be 0 < / < F a n d represent the righthand side pn(t) by a Fourier
expansion with the frequencies = 2/ , 2(2/ ). 3(2/7),... ; the system
response may then be obtained as follows: During the time interval
0 < / < Fthe solutions (14) and (15) for the different frequencies must be
multiplied by the corresponding Fourier coefficients and then super
imposed. For time / > T. formulae (16) apply with the values () and
(T) due to the response obtained at the end ofthe preceding time interval.
Once the components a(t) are known, the vectors s of the structural
displacements can be calculated by eqn. (11). In general, the structural
displacements will be an aperiodic function of time.
6.
A PPLIC A TION TO THE PRESSURE SUPPRESSION SYSTEM
The SINGS computer program has been written to implement the

operations described in the sections above. A steam bubble collapse in the
307
water pool of the pressure suppression system was assumed and SING-S
was used to compute the dynamic response ofthe spherical containment
shell, the coupling with the fluid dynamics of the adjacent water pool being
taken into account. Major input data are the stiffness and mass matrices of
the flexible structure.
In earlier work Gller investigated the structural dynamics of the
pressure suppression system of the Brunsbttel reactor by finite elements. 13
Coupling with the fluid dynamics of the water pool was not included.
Instead pressure distributions measured during steam relief tests 14 were
given. Comparing the calculated displacements with corresponding values
measured during the same tests. Gller found that structural modeling by
standard finite elements (STRUDL/DYNAL) was inadequate. Therefore
the spherical shell has been investigated by a semi-analytical method taking
into account about 2000 mode shapes, which is far beyond the computation
capacities of standard finite element programs. 9 , 1 5 Now the computations
agreed very well with the experimental data. Furthermore. Gller was able
to show that major deformations occur only in the loaded regions.
Consequently, for application ofthe SING-S program to the spherical
containment, the stiffness matrix was determined by the above semianalytical method. However, only that part ofthe spherical containment
was taken into account which at the same time constitutes the outer wall of
the condensation chamber. (At the upper and lower edges of that part the
shell was assumed to be simply supported.) The other parts of the spherical
containment could be neglected, since no dynamic loads act there and
according to Gllers' findingno major deformations are to be expected.
Also, the flexibility ofthe other walls ofthe condensation chamber had been
neglected, since investigation of those elements was not the primary goal of
the work. Furthermore, these flexibilities are relatively slight; hence, their
influence on the dynamics of the spherical containment was estimated to be
rather small.
For excitation of the system three simultaneously collapsing steam
bubbles were assumed to be located at the ends of condensation tubes at the
circumferential angles of 0, 120 and 240. Consequently, the investigations could be restricted to a 60 section of the containment shown in
Fig. 1. The initial bubble radius was assumed to be about 0-5 m. The bubble
collapse (i.e. the decrease of the bubble radius versus time) was modelled
according to Class 1 and is shown in Fig. 5. The collapse time is 0-1 s. The
times for bubble growing are much longer by comparison. Therefore,
dynamic effects due to this period are negligible.
In the SING-S program the decreasing bubble was simulated by a
308
R. KRIEG
5 02
0 04
FIG.
5.
0.10
0 06
Time (sec)
Steam bubble radius versus time during bubble collapse.
transient point source (sink). The dipole panels necessary to describe the
fluid dynamics of the water pool constituted the pool boundary as shown in
Fig. 1. The number of dipole panels was 696, while the number of the
components of X was 247.
The eigenfrequencies calculated with SING-S are
9-4, 13-0, 17-5. 17-8, 19-3, 21-4, 2F9, 24-4, 24-8, 25-0,... Hz
The mode shapes ofthe shell vibrations (coupled with the fluid dynamics of
the adjacent water pool) which belong to the four lowest eigenfrequencies
are shown in Fig. 6. For simplification only the wetted shell region is
plotted. The radial shell displacements versus time at six different points are
shown in Fig. 7 (location of these points. Fig. 1). It can be seen that after
bubble collapse (which takes 0 1 s ) the amplitudes of the radial
displacements are roughly constant. Decreases due to damping (material
damping in the shell, viscosity in the fluid) are not included. Comparing the
results for different points shows that with increasing distance from the
prescribed source (simulating the bubble collapse) a short time delay can be
observed. This is an indication that local disturbances propagate at a finite
velocity in coupled systems, although fluid compressibility is neglected, i.e.
the velocity of sound is finite.
The lowest eigenfrequency measured during the Brunsbiittel tests was
about 11-5 Hz, 1 4 which is some 20% above the calculated value of 9-4 Hz.
The reason for this discrepancy is supposed to be additional stiffening ofthe
lower containment section ofthe Brunsbiittel reactor. This stiffening, which
spreads over the three bottom rows of panels, was not taken into account
309
17-5 Hz
TWTTTT
I f - 1 "4
it I t
T_:_T#t
itti
\\\ V HTT
mi- xijffi
1 3 0 Hz
FIG. 6.
17-8 Hz
Vibration modes of the spherical shell coupled with the adjacent water
pool of the pressure suppression system (components of X: 247).
310
R. KRIEG
Boundary Point 5
Boundary Point 6
02
_.
E
E
04
06
08
10
02
02
04
Boundary
0.6
oint 2
VV
02
04
io
06
Boundary Point 3
Boundary Point 4
^ V
-3
-3
-=^
1.0
^v/u

04
06
Timels )
02
04
06
08
10
Boundary Pomi 1
02
04
06
08
io
Time ( s )
FIG. 7. Radial shell displacements versus time for points 1-6 shown in Fig. 1
(components of X: 247).
in the theoretical investigation. Therefore, in a second computer run. the
three bottom rows of panels were assumed to be rigid. Furthermore, the
number of components of X was increased to 394, thus allowing for a better
description ofthe mode shapes. Now the eigenfrequencies calculated with
SING-S are
12-5, 14-9. 17-9. 20-4. 21-5. 22-3, 22-8. 23-6. 25-0. 25
Hz
The mode shapes ofthe four lowest eigenfrequencies are shown in Fig. 8.
where both the wetted and the unwetted shell regions are plotted. Now the
discrepancy between the measured and the calculated lowest eigenfrequen
cies is about 10%, but with an opposite sign compared with the first
calculation. This means thai modelling of the stiffened containment section
as an almost, but not completely, rigid element seems to provide true values
for the eigenfrequencies. Other reasons for minor overestimates of the
eigenfrequencies are due to neglecting the flexibilities of the cylindrical and
conical walls of the condensation chamber and due to the finite number of
degrees of freedom in the theoretical model.
In order to study the influence of the coupled fluid dynamics of the water
12-5HZ
17-9 Hz
14-9 Hz
2 0 4 Hz
31 1
FIG. 8. Vibration modes of the spherical shell coupled with the adjacent water
pool of the pressure suppression system (components of X: 394).
pool on the structural dynamics, the above spherical shell was investigated
without this coupling. The corresponding eigenfrequencies are
520. 56-0. 58-8, 59-1, 59-5, 6 0 0 . 60-2. 60-9, 6 1 0 . 61-1.
Hz
The mode shapes of the four lowest eigenfrequencies are shown in Fig. 9.
Comparing the above results shows that coupling with the fluid dynamics of
312
R. KRIEG
mmE
mti T
T
ff
,|T T TT
T T
';'
T1
" 11-
TWU
*
T
^ -r
TTT TTT * *
T T T T TT
VMXV
UUIH
u\UUU
mWWW
T.
- * TT
, T TTT *
:
'
' T TTTTT *
4 * TT
T T
r *
!
*
;
m im
T
! *T T t T
WTULUUI'I * *
T
"
T IT "I*
TT
T
t T )Tlfi
t IT IT jrj
;3$0^
H $ t t t fr fri
+ttttw~.
WtL
TuutL
mmfw
520 Hz
588Hz
56 0 Hz
591 Hz
FIG. 9. Vibration modes of the spherical shell without coupling.
the adjacent water pool greatly reduces the eigenfrequencies ofthe spherical
shell. Furthermore, coupling completely changes the mode shapes.
Displacement peaks occur close to the pool bottom, where the influence of
inertia due to the fluid masses is stronger than close to the free fluid surface.
In the unwetted region the displacements disappear almost completely. (In
the figures, displacements lower than about 10 % ofthe maximum value are
313
not plotted.) Moreover, it can be noticed that in both cases, without and
with fluid-structure coupling, the eigenfrequencies are very close to each
other (while the corresponding vibration modes, in general, are entirely
different). According to Gller. 15 this is due to the local flexibility
behaviour ofthe system, where the displacements at a shell point essentially
depend only on the load at the same point and immediately around it.
7.
DETERMINATION OF THE DYNAMIC STRESSES A N D

C O N C L U D I N G REMARKS
On the basis of the transient displacement field calculated before, the

determination ofthe resulting stress fields is straightforward.
The remaining problem in assessing the integrity of the system is
primarily of a statistical nature. According to the experimental findings,
bubble sizes and collapse times (onset and duration of collapse) vary within
wide limits. Assuming that representative bubble sizes and collapse times
may be given (i.e. no feedback ofthe coupled fluid-structural dynamics on
the condensation processes), the resulting stress fields in the spherical
containment shell can be calculated by superimposing the fields due to the
single events.
However, if damping is neglected, as in SING-S, it should be emphasised
that for bubble collapse times with uniform probability distributions the
expectations for the maximum displacements and stresses increase
monotonically with the number of bubble collapses. Consequently,
assessments of the integrity of the containment must take into account
damping effects. The author believes that reliable modelling of this
damping should primarily be based on experiments, not on theoretical
investigations.
REFERENCES
Class. G. 'Theoretische Untersuchung der Druckpulsentstehung bei der
Dampfkondensation im Druckabbausystem von Siedewasserreaktoren-Rechenprogramm KONDAS'. Kernforschungszentrum Karlsruhe, KFK 2487, Okt.
1977.
Guyan, R. J., Ujihara, B. H. and Welch, P. W. 'Hydroelastic of axisymmetric
systems by a finite element method'. Proc. Conf. Matrix Methods in Structural
Dynamics, Dec. 1969.
314
R. KRIEG
3. Khabbaz, G. R. (1971). 'Dynamic behaviour of"liquids in elastic tanks'. A IA A

Jl, 9, Oct.
4. Chung, T. J. and Rush. R. H. (1976). 'Dynamically coupled motion of
surfacefluidshell systems'. J. A ppi. Mech.. Sept.
5. Schweiger. W. and Mayr, M. (1976). 'On the solution of acertain fluid structure
coupling problem using the boundary integral equation method". Mech. Res.
Comm., 3.
6. Liang. S. S. and Luk. C. H. "A new approach to suppression poo]
hydrodynamics', 4th Int. Conf. Struct. Mech. in Reactor Technology. A ug.
1977. San Francisco.
7. Di Maggio. F. L., Bleich. H. H. and McCormick. J. M. (1978). 'Dynamic
response ol' a containment vessel to fluid pressure pulses'. Comp. Sinici.. 8.
8. Class, G. 'Untersuchung der bei der Dampfkondensation im Druckabbau
system von Siedewasserreaktoren entstehenden Wandbelastungen". Kern
forschungszentrum Karlsruhe. KFK 2549. 1978.
9. Gller, B.. Hailfinger. G. and Krieg. R. 'Vibrations of the pressure suppression
system of a boiling water reactor'. Int. Conf. Vibration in Nucl. Plant, May
1978, Keswick. UK.
10. Krieg. R. (1978). 'Coupled problems in transient fluid and structural dynamics
in nuclear engineering. Pari 1: Safety problems solved by flow singularity
methods'. A ppi. Math. Modelling. 2.
11. Krieg. R.. Gller. . and Hailfinger. G. 'Transient, threedimensional potential
flow problems and dynamic response of the surrounding structures. Part I :
Description of the fluid dynamics by a singularity method: Part II: Simultaneous
coupling between fluid and structural dynamics' (to be published).
12. Krieg, R. 'Threedimensional flow fields caused by transient dipoles uniformly
distributed on rectangular plane elements". Kernforschungszentrum Karls
ruhe. KFKExt. 8/762. 1976.
13. Gller. .. Laursen. B.andSchlechtendahl.E.G. (1975).'Results of calculation
of the dynamic behavior of pressure suppression system during blowdown'.
Trans. Int. Seminar ELCA LA P, No. Q l / 3 , Berlin, Sept.
14. Kadlec. J. and Mller. R. A . (1975). 'Dynamic loading ofcontainment during
blowdown. Review of experimental data from Marviken and Brunsbiittel',
Trans. Int. Seminar ELCA LA P. No. Ql/2, Berlin. Sept.
15. Gller, . 'Schalendynamisches Verhalten des kugelfrmigen Containments
eines Siedewasserreaktors bei Dampfkondensation im Druckunterdrckungs
system'. Kernforschungszentrum Karlsruhe (to be published).
10
Computation of Flow-induced Vibrations in Piping
Systems
R . J. GlBERT
(CEA-DEMT-CENjSaclay,
1.
Gijsur-Yvette.
France)
INTRODUCTION
Internal flow induces vibrations in piping systems. The main sources of

these vibrations are located at the singular points of the circuit
(enlargements, bends, valves, etc.).
The unsteady flow field downstream from such singularities generates
both 'local' wall pressure fluctuations and acoustic waves (generally plane
waves). The latter are transmitted through the whole circuit.
In the first part of the present chapter the 'source problem' is studied:
local pressure fluctuations and associated acoustic sources are experimentally characterised for several current singularities. The second part is
devoted to the 'response problem' of a piping system to such sources: A
theoretical formulation for the modal characteristics and transfer function
computation of piping systems conveying fluid is described. The results of
an experimental test are compared with TEDEL computing code
calculation, which uses this formulation.
2.
PRESSURE FLUCTUATIONS INDUCED BY FLOW
Flow through industrial circuits is turbulent. However, the 'steady

turbulence" in straight tubes is quite different from some unsteadinesses
arising at singular points of the circuit (see Fig. 1).
The former is a small-scale phenomenon, the intensity of which is rather
low. The latter has larger correlation lengths and higher levels. Therefore its
mechanical effect on structures is much more important.
Singularities can be distinguished into two kinds: those with variable
315
316
R. J. GIBERT
SUDDEN ENLARGEMENT
DIFFUSER
GATE VALVE
JUNCTI ONS
FIG. 1. Typical flow singularities in industrial piping systems.

boundary conditions (pumps, blowers, etc.) and those with fixed boundary
conditions (fixed obstacles in flow, geometrical variations of pipes).
Many studies have been devoted to obstacles in flow as heat-exchanger
bundles and to pumps or blowers.
Pipe singularities as sudden enlargements, diffusers, valves. T junctions,
etc.. have been less studied (see refs. 6-9). However, their contribution to
the fluctuating pressure level in piping systems is important. Moreover,
their characteristics are perhaps simpler. A methodical study of these
singularities has been performed by the C.E.A., D epartement des Etudes
Mcaniques et Thermiques.
Experiments have been worked out on atmospheric air rigs and water
loops. Theoretical considerations have also been developed. The results
obtained have been collected in a formulary (see refs. 1, 2 and 5 for more
details of methods used). Some examples are presented in this paper.
2.1. The Two Effe cts of a Singularity in a Circuit
In a circuit including a singularity two zones can be distinguished where
the power spectral densities (PSD ) of wall pressure fluctuations are quite
diffrent: (I) Immediately downstream from the singularity the strongly
COMPUTATION OF FLOW-INDUCED VIBRATIONS IN PIPING SYSTEMS
317
perturbed flow field generates high-level pressure fluctuations with large

frequency range PSD. The correlation between two measurements
decreases with the distance between the sensors. (2) In the rest ofthe circuit
pressure PSD with acoustical resonance peaks are observed. Two
measurements are perfectly correlated, independently of the distance
between the sensors. The singularity, then, can be regarded as an acoustical
source which excites the whole circuit.
Both local and acoustical aspects of singularities must be characterised.
2.2. Analysis of the Local Wall Pressure Fluctuations
2.2.1. Mean Flow Parameter Analysis
Pressure fluctuations are characterised by a few statistical quantities
(such as mean square values and PSD). These quantities are governed by
simple laws, provided that they are reduced to some suitably chosen mean
flow parameters. These dimensionless laws depend very slightly on the
Reynolds and Mach numbers (in the range studied, 0-5 IO5 < (Re) <
2 6 and 0-01 < M < 0-5. no influence has been noted). They depend
only on the nature ofthe singularity, although they remain very similar as
from one singularity to another.
The main reducing parameters are: a mean pressure, AP. which is, in
some cases, the static pressure jump in the permanent flow across the
singularity; a mean velocity which generally is taken as the maximum
permanent flow velocity across the singularity; one or several characteristic
dimensions (an example is given in Fig. 2).
2.2.2. Characteristics of the Local Fluctuations
(a) The dimensionless mean square value, /, varies with the
distance from the singularity and presents a maximum of about 0-1
in the case of singularities involving a sudden separation of stream
lines.
(b) The dimensionless power spectral density of pressure fluctuations is
defined by:
&(s)
Sr( V) V
AP2 A
where = frequency, .( v) = measured PSD and (.v) is the

Strouhal number =(vA/V). ,<F(s) has a slowly decreasing shape
which can generally be represented by
&(s) =
1 + (.v/.vc)2
PERMANENT FLOW PARAMETERS
GEOMETRICAL PARAMETERS
Enlargement ratio: D/d

Reference length: D-d
Upstream cross-section: S,
Downstream cross-section : S2
Reference velocity: V
V - mean value of the flow velocity in the
upstream tube
Reference pressure: AP
AP = permanent pressure jump across the
enlargement
SPECTRAL QUANTITIES
LOCAL
ACOUSTICAL
FLUCTUATIONS
SOURCE
Strouhal number
(D - d)
s = PSD:.^(.v) =
y ( v)
() 2
1
PSD for pressure discontinuity: &&F(s) = Sr&P( v) 2
AP
F
D-d
/.,
f ..JAPI
|0|
Coherence, C -, U e x p - 0 - 8 5 -
Phase, [ ,
Al
I'
2
(F7
PSD for mass flow rate discontinuity:
Al
y
^ ( - 0 = ^?()
AP: distance between the intercorrelated sensors

0: angle between the intercorrelated sensors
cv
S, ; AP2
c: velocity of sound
Measurements have been performed for 0-5 10s < (Re) =
Vd
V
< 1-5 IO6 and 005 < - < 0-5.
FIG. 2. Sudden enlargement.
D-d
F
D-d

ZONE I
319
fis)
6
: ZONE OF MAX. FLUCTUATIONS

/(%)
1
10-'-=.
20
d/D = 0 5 ^ /^^**^
/V
d/Q=0-625
/ //'
d/D=0 71
i 2
_v(D-d)
L/(D-d)
1
10
t)-2
6 8 )-'
6 8 1
FIG. 3. Local pressure fluctuations.
(c)
where sc is the 'cutofT Strouhal number: ,vc ~ 0205 for the zone
of maximum fluctuations (see Fig. 3).
CrossPSDs are analysed in the form of coherence functions (ratio
between PSD modulus and the square root of the product of the
PSD of both correlated signals) and phase functions. The
coherence functions of pressure fluctuations present a plateau, the
level of which varies exponentially with the distance / between
both correlated sensors (c = ~ //; ).The correlation length,/has the
same order of magnitude as the characteristic length . A longitudinal
and transversal correlation length can be obtained according to
whether the correlated sensors are located parallel or transverse
to the mean flow. For example, the longitudinal correlation length
ofthe fluctuation downstream from a sudden enlargement is equal
to the diameter of the upstream tube. The 'longitudinal' phase
curves are straight lines. This shape is typical of a fluctuating
'transport' phenomenon.
The slope ofthe phase lines allows us to obtain the characteristic
velocity, v, of this phenomenon. The dimensionless value of t' is
equal to about v/V = 07.
Transverse fluctuations are in phase.
2.2.3. A coustical A nalysis

Theoretical consideration. The acoustical effect of a turbulent flow field
320
R. J. GIBERT
was shown by Lighthill several years ago. 3 He developed a theoretical

formulation available for threedimensional external problems. The intense
unsteady flow field downstream from a singularity has also an acoustical
effect but with internal and monodimensional characteristics:
The main part ofthe energy of local fluctuations is in the lowfrequency
range (below the cutoff Strouhal number sA . The comparison between the
acoustical wave length, A c, for the cutoff frequency vc( vc = scV/A ) and the
characteristic dimension, (A ), gives
scM
where M = mean flow Mach number.
As s < 05 and M ; 1 for the most current industrial circuits, then
/.JA : 1 : and the tube diameters of the circuit having the same order of
magnitude. vc is much lower than the acoustical cutoff frequency of tubes.
Consequently, only plane waves are transmitted through the circuit and
each singularity can be characterised by two spatial discontinuities: a
discontinuity of the acoustical pressure (averaged in a tube crosssection).
Ap(/)t and a discontinuity of the acoustical mass flow rate, A q(t).
Integral expressions of these functions have been written in ref. 2, using
the local fluctuations of pressure and velocity.
Ap(t) and A q(t) are random functions of time which are characterised by
their PSDs, r^(v) and
y^(v).
Acoustical source characteristics. Using the same mean flow parameters
as before, dimensionless PSDs &\,(s) and J*\H(s)% can be defined.
A dimensional analysis shows that 3F&p(s) and SP\H(s) can be written as
&)
= M2s2&(s)
FA/,(s) = F '(s)
^F(s) and F'(s) are dimensionless functions which have a 'turbulent'
spectrum shape. They only depend on the local perturbed field.
The following graph shows the shape of &^(s) and 3F\q(s). In the low
Strouhal number range the acoustical source associated with the singularity
can be well represented by a A p function.
Provided that M is not too small in comparison with / and for the high
t For singularities such as junctions of several tubes, several Ap functions need to
be defined.
&&.q(s) = (yAll(v)/Ap2)(c/S)1(V/A)
where c = sound velocity and 5 =
characteristic cross-section.
COMPUTATION OF FLOWINDUCED VIBRA TIONS IN PIPING SYSTEMS
321
'log
Vs)
2-W5>
log s
Strouhal number range, the acoustical source can be represented by a A q
function.
All these theoretical results have been experimentally verified. A method
has been developed in order to deduce the A p and A q functions from the
measurements of acoustical pressure at different locations in the circuit. In
particular, the acoustical transfer functions ofthe test circuit are given by
the computer code VIBRAPHONE. 4 Good isolation from other acoustical
sources (especially pump or blower) and welldefined boundary conditions
are needed. Figure 4 illustrates the results obtained in the case of a sudden
enlargement.
PRESSURE DISC ONTINUITY
MASS FLOW RATE DISC ONTINUITY
10-:
8
\a(si/M2
IO" 3
8
6
"
<~d/D=0 5
^
"N_
d/D = 062S
d/D = 0 71
d/D=0 75
"'
8
6
.v(Dd)
. v(D-d)
2
"'
4
_L
6 810-2
-L
8TJ-'
-io-'
IO" 2
6 8 10-1
FIG. 4. A coustical source of a sudden enlargement.
6 8 1
322
R. J. GIBERT
The main experimental results are given below.

(1 )
(2)
3.
For the A p function : its dimensionless mean square value /

generally is lower than the local (// 3 ) max. (001002). The cut
off Strouhal number of JÂ p(s) is also lower (001005).
For the A q function. 3F\H(s) presents a plateau. Its level varies
proportionally with M2. The maximum of F&q(s)jM2 is about 05
I0~ 2 10~ 2 for singularities generating a large perturbed zone.
RESPONSE OF PIPING SYSTEMS TO FLUCTUA TING
PRESSURE SOURCES
3.1. Introduction
3.1.1. Statement of the Problem
The third section of this chapter is devoted to the 'response problem' of a
piping system to the pressure fluctuations which have been analysed above.
Both local and acoustical fluctuations lead to fluctuating forces which act
on the walls ofthe circuit. The resulting vibrations may be evaluated for any
given piping system, by computing its mechanicalacoustical eigenmodes
and by using results of ref. 1. We present below a theoretical formulation
for the computation of the modal characteristics and of the transfer
functions of piping systems conveying fluid. Finally, an experimental test of
the whole method is given.
3.1.2. Modal Characteristics Calculation of Piping Systems Conveying
Highdensity Fluids
The effect of internal axial permanent flow on cylindrical structures has
already been analysed in the literature (e.g. ref. 10). Results are usually
presented by plotting the resonance frequencies in the complex plane as
functions of the flow velocity (V). The general shape of such plots is
sketched in the diagram given below.
For most nuclear reactor circuits the range of interest is usually limited to
the initial part of the curve (near V 0). Therefore the main additional
effect of the fluid flow is to damp the vibrations without changing
significantly their frequency. Moreover, in actual circuits the damping rate
due to this effect is often less than that caused by other effectsin
particular, imperfect connections between structures.
Consequently, it is a good approximation to perform a modal calculation
ofthe structuresteady fluid system and to use an estimate ofthe damping
rate derived from independent considerations (calculated or experimental
323
Im f)
increasing values
of velocity V
R e (Q)
| v=o
data). Furthermore, in the case of piping systems the mechanical-acoustical

calculation is simplified by using the fact that generally the acoustical
wavelengths of vibrations are very large with respect to the crossdimensions of the tubes.
3.2. Description of the Theoretical Formulation
3.2.1. Equations of the Problem
As the structure is concerned, the beam hypothesis (no strain ofthe tube
cross-section) may well be applied. Tube displacements are characterised by
a displacement vector o(.v, /)and a rotation vector a(.v. t) of the neutral fibre
and normal cross-section, which depend upon the curvilinear abscissa s.
In the simple case of a straight tube with open extremities and constant
cross-section, only the transverse component of t he displacement f5,(.v)acts
on the fluid. Since the acoustical wavelengths are large, compared with the
cross-dimensions, fluid moves with a uniform velocity <5,(.v) through the
whole cross-section S(.s).t Hence, the fluid effect can be taken into account
by adding the mass per unit length of fluid, p0S (p 0 = fluid density), for
transverse motions of the tube.
When the circuit presents geometrical singularities (namely bends,
changes in the cross-section, closed extremities, junctions between several
tubes), wall motions excite longitudinal plane waves (variations of the
acoustical mass flow rate are generated and located at the singular points of
the circuit). Conversely, a plane wave system, present in the circuit, induces
forces acting on the walls ofthe singularities. This plane wave effect must be
added to the transversal effect described above, and one is led to solve a
monodimensional acoustical-mechanical problem.
t That is not rigorously verified near the extremities ofthe tube. However, necessary
corrections remain small, provided that LjD 1 (L = length of the tube;
D = diameter of the cross-section).
324
R. J. GIBERT
In the Appendix an example is given in the simple case of a rigid bent tube
with open extremities and containing a quasi incompressible fluid. This
example shows the relative influence of the longitudinal and transversal
effects in fluid on the translation modal masses of the system.
Let R(s) be the local radius of curvature, dS/ds (s) the local variation rate
of the cross-section, and I and t the local unit vector system associated with
the osculatory plane ofthe neutral fibre. The local coupled equations can be
written as
KJ
+ MJ +
Sc2p
P o So,
dS
- -St + p \= 0
R
ds
at
(1)
dS-
[S^f--^^TTPcjS
+ Po - = 0
s\ es J
c2 dt2
(Re)
ds
A"mand M,,, represent the stiffness and mass operators ofthe tube in absence
of fluid; is the acoustical pressure averaged over the crosssection S(s);
and c is the velocity of sound.
From these equations simplified expressions may be easily derived for
several particular cases, such as a junction between tubes, a local variation
in the curvature or the crosssection of the tube.
3.2.2. Description of the Computer Code TED EL
Short description ofthe mechanical version of TED EL. TEDEL is designed
to compute threedimensional piping systems. The structure is modelled by
connecting straight beams or more complex elements (e.g. bends. T
junctions, etc.). Computations of several types may be performedin
particular, computation of the resonance frequencies and mode shapes.
and computation of the time response to any excitation.
The code uses the classical finite element method. The elements have two
nodes and there are six variables per node (three displacements and three
rotations). The form function which is used is a cubic polynom.
Modifications introduced for taking into account the fluid effects. To be
introduced into the TEDEL formalism, the Fourier transform of system (1)
must be written as
(K 2) = 0
(2)
Here ' and M are symmetrical stiffness and mass operators ( =

pulsation). For that purpose a supplementary variable. =/2,
is
needed. Hence, this leads to the consideration of eight variables per node
325
(three displacements, three rotations, and p). Discretisation of the

modified system (1), by the classical finite element method provides
elementary symmetrical stiffness and mass 16 16 matrices. These
matrices involve a 12 12 submatrix and a few additional terms related to
the fluid-structure interaction, as below.
Elementary stiffness matrix. Fluid does not provide any stiffness; then
the only additional terms are given by the discretisation of the equation
2 = 0
Elementary mass matrix. Three kinds of elements are distinguished:
Straight tube elements: additional terms are associated with the
transversal inertial effectf and with the longitudinal compressibility effect
of fluid.
Bend elements: terms due to the change of direction are added to the
preceding ones.
Elements with cross-section variations : terms due to the change of crosssection are added. Such elements can also represent a closed extremity anda
junction between several tubes.
In a further step elementary matrices are assembled in a classical way and
the boundary conditions are introduced. As the fluid is concerned, such
conditions may be a closed extremity of the circuit or a node, either for the
acoustical pressure or for the acoustical mass flow rate. The last operations
provide the final and M matrices of the whole system. The last step
consists in deriving resonance frequencies and mode shapes by using
classical algorithms.
Calculation oj the transfer junctions and characterisation of the sources.
Starting from the computed modal characteristics, the code allows one to
obtain any desired transfer function ofthe system, provided that modal
damping coefficients have been introduced into the code as user's data. To
perform such a computation, the nature ofthe source to be considered must
be known. The second section of this chapter shows that the main sources of
vibrations are the flow singularities present in the circuit.
t The transversal inertial coupling between coaxial tubes separated byfluidcan also
be taken into account. The corresponding terms ofthe mass matrix are obtained by
an analytical bidimensional calculation. Note that this coupling effect is very
important in the case of a narrow gap between tubes.
326
R. J. GIBERT
Local and acoustical effects of such singularities have been outlined above.
The local pressure fluctuations generate forces on the walls, (), of the
locally perturbed volume, ( V), which takes place next downstream from the
singularity. For the first resonance modes of the system the strains over ()
are small enough to be neglected. In addition, acoustical wavelengths are
also large in respect ofthe dimensions ofthe perturbed volume. Therefore,
by integrating the equation ofthe fluid motion, it can be shown that the
force field acting on () is equivalent to the force field acting on the cross
sections, (S), which separate ( V) from the rest ofthe circuit (see the diagram
below).
As the force field acting on (5,) depends only upon the acoustical pressure
averaged over (S,), the local excitation problem can be reduced to an
acoustical plane wave problem. Therefore, excitation effects on the whole
circuit, caused by a singularity, can be derived from the characteristics of
the acoustical source which is associated with the singularity. In the case of
low Mach number flows, such as liquid flows encountered in nuclear reactor
circuits, their acoustical sources may be characterised by a discontinuity of
the acoustical pressure (see Section 2).
The TEDEL program allows the calculation ofthe transfer functions
H(v,s) associated with a unit pressure discontinuity, located anywhere in
the circuit. Using such functions, it is also possible to compute the power
spectral density ,9"R of the response (displacement, stress, acceleration,
pressure) at any point of the piping system, to several decorrelated
singularities, according to the relation
ff,(v,i)|Vjv)
(3)
y \ (v) represents the power spectral density ofthe pressure discontinuity

associated with the singularity indexed by i. It can be estimated from the
results given in ref. 1.
4.
327
DESCRIPTION A N D RESULTS OF A N EXPERIMENTA L

TEST
4.1. Description of the Experiment

The piping system tested consists of a stainless steel tube circuit 2 mm
thick comprising four 90 bends, conveying water (diameter ofthe tubes is
D = 18 cm; curvature radius ofthe bends is R = 275 cm; RID = 154). The
tubes are joined together by very stiff flanges. The rig is supported by rubber
SK, , SK = rubber bellows
S T S S 3 = rubber extensible springs
joining flanges
accelero meter
fluctuating pressure transducers
3025 >
t_
rf~\
gate valve
FIG. 5. Configuration ofthe circuit.

extensible springs (steel mass of the whole is M s = 250 kg; fluid mass is
= 300 kg). This pipe is inserted in the 'Gascogne' water loop. Connections
are made by using rubber bellows which isolate mechanically the rig from
the rest of the circuit. In the same way two cavities with a compressed air
level are located upstream and downstream from the rig, providing an
acoustical isolation of it (see Fig. 5).
Vibrations are induced by a half-closed gate valve, inserted between the
downstream bellows and the downstream cavity. The level ofthe acoustical
source associated with the valve is much higher than that from any other
source in the whole circuit, at least in the low-frequency range, which is of
interest here (0-10 Hz). Pipe motions are measured by accelerometers, and
pressure fluctuations by wall pressure transducers (their location is given in
Fig. 5). Pressure measurements are used, in particular, for evaluating the
spectral characteristics of the acoustical source.
4.2. Modal Analysis of the Piping by TEDEL
The modelling takes into account the stiffness ofthe springs and that of
the bellows (which have been estimated by static tests), and also the effect of'
328
R. J. GIBERT
MODE 1
=382
m =112 Kg
battows
/ C > v . ^
max
^~\^^
/cT
upstream cavity
downstream cavity '
^\,
,
/c,
MODE 2
acoustical pressure field

v=5B8Hz
mr460Kg
FIG. 6. Computed modal shapes.

the flanges. The latter introduce a reduction ofthe flexibility ofthe bends
which it is worth emphasising. This effect has been calculated separately, by
using the threedimensional shell computer code TRICO.
The fluid boundary conditions are assumed to be fluctuating pressure
nodes, located at each cavity. TEDEL provides the resonance frequencies
and the mode shapes (displacement and pressure) of the system, in the
010 Hz frequency range. Mode shapes fall into two categories:
Parallel to the pipe plan.
v, = 382 Hz;
v2 = 5
;Hz;
v3 = 694Hz;
v4 = 94Hz
Figure 6 shows the shape ofthe first two modes. In particular, along the
straight portions ofthe circuit a linear evolution ofthe fluctuating pressure
is noted. This is a consequence of the large value of the acoustical
wavelengths in comparison with the length ofthe pipe. A discontinuity in
329
the slope of the pressure occurs at each bend; this is equivalent to a

discontinuity in the fluctuating mass flow rate.
The first mode corresponds to a swing motion of the bended part of the
pipe. For the next modes an important strain of the straight parts can be
noted. For all these lowfrequency modes stiffness of support springs and
bellows are revealed as important parameters.
Normal to the pipe plan. It is expected that such modes are not excited by
the source. Indeed, the level of such resonances is always very low in the
experimental data.
4.3. Acoustical Source Associated with the Valve
Tests performed with the valve in the fully opened position show that the
noise coming from the whole loop remains imperfectly softened by the
cavities (especially in the lowfrequency range of interest here). Moreover,
some pressure transducers are sensitive to local turbulence. Interpretation
ofthe results is much simpler when the valve is halfclosed, because in that
case the background due to disturbing sources and local turbulence
becomes much lower than the noise due to the valve. Indeed, the fluid mean
velocity is much higher at the valve location than anywhere in the whole
circuit.
Various levels ofthe valve gate have been tested, and for each level several
mass flow rates have been studied. The power spectral density of the
acoustical pressure discontinuity associated with the valve can be easily
obtained from the spectra ofthe pressure transducers which are located in
the straight tube, upstream from the valve: TEDEL shows that for the
present experiment the transfer functions of the pressure sensors do not
differ significantly from the pure acoustical transfer functions of the whole
circuit. In the 010 Hz interval these functions vary slowly with the
frequency. Then the pressure discontinuity spectrum J*^(v) is deduced
simply by dividing the experimental spectrum by the squared modulus of
the corresponding transfer function. This is done with the adimensional
quantities:
Strouhal number
adimensional power spectrum
xd
V
3F^VV(s)' ~=
tl
/22
(PoV'YD
where D = tube diameter and V = mean velocity ofthe flow at the valve. It
is worth emphasising that &\p(s) does not depend upon the mass flow rate
330
R. J. G I B E R T
and is only a function of the valve gate level. Consequently, p(s) proves to
be a specific quantity which characterises the singularity. The shape of
tFAp(s) is given in the diagram below.
<v>
10"
h/D=0.2
h/D = 0.
10 " 3
h/D = 0.4
10 - 4
10-3
10-2
10-1
These results are in good agreement with other measurements,

performed in various experimental conditions (especially in air or water test
loops) and for various singularities (see ref. 1).
PSD(gVHz)
7
8
9
Frequency Hz
10
3
6
7
8
9
Frequency Hz
10
3
6
7
8
Frequency Hz
9 '
FIG. 7. Comparison between experimental and computed results (gate aperture of

valve HjD = 04).
Experimental power spectral density;
computed power
spectral density.

,
6
7
8
9
Frequency Hz
6
7
8
9
Frequency Hz
331
PSD(g 2 /Hz)
6
7 ~8
Frequency Hz
FIG. 8. Comparison between experimental and computed results (gate aperture of

valve H/D = 0-21).
Experimental power spectral density,
computed
power spectral density.
4.4. Computation ofthe Response ofthe Circuit to the Source; Comparison
with Experiment
Experimental damping coefficients are about 10~ 2 . They are
introduced in TEDEL for calculating the transfer functions corresponding
to a unit pressure discontinuity, which is located at the valve position. Then
the spectrum of the source (see subsection 4.3) is used to derive the PSDs of
vibrations in the whole circuit. Figures 7 and 8 summarise the comparison
between the measured and computed spectra for three accelerometers and
two values of the valve gate opening.
CONCLUSION
There is good agreement between computation and experiment, in spite of
some small differences concerning the frequency ofthe resonances. This is
probably caused by an imperfect knowledge ofthe actual stiffness of flanged
bends and pressurised bellows. Nevertheless, it appears that a quantitative
estimation of piping vibrations can be obtained with a sufficient accuracy by
using the TEDEL program and the source data presented in ref. 1.
However, complementary studies are still needed to improve knowledge of
damping effects in piping systems conveying fluid.
332
R. J. GIBERT
REFERENCES
1. Gibert, R. J. (1976). 'Etude des fluctuations de pression dans les circuits

parcourus par des fluides', Note CEA No. 1925.
2. Gibert, R. J. (1974). 'Etudes exprimentales de deux singularits d'un circuit".
Note CEA No. 1735.
3. Lighthill, M. J. (1952, 1954). 'On sound generated aerodynamically'. Proc.
Roy. Soc, 211, 564-87; 224, 1-32.
4. Gibert, R. J., Livolant, M. and Baylac, G. (1971). 'Calcul de l'amplification
acoustique dans les circuits complexes', La Houille Blanche No. 5.
5. Gibert, R. J. and Villard. B. 'Sources acoustiques associes la runion de deux
coulements dans un circuit', Congrs G.A.S.E. 75, October 1975.
6. Morfey, C. L. 'Sound transmission and generation in ducts with flow'. Int. of
Sound and Vibration Research, University of Southampton, March 1970.
7. Fortier, A. 'Ecoulements instationnaires dans les circuits. Congrs A1RN.
1969.
8. Henry, R. (1971). 'Mise en vidence et analyses des instationnarits lastiques
dans les coulements internes", Thse, Paris VI.
9. Rouse, ., Asce, F. and Jezdinsky, V. (1966). "Fluctuations of pressure in
conduit expansions', J. Hydraulics Div. Proc. Am. Soc. Liv. Engrs.
10. Paidoussis, M. P. 'Vibrations of cylindrical structures induced by axial flow".
CEA/EDF Conference, Sept. 1972. Ed. Eyrolles.
APPENDIX: TRANSLATION MOD ES OF A RIGID BENT TUBE

WITH OPEN EXTREMITIES A N D C O N T A I N I N G A QUASIINCOMPRESSIBLE FLUID
T h e system studied includes two straight parts separated by a 90 bent part,
the length of which is negligible in c o m p a r i s o n with the length, /, of each
straight p a r t .
T h e system is s u p p o r t e d by springs at each extremity. Their stiffness
coefficient is K for b o t h symmetry directions Ox, 0>' of the system (see
d i a g r a m ) . (Only in-plane m o t i o n s are studied here.)
T w o translation m o d e s a p p e a r :
(1)
OX translation (<5J:
(a) T h e stiffness force acting on the tube, due t o the springs, is
F2 = - 2 *
(b) T h e inertial force due t o the tube itself is
Fn -
333
(c) The inertial force due to the fluid can be separated into two
parts:
(i) The force acting on the straight parts (due to the transverse
motions =( N /2/2) x ) projected on OX is:
-$
^2
p0Sl^
dx 3L_ =
pSlx
where S = tube cross-section and p0 = mean fluid density,

(ii) The force F i 3 acting on the bent part depends upon the
pressure p(l) ofthe fluid inside the bend. To determine this
334
R. J. GIBERT
force, the equation of fluid motions can be integrated in the

volume, (V), o the bend:
pido +
Hi
III
cp\ = 0
ct
The second term of thisequation is a volume term and can be neglected. The
surface, (), ofthe volume (V) includes the wall, (), of the tube and the
crosssections, (S) and (S2):
pnda +
(,)
pn2 da
i.v :)
pn do 0
i.S'|l
The first term is the force F, 3 ; the two other terms only depend on the
"plane wave' pressure p(l) in the circuit at the level of the bend:
p(l)S(n2n,)
F,3=
(A.l)
Symmetrically the motion of the bend generated a mass flow rate

discontinuity, A q, which can be deduced from the mass conservation
equation integrated in (V). The term due to the compressibility being
neglected, we obtain
,,^ = 0
Aq +
(/)
Using
0^ = 0
we have
</ = p0S<5(n2 - , )
(A.2)
This discontinuity, Aq. generates plane waves in the circuit. An acoustical

monodimensional calculation, taking into account the boundary con
ditions = 0 at each extremity ofthe tube, gives the 'plane wave' pressure
p(s) at any point of the tube. In particular.
11
P(l) = 2^ A 4 = :bcA(rl, - n , )
assuming that the wavelengths of fluctuation are large in comparison with /
(quasi-incompressible fluid hypothesis).
Using this expression of p(l) in eqn. (A.l). the projection of F l 3 on O.v is
given by
F3 = ~Pi)Sl'x
335
P(S)
P(D
21
Therefore the modal mass of the OA " translation is

Mox =
Fn + Fu + F,.
= Ms + 2pSl
The associated natural frequency is:

2K
Mn
(2)
2K
Ms + 2pSl
OF translation (y):
(a) The stiffness force acting on the tube is F r = 2Kdr
(b) The inertial force due to the tube itself is Fn =
MsSy
(c) The inertial force due to the fluid is
(i) force acting on the straight parts: F,2 = pSl.
(ii) force acting on the bent part: F i 3 = 0 (since (n2 n,)
= 0).
The modal mass ofthe 0y translation mode is:
M o v = Ms + pSI
The associated natural frequency is
2K
Ms + pSI
CONCLUSION
This example shows the mechanism ofthe coupling between the motions of
a circuit and the contained fluid. It particularly shows the importance of
336
R. J. GIBERT
taking into account the longitudinal aspect correctly, which strongly

depends on the acoustic boundary conditions ofthe circuit and then cannot
be evaluated by local considerations. In our example we note that certain
simplified hypotheses (fluid fixed to the wall of the tubes) lead to wrong
results even for calculating the inertial effect of fluid.
11
Analysis of Aircraft Impact Problems
N. J. KRUTZIK
Kraftwerk
Union A.G., Offenbach (Main), West Germany
1.
INTRODUCTION
The special load case of an aircraft crash is at present attributed a designgoverning importance for nuclear power plants built in particularly high
risk areas. The underlying principle is that the hazards presented by nuclear
power plants for the environment and the population should be no greater
than the hazards presented by other industrial installations. Building
structures and systems components should therefore be so dimensioned or,
if necessary, reinforced, that they remain functionally competent in the
event of such an occurrence, at least to the extent laid down in the relevant
safety specifications.
The relevant design specifications are approaching technical maturity
and are being increasingly incorporated into current engineering practice.
In view ofthe technical and economic background, the application of these
specifications requires the use of exact analysis methods which make it
possible to utilise to the limit the reserves inherent in the design and the
materials used. The relatively low probability of occurrence of an aircraft
crash also points in favour of this analytical approach. It follows that the
analyses formerly used assuming elastic conditions ofthe materials should
now be followed by methods which make it possible to take into account in
the dynamic analysis the effects of regional plastification of structures. The
extent of energy dissipation and the increase in the damping properties
which may be anticipated under these circumstances have a major effect on
the dynamic response ofthe building and of connected structures.
Starting from special anisotropic models describing the deformation
behaviour of reinforced concrete, the load functions specified for the
impact of a flying object can be modified so that they take into account the
plastic behaviour of the materials concerned.
337
338
N. J. KRUTZIK
Dynamic structural calculations based on the use of these modified load

functions lead to realistic results. In this chapter we shall compare the
dynamic reactions determined in this manner (time histories, response
spectra of the accelerations, section forces and moments, and displacements) with the results of calculations based on linear elastic assumptions.
The influence of parameter variations (model representation, soil
parameters, damping, etc.) on the results ofthe calculations is discussed.
The dynamic responses ofthe structures are compared with those of other
dynamic load cases (earthquake, external explosion), and various aspects of
the results are discussed.
1.1. Characterisation ofthe Load Case
The load case of an aircraft crash consists of an impact load of short
duration in a single arbitrary direction in space. Unlike the earthquake load
case, the shock is not transferred to the load-carrying structures via the
foundation slab but is applied at any arbitrary point of the building. The
most unfavourable location of this point is unknown and must first be
determined.
Depending on the point of impact and on the dynamic properties of the
structural regions near this point, one should therefore expect strong effects
of local resonance amplification. In view of this fact and ofthe magnitude
and duration of the load, one should as a rule expect local plastic flow as
well as an elasto-plastic material behaviour in adjacent regions (including
joints and floor slabs).
This results in a dissipation of energy and thus in a reduction in the global
thrust acting on the system in the transition zone from the plastified to the
still elastic region. It is this thrust and its temporal characteristics (which
depend on the local load-carrying behaviour) which constitute the actual
excitation or load function of the global structure.
The load functions so far stipulated by various authorities make no
allowance for the local plastification ofthe concrete. Instead, the impact is
assumed to act against an infinitely rigid wall. The load functions derived
under these assumptions are usually very conservative.
2.
LOAD FUNCTIONS
2.1. Survey
The current specifications relating to the load case of an aircraft crash are
characterised by the fact that they are not defined by the authorities
ANALYSIS OF AIRCRAFT IMPACT PROBLEMS
339
150r-
FIG. 1. Load time impact functions induced by different types of aeroplanes.

indirectly in terms ofthe type of aircraft, speed in flight, etc., but are laid
down directly in the form of a load-time diagram including a binding
specification of the impact area and of the angle of impact (Fig. 1). Such
data do. indeed, provide unambiguous load definitions, but these
definitions are only valid if one assumes that the concrete structure at the
point of impact is infinitely rigid. In view of the conservative nature of these
load data, and the consequences which this has not only for the
dimensioning of the load-carrying building structures, but also for the
design and the dimensioning of the mechanical equipment components
installed in the building, these load data require revision.
In order to define at least a provisional basis for the consideration of this
load case, a load function has also been defined in the Federal Republic of
340
N. J. KRUTZIK
Germany, and has in the meantime been specified as binding for locations
in Germany (Fig. IB).
This load function postulates the impact of an aircraft ofthe Phantom
RF-4 E type with a full-load mass of 200000 N and an impact velocity of
215m/s. The further stipulations are that the structure against which the
aircraft crashes is infinitely rigid, that the impact is normal to the surface
and that the load is applied to a circular region with an area of 7 m 2 . No
consideration is given to the building region, components, stiffness ofthe
region, or point of impact.
2.2. Load Functions with the Assumption of a Rigid Wall
All the load data relating to the impact of flying objects against concrete
walls go back to the work of Riera, 13 in which the behaviour of the flying
objects was considered only globally. In subsequent publications on this
subject the original assumptions postulated by Riera were altered and
improved.
The simplifications contained in these models lead as a rule to excessively
conservative load functions. Considerable improvements for determining
the properties of flying objects were introduced in the method used by the
Institut fr Reaktorsicherheit. 4 Using this method as an example, we shall
now briefly describe the procedures used so far (using a rigid wall).
2.2.1. Model Assumptions
The models are based on one-dimensional flying objects in which it is
assumed that all the parameters relating to the flying object can vary only in
the longitudinal (axial) direction. The flying object is represented by
stepped axially symmetrical cross-sectional elements of diameters equal to
the mean or the maximum diameters ofthe actual longitudinally variable
cross-sections of the flying object.
With such models, in conjunction with the assumption of a rigid wall, it is
possible to calculate only the total force or the mean pressure acting on a
certain area. It is assumed that the wall does not undergo any global
displacements during the impact and that the area where the impact takes
place does not undergo any local deformations. The longitudinal axis ofthe
flying object coincides with the direction of flight, and, at the moment ofthe
impact, it is perpendicular to the surface of the building structure.
The stress-strain and the force-deformation relationships vary during
the course of the impact. The structural and material deformation
behaviour is represented by a simplified force-deformation diagram. 4
which contains the material as well as the structural deformation
341
parameters. The load reduction rate is always equal to the load increase rate
(slope) in the elastic range.
The equations of motion describing the impact are solved by the finite
difference method (FDM).
The total force, F, acting on the building for the load case with
detachment of fragments is obtained by superposing the forces exerted by
the body ofthe aircraft (braking force) and those exerted by the impacting
engines (fragment impact loads).
2.2.2. Discussion of Known Load Functions
The impact load to be expected in the event of an aircraft crash depends
on the rigidity of the impact area and on the mass and velocity of the
aircraft. As increasingly heavier and faster aircraft are developed and put
into service, it may be anticipated that the load data will be repeatedly
changed.
Figure 1 presents a survey of load functions which were defined in the
years 1965-75. It can be seen that these functions differ considerably from
each other with respect not only to their force-time characteristics, but also
to the duration ofthe impact process. The shortest duration (007 s) is that
of the load function specified by the 1RS (Phantom RE-4 E). The load
functions put forward by Riera (Boeing 707 and 727) have durations longer
by a factor of about 4-5. The data for smaller aircraft such as the Starfighter
and the FB 111 are in the intermediate range. These differences result in
very large differences in the calculated forces acting on the building.
2.3. Load Functions with Elasto-plastic Deformation Behaviour in the Area
of Impact
The global load acting on the structure subjected to the impact of an aircraft
depends on the rigidity and the deformability both ofthe aircraft and ofthe
area of impact, as well as on the mass and impact velocity of the aircraft.
Very little information is available on the non-linear deformation
behaviour of reinforced concrete. This makes it necessary to use special
space methods of investigation by which the loads and the section forces
and moments in the non-linear range can be determined by means of special
anisotropic models for reinforced concrete based on the actual reinforcement distribution, the wall thickness and the properties of the concrete
itself.
It has been found by means of such studies that, in the case of outside
walls of a design with dimensions typical for nuclear power plants, the
plastic deformation of the wall results in a substantial absorption of energy.
342
N. J. KRUTZIK
AXISYMMETRIC STRUCTURES
D
RECTANGULAR STRUCTURES
D
FLAT 1
i PLATE I
PWR-STANDARD
SPHERICAL SHELL 1 8 0 m
FIG. 2.
BWR72
FLAT PLATE 1 8 0 m
BWR69
FLAT
PLATE I 50m
Modelling and reinforcement of impacted structural regions.
This may result ina reduction in the amplitude ofthe load-time function by
about 10%, and in a reduction of the global thrust acting on the
investigated reactor building by about 50 %.
2.3.1. Solution Method
In view ofthe large deformations ofthe investigated structures and ofthe
desirability of describing the impact process itself and the resulting
deformations and stresses by means of the same method, the method most
suitable for such investigation is the finite difference method (FD M). The
calculation operations were carried out by means of the program system
PISCES-1I D ELK, 1 2 which allows the use of mixed Eulerian Lagrangian
coordinates.
The investigations were carried out for geometrically undisturbed
regions of plane and curved outside walls with thicknesses of 1 5 m and of
1-8m commonly found in reactor buildings. In order to simplify the
problem and to reduce the cost ofthe calculations, the structural regions
investigated were assumed lo be axisymmetric (Fig. 2). The reinforcement
data were adopted from the actual drawings, with reinforcement densities
in the individual directions ranging from 0-5 to 2 %. The body ofthe aircraft
and the relevant region ofthe outside wall were subdivided into appropriate
zonal grids (Fig. 3). The deformability of the global structure at the
343
boundary of the local region considered was also taken into account. In
accordance with the resultsofa separate static calculation, the constants at
this boundary were so chosen that the deformations induced by a load of
llOOOMPa amounted to about 13mm. It was found that taking into
account the deformability of the global structure beyond this demarcation
boundary has almost no effect on the results of the calculations. This
deformability may therefore be neglected for practical purposes.
;so
100GRID PLOT
50
AIRCRAFT
00
50
00
50
10 0
15 0
FIG. 3. Modelling of the aircraft and the crash zone of the shell.
In order to make possible a direct comparison of our results with those
described in ref. 4, the same idealisation model for the body ofthe aircraft
was used.
2.3.2. Results
The events and the forces within the plastified or partially destroyed
region in the vicinity ofthe impact area are of no interest in investigations ofthe
dynamic response ofthe global system, provided we can define the forces
and moments acting on the global structure along the boundary of an
appropriately defined (e.g. circular) zone enclosing this plastified and partly
destroyed region. The governing load for the global system is the thrust
actually acting along this boundary, including the temporal characteristics
of this thrust. This boundary can easily be defined by following the
344
N. J. KRUTZIK
deformations and the strain conditions within the plastified and partly
destroyed region.
Deformations. The displacements of the deformable concrete structure
were calculated for the postulated impact with a load function according to
ref. 4 action on sections of a plane wall and of a spherical cap, respectively.
_
GRID PLOT
40-
20r
0 0
1
/ 77 / 7 7
/
I ! '\
NIH"
LrtTtS
M
!!
l|l|i|i|i|
!
''
'
0-0
'ili
50
FIG. 4. Deformation of the structure at F = 004 s. flat plate of 1 -5 m thickness.

The deformations of the zonal cells of the concrete and of the body of the
aircraft for the case of a plane slab are shown in Fig. 4.
Representative thrust. In both the investigated partial structures, the
transition zone between the plastic and the elastic states ofthe material is
found at a distance of about 7-3 m from the edge of the main impact area.
The time histories of the resultant thrust are shown in Figs. 5 and 6.
Comparing these results with the time histories calculated for the impact
against a rigid wall (Fig. IB), it can be seen that taking the plastic
deformations of the concrete into account results in a reduction in the
calculated peak thrust of" about 50%.
The time histories shown in Figs. 5 and 6 were used (in a smoothed form)
as excitation functions for the dynamic structural investigations discussed
below.
345
1 5" + 4
10"+4
5 0^+3
00
CO
0 02
0 04
0 06
/ (s)
FIG. 5.
Time history of shear force at 7-3 m, plate of 1-5 m thickness.
I 5if + 4 -
10" + 4
, 50F+3
00
0 0
002
0 04
0 06
/ (s)
FIG. 6.
Time history of shear force at 7-3 m, spherical shell of 1-8 m thickness.
346
N. J. KRUTZIK
2 O
't
'ho.. ^ -
V.
-M
05
10
20
30
40
50
60
70
80
90
100
MHZ]
FIG. 7.
Dynamic load factor for load function. 4
0 50r
040
030020-
\
\
o io009:
0 080 07:
0 06:
0 050 04-
Duration of load function Cd = 0 0 7 s
0 03-,7"
0 02-
001
'd
= 14-3 cps
10
20
FIG. 8.
30
40
FREQUENCY (Hz)
50
60
Time to maximum response.
70
347
ANALYSIS OF A I R C R A F T I M P A C T PROBLEMS
2.4. Analysis of the Load Functions

Analysing the load-time function according to ref. 4 for its frequency
content, the conclusion is reached that the structural regions likely to be
most strongly excited are those with eigenfrequencies of about 25-30 Hz.
and that the dynamic calculations must take into account eigenfrequencies
up to at least 75 Hz (Fig. 7). Global eigenfrequencies of typical building
structures are unlikely to be excited. In aircraft crash calculations the elastic
bedding ofthe building in the soil may not be neglected, but the effects of
embedment parameter variations are of minor importance.
It can be shown that the time elapsed up to the maximum response of all
structure regions with eigenfrequencies above 10 Hz is shorter than the
duration of the load function (0-07 s). For structures with low
eigenfrequencies the maximum response could be expected around 0-4 s
(Fig. 8).
Allowing the load-time function to act on two structural regions with
different stiffnesses (single mass oscillating systems, one with a high and one
with a low eigenfrequency) we obtain characteristic dynamic responses.
The time histories of the displacements (Fig. 9) show that the lowfrequency structure (2 Hz) responds relatively late (about 2 t A. The
amplitude ofthe displacement time history is small.
The high-frequency structure (50 Hz) responds simultaneously with the
load function and reproduces the applied load almost exactly (Fig. 10).
0 40
*_Q
4 = 2 (Hz)
7"s = 0 50 ( s )
-0 4 0 L
0 00 0 05
0 10 0 15
0 20 0 2 5
TIME
0 30
0 35
0 40 0 45
(s)
FIG. 9. Time history of displacement (= 2 Hz).
0 50
348
N. J. KRUTZIK
040
T-f~j
/s = 5 0 ( H z )
030
71 - 0 0 2 ( s )
0 20
O 10
JL.
-0 00
l l . . L - . . , l
-1
J.
0 45
0 50
- O 10
-O 20
- 0 30
-0 4 0 L
000
005
010
015
020
0 25
0 30
0 35
0 40
Time ( s )
Time history of displacement (f 50 Hz).
FIG. 10.
22
DLF
i=L
20
/ M-
>
AIRCRAFT C RASH
\\
V
\
/ // vi
- 1 '/ / ^
12
i o
08
06
04
0 2
00
I I
1
I
3
I
7
10
20
30
5 0 7 0 100
/[Hz]
FIG. 11. Dynamic load factor for different types of load functions.
349
Figure 11 shows the dynamic load factor (DLF) curve for various load
functions. It can be seen that the load functions based on an elastic material
behaviour result in smaller DLF values.
DYNAMIC STRUCTURAL INVESTIGATIONS FOR
CHARACTERISTIC BUILDINGS
For nuclear power plants located in regions with low earthquake intensities,
the design of load-carrying structures as well as of mechanical equipment
components may be governed by the load case of an aircraft crash. The
impact may take place at any point ofthe building. Starting from the point
of impact, there takes place a wavelike propagation of vibrations. In
monolithic buildings these vibrations are transmitted to the coupled
structures directly, whereas with separate structures they excite the internal
structures indirectly via the common foundation slab.
The response ofthe building structures to the impact load induced by an
aircraft crash always definitely depends on the point of impact and on the
assumed plane of action ofthe load. The first step in the investigation ofthe
consequences of this load case is therefore to find the governing points of
impact and to determine their topological arrangement in the building
structure as well as the direction of action ofthe impact load. This step may
be very laborious and time-consuming, and in some cases this problem
cannot be solved at all. If the coupling points of the most important
mechanical equipment components are not known in advance and/or if the
investigations of the effects of loads applied at several assumed points of
impact cannot be limited to only a few points ofthe structure, it becomes
necessary to carry out repeated analyses of the dynamic response of the
global system for various points of impact. This leads to extensive and costly
studies.
The level of stress of the system outside the area of impact is relatively
low. so that, as a rule, the global structure should occur outside the near
vicinity of the impact area; this could have a significant effect on the
response of the structure. In such a case it would be necessary to carry out a
non-linear investigation ofthe entire system. However, the state of stress
can bedefined and checked at any time by following the time histories ofthe
deformations. On the basis of state of strain of the elements in the local
region, it can also be definitely established whether partial or total
destruction is to be expected in the impact area.
350
N. J. KRUTZIK
3.1. Model Representation

The type of required idealisation ofthe structures to be analysed depends
on the aim ofthe investigation and on the information required. In many
cases the node distribution is largely determined not only by the range of
frequencies which must be studied (up to about 80 Hz), but also by the
number of locations (coupling points of substructures and of equipment
components) at which the information is required.
The dynamic response of the building structure to the impact of an
aircraft crash also depends decisively on the local dynamic behaviour ofthe
wall structures in the immediate impact area. This and the other relevant
conditions impose a degree of discretisation which, bearing in mind the size
ofthe structures and the resulting number of dynamic degrees of freedom,
may result in computation tasks near or beyond the capacity of even the
largest computer systems available at present. The borderline for the degree
of discretisation of the structure regions can be determined from the wall
thicknesses involved. For walls of box-like buildings it has been found 21
that for the usual wall thicknesses in the range of 1 -52 m the size of the
elements should be smaller than 10m 2 .
Particularly laborious are investigations of building structures which
must be idealised by means of three-dimensional analytical models (boxlike buildings). Characteristic of all the buildings considered is the fact that
internal load-carrying structures and the external walls stand on a common
foundation slab through which excitations are transmitted from the
external to the internal structures partly or wholly.
The elastic bedding in the soil is taken into account, starting from the
theory of the elastic half-space, by means of springs with equivalent
rigidities arranged at appropriate points under the foundation slab.
3.1.1. Axisymmetric Buildings
Most reactor buildings of boiling water type (BWR) and all reactor
buildings of pressurised water type (PWR) may be considered with good
approximation as axisymmetric structures. They usually consist of several
concentrically arranged cylinders connected and braced by floors (annular
slabs). These floor slabs may be connected with the outside shell
monolithically or they may be supported there only on vertical supports.
Idealisation by means of shell elements is the obvious choice. We choose
elements with a continuous mass distribution. Unsymmetrical structure
regions of additional loads are appropriately distributed over the
axisymmetrically idealised structures. The rigidity of the annexes is
neglected, whereas their masses are distributed over the corresponding
351
Node
1 Shell elements
1981 Mass elements
114 Springs
0
1
V
108 OV
'^ÔV o v
ov
~ov
O Harmonic
1 Harmonic
Vertical
Torsion
Horizontal
ov
FIG. 12. Idealised structure of a PWR building.

annular elements. The internal concrete structures are idealised by means of
equivalent cylinders. The safety containment is represented by a spherical
shell. The interaction ofthe soil and the building is taken into account in the
same manner as for the models described earlier. The mathematical
computation model is shown in Fig. 12.
The elastic bedding in the soil is represented by equivalent spring
stiffnesses arranged as concentric annular elements under the foundation
LO
70
IO
60
50
40
30
INNER CONCRETE
PRESSURE VESSEL
Biological
199
399
^shield
198
398
<>197
196 T 295
tl95
397
194
20
10
-10
?193 294
192
{>99 t 2 9 3
98
396
9 7 129
96
95
94
T 93
92
*039
STEAM GENERATOR
REACTOR BUILDING
FIG. 13.
Idealised structure of a rectangular reactor building.
r.
-
353
slab. The structures ofthe pressure suppression system, ofthe biological

shield and of the reactor pressure vessel are also represented by shell
elements.
3.1.2. Box-shaped Buildings
The structure of a box-shaped building consisting of vertical and
horizontal slabs is represented accurately by finite elements of equivalent
stiffness. The requisite degree of discretisation is determined by means of
preliminary investigations. The masses of the structures, including selfweight and live loads, are represented by an appropriate system of discrete
point masses. The internal structures, coupled to the external structures via
the foundation slab, can be represented by means of an equivalent
spring-mass system (Fig. 13). The coupling conditions and the elastic
bedding in the soil are represented by equivalent springs based on the
theory of the elastic half-space.
3.2. Inherent Behaviour
Some typical eigenfrequencies and the corresponding mode shapes are
shown for axisymmetric buildings in Fig. 14 and for box-shaped buildings
in Fig. 15. For the axisymmetric buildings the frequency range was
determined, starting from the load distribution and the frequency content
of the load function, for eleven harmonics up to a numerical value of the
eigen frequency of about 80 Hz. For box-shaped buildings all the local and
global eigenfrequencies are obtained simultaneously from the threedimensional model. The range of frequencies is analogous.
3.3. Modal Damping
The modal damping of the coupled oscillating structure was calculated
for each mode shape according to its energy participation by the following
formula:
D,ELn
D
F,.
where D, is the damping of the z'th substructure; E, is the deformation
energy of the /th substructure in the nth mode; and is the number of
substructures.
In conformance with the relevant specifications, the modal damping
ratios were limited to 15% in the horizontal direction and to 30% in the
354
N. J. KRUTZIK
^MTJ*^4-"
il
O. Harmonic,f, =1-92Hz
O. Harmonic,f 2 = 2-63 Hz
1 Harmonic,f, =1-33 Hz
1.Harmonic,f5=6-26Hz
1. Harmonic, t|7= 22-33 Hz
1. H a r m o n i c , f ^ 24-73 Hz
FIG. 14. Global and local mode shapes of a PWR building (standard).
vertical direction. The structural damping ratios were assumed as 2 % for
steel and 5 % for concrete. The proportions of soil damping were
determined according to the theory of the elastic half-space.
3.4. Selection of the Governing Points of Impact
Considering the shape and the behaviour ofthe building structures, it
may be expected that there are many (but not infinite) points of" impact
resulting in maximum response in certain structure regions. However, in
order to limit the extent of the calculations, it is often possible to specify
ANALYSIS OF A IRCRA FT IMPA CT PROBLEMS
355
îteliii^ 1 ' 1
.
c\,
' "1
" " Sf l & ^ S i t ^ i r ^ l

4. Modeshape.f= 0- 76 Hz
Tr-^ r - f u i r - i ^q
.iv T" r-'F.-rtrr-4-
76. Mode shape.f = 18-7 Hz

FIG. 15.
6.Modeshape.f = 1-85 Hz
- - - =U'^1^-X--i^.pv./ .'
l^-tf'-fcrffT '1 "
129Modeshape.f=30-5Hz
Global and local mode shapes of a BWR 69 building.
356
N. J. KRUTZIK
certain regions and directions of impact which cover all other regions and
directions within a practically acceptable range of scatter. On the other
hand, it is only in rare cases that it is possible, by means of parametric
studies, to define one single region of impact resulting in approximately
maximum responses at all the relevant locations.
In the studies of the box-shaped building it was proved possible, on the
basis of eleven case investigations, to find one impact region which was
finally used as the basis for'complete calculations and for obtaining relevant
data for detailed investigations of the coupled substructures. This impact
region corresponds to a corner in the top part ofthe building (Fig. 13). For
the axisymmetric building, however, it is necessary to investigate four to six
points of impact (Fig. 12).
3.5. Load Distribution
In calculations with three-dimensional models the load distribution is, by
definition, direct over an area of approximately 7 m 2 . In investigations with
axisymmetric models the load is distributed in the peripheral direction by
means of appropriate Fourier series. The size ofthe impact area is the same.
To achieve a representative load distribution for the buildings considered in
this study it is necessary to use Fourier series with some 10 to 12 rows.
3.6. Dynamic Response of Typical Structures
For the evaluation of the consequences of an aircraft crash and for the
design of substructures and equipment components it is essential to have
complete information on the local load functions in the form of time
histories and response spectra of the accelerations. No less important is
information on the stresses and the section forces and moments which may
be expected in the most severely stressed regions of the structures. The
investigations were carried out with the assumption of linear elastic stress
states in the structural regions remote from the area of impact, as defined
earlier, and with the following alternative assumptions for the region
immediately adjacent to the area of impact: undeformable area of impact;
elasto-plastically deformable area of impact.
The effects of plastic deformation on the results of the dynamic
calculations, and the consequences for the design of substructures and
equipment components, are discussed below for the examples of the
building concepts defined earlier. The calculations are based on the same
analytical models (Figs. 12 and 13). The inherent behaviour, the modal
parameters and the structural damping ratios are similar. The calculations
for each building are based on the load-time function specified according to
357
ref. 4 and on the function determined taking into account the non-linear
behaviour (Fig. 5). The reductions in the dynamic response ofthe global
structure by virtue of taking local plastic effects for characteristic regions
are compared and discussed. The computations were carried out using the
programs MESY and STARDYNE.
3.6.1. Calculations for an Undeformable Wall
In calculations based on the assumption of a locally stable wall in the area
of impact, the entire momentum of the load-time function specified for an
infinitely rigid area of impact is transferred to the global structure, which
undergoes elastic deformations both locally and globally. The assumption
of elastic deformation parameters leads to an anticipated dynamic
response force F d d = / ( C e l ) , which is substantially greater than the force
F d p | = / ( C p l ) calculated taking plastic deformations into account
It should also be noted in this connection that calculations based on the
impact as defined in ref. 4. i.e. assuming a rigidity of the area of impact
(infinite rigidity) greater than it can be in reality, result in a higher value of
the momentum to be transferred.
3.6.2. Calculations for a Deformable Wall
Taking the elasto-plastic deformations in the area of impact into account
results in much smaller values of the total transferred impulse and the
resulting dynamic response force.
The total impact force acting on the building along the transition
boundary to the elasto-plastic region (R = about 7-3 m; Figs. 5 and 6) is
smaller than that acting along the boundary ofthe main impact area (R
= 3m), assuming linear elastic deformations (Fig. IB).
3.6.3. Time Histories of the Accelerations
The time histories of the displacements and accelerations which may be
expected in characteristic regions ofthe building structures as a result of the
loads discussed above are presented in Figs. 16-20 for the example of a
reactor building. These data are presented with reference to the three
directions (radial, tangential and vertical) of the global coordinate system
describing the building structures. It can be seen that these phenomena
decay practically completely within about 0-5-2 s, so that the behaviour of
the building structures need not be followed any further.
In the calculations for impact point 1 only the zero harmonic was
considered, whereas in the calculations for impact points 2, 3 and 4 a total
358
N. J. KRUTZIK
20
0 o
: \
2 0
2 0
0 0
-20
20J
00
-20-
O)
t/>
*./W'
tr
20
o
o
00
<
-20
20
0 0 >Ued\j\pQ&*\f*ev~^*
-20
20
0 0
^&/Âe^
-2-0
RADIAL
VERTICAL
0-08
024
0-40
056
072
0 88
104
00
016
032
048
0 64
0 80
096
TIME(s)
FIG. 16.
Time history of acceleration at the level ofthe foundation plate.
359
000 u016
0-32
0-48
0-64
0-80
0-96
008
024
0-40
0-56
0-72
0-88
104
TIME (s)
FIG. 17. Time history of acceleration at the level ofthe inner cylinder, upper edge.
360
N. J. KRUTZIK
002
000
-002
004
002
000
-0 02
0-02
F2
000
S -002
o
< -004
UI
_l
o.
o
002
0-00
-002
-004
002
000
-0-02
-004
0-02
000
-002
-004
000
FIG. 18.
0 16
0 08
0 24
032
0 48
0 40
0 64
0 56
TIME (s)
0 72
080
0 96
0-88
104
Time history of displacement at the level ofthe foundations plate.
361
000
FIG.
19.
008
016
024
032
040
048
0 64
0-56
0 72
TIME (s)
0 80
0-88
0 96
104
Time history of displacement at the level of the inner cylinder, upper

edge.
362
N . J. K R U T Z I K
HORIZONTAL
VERTICAL
0 0 0 0 0 8 016 024 0 32 0 40 0 48 0 56 064 0 72 0 8 0 088 096

TIME! s )
FIG. 20.
Time history of acceleration at outer wall region.
J L J_
50 000 100 000
m /s2
FIG.
50 000 10O0O0
21 Maximum value acceleration.
104
J L _L
50000 100000
J_J L J _
5 0 0 0 0 100 000
FIG. 22.
tt7~t
T
il
Maximum value acceleration.
x"
'
JL,h^n
,
'yf/
^S^*
'I
"
1 I
"+
"
*-*
50 0 0 0 100000
<
50000100000
FIG. 23.
Maximum value acceleration.
363
364
N. J. KRUTZIK
50
100
tm/s2)
LOCATIONS
0o
,15-
,45
,90u
180
FIG. 24.. Maximum values of accelerations.

of eleven harmonics were taken into account. The time interval was 1 s and
the time step was 2-5 ms.
The maximum and minimum values ofthe accelerations, obtained from
the time histories for each impact area considered, are presented as
horizontal and vertical vectors for the nodal point concerned (in the
azimuthal plane 0; Figs. 21 to 23). In Fig. 24 relating to an impact at point
5 taken as an example, the distribution ofthe accelerations in the peripheral
direction (0. 15, 45, 90 and 180) is presented. It can be seen that
365
ANALYSIS OF A I R C R A F T I M P A C T PROBLEMS
the accelerations decay rapidly beyond the immediate vicinity ofthe point
of impact.
3.6.4. Response Spectra
The form in which the load data are usually required for the design of
substructures and of equipment components is the evaluated form of the
time histories, called the response spectra.
F
o
\ t
24
idi
22
20
18
STIFF
PLASTIC
14
12
10
> \
8
6
-1
4
2
cSSS ---" .-_-: - t

3
p>^
5 6 7 8 9 1 0
'V
20
/"*^
- ' "-"-v -
30
40 5060
8 0 f (Hz)
FIG. 25. Dynamic reaction at the level foundation plate (A) (maximum radial and
vertical).
Figures 25-30 show the response spectra for characteristic regions ofthe
building structure and of the steam supply system coupled to it, for both
load functions. It may be stated without limitation that when the local
elasto-plastic effects are taken into account, the maximum values of the
acceleration spectra are reduced by about 30-50 %, and it can also be seen
that the frequency regions ofthe maximum acceleration peaks are slightly
shifted. The acceleration peaks are generally lower. Owing to continued
cracking, which induces brief shocks in the structure, acceleration peaks
also occur at higher frequencies. Considered overall, however, the response
spectra are smoother.
366
N. J. KRUTZIK
26
24
22
Jdi
fijt
STIFF
PLASTIC
z-\
/f\\l/"N
// \ Li?
,^ N
--1" zj^
_^ss=3 5=;
FIG. 26.
|
20
5 6 7 8910
\~ zz~s~
30 40 5060
8 0 f (Hz)
Dynamic reaction at the level of the connection point of the RPV (B)
(maximum radial and vertical).
26
24
20
Idi
DI
18
STIFF
--PLAS TIC
16
14
12
IO
/ / /"\
Itz ^ *
\
'
y '//
/
5= -f - /
2
27.
'*
5 6 78910
t\
^-v
"^
FIG.
2X
A
L ^
j
20
30 40 5060 80 MHz)
Dynamic reaction at the level of the inner cylinder (upper edge) (E).
367
\J
-Wl
V, f
12
10
-STIFF
PLASTIC
__1. _
'"VL
\(2>
-^v
2
~~L~
^ i \Z
<.
^zzr-zzTzSF-j*:
5 6 78910
20
30
40 5060
80 )
FIG. 28. Dynamic reaction at the level of the foundation plate (A).
- J
j\ I
J '
24
22
- S T I F F - - P L AS TIC
II
2
1
J>
-\
3 4 5 6 7 8
\
\
. V\\ \
/
\
/ /
\>
/
10
^-.
18
'W
'/
.'
910
20
*\ \\ \
^
3040506080
A (Hz)
FIG. 29. Dynamic reaction at the level of the concrete ring (B).
368
N. J. KRUTZIK
~~|
(
.-
12
STIFF - -PLASTIC
10
/
/
= =
1
a.
4
2
z_g
;>>
5 6 78910
20
~ ^
30
40 50 60
80
/(Hz)
FIG. 30. Dynamic reaction at the level of the RPV support skin (D).
INFLUENCE OF MODEL AND PARAMETER VARIATIONS

ON THE VALIDITY OF THE RESULTS
Possible uncertainties in the assumptions regarding the input parameters
and the description of the investigated structures by means of models
should be covered by enveloping. The requisite smoothing ofthe calculated
response spectra depends on the influence and the range of scatter ofthe
following parameters: model representation: taking higher eigenfrequen
cies into account; soil parameters; damping parameters.
4.1. Model Representation
Investigations ofthe influence ofthe type of model representation on the
results of dynamic calculations for the load case of an aircraft crash have
shown that the local behaviour and the dynamic response can be
determined more accurately using shell models than using stick models.
The impact force applied to a small area induces local oscillations which
cannot be depicted by means of stick models. Figures 31 and 32 show
comparisons of the results for representative regions (Fig. 12) of an
axisymmetric building obtained for two alternative sets of calculations
namely, using a stick model and using a shell model. Figures 33-37 present
369
"0
8
Damping = 0 0 2
16
24
32
40
48
C
FREQUEN Y (Hz)
Shell model
FIG. 31.
Beam model
Response spectra, comparison.
40
f
32
i
f /
1 I
24
j j
f "
JI
1 \
1 \
16
*i
/ *
/' / / "
08
\
\
\\
i f
0
8
Dompinq=0 0 2
16
24
32
40
48
C
FREQUEN Y ( Hz )
Shell model
Beam model
FIG. 32.
56
64
Response spectra, comparison.
72
80
370
N. J. KRUTZIK
8 00
16 00
24 00
FIG. 33.
800
1600
32 00
40 00
48 00
56 00
64 00
80 00
Response spectra, comparison. X\.
2400
3200
1000
48 00
FREQUENCY(Hi)
56 00
64 00
Damping = O 02
FIG. 34.
72 00
FREQUENCY (Hz)
Damping =0 02
Response spectra, comparison, Xs.
72 0 0
80 00
371
100
16 00
24 00
32 00
40 00
48 00
56 00
64 00
72 00
80 00
FREQUENCYlHz)
Damping 0 02
FIG. 35.
100
16 0 0
Response spectra, comparison, X\.
24 0 0
32 0 0
40 00
48 0 0
56 0 0
64 0 0
FREQUENCY (Hz)
Damping =0 02
FIG. 36.
Response spectra, comparison, Xi.
72 0 0
80 0 0
372
N. J. KRUTZIK
similar comparisons for a box-shaped building idealised by means of a

beam modeland a three-dimensional model. It can be seen that not only the
outside structures but also the inside structures show quite different
dynamic response patterns in the high-frequency range. The results relating
to the directly loaded regions ofthe outside structures differ radically (Fig.
37).
With shell models (or with three-dimensional models) it is unavoidably
necessary to introduce some simplifications, because the internal structures
32
40
48
FREQUENCY (Hz)
FIG. 37. Response spectra at impact area, outer containment (2%).

are often quite complex. Minor deviations from the exact axisymmetric
shape, due to gaps, openings, branchings or annexes, can, however, also be
represented axisymmetrically provided only the shear areas, moments of
inertia, and moment distributions of the actual and the idealised structures
result in the same inherent behaviour.
4.2. Taking into Account Higher Eigenfrequencies and Higher Harmonics
The force applied over a relatively small area (7m 2 ) induces local
oscillations. Consequently, higher mode shapes are excited particularly
strongly. In axisymmetric models this is taken into account by means of
higher-order harmonics. Figures 38-40 present, for typical regions (Fig. 13)
of a PWR building, comparisons for response spectra calculated taking
into account two harmonics and eleven harmonics, respectively.
373
10
6
<
cr
L
_l
LU
<
/
/ y
.s
0
8
16
24
32
40
Damping =0 0 2
C
FREQUEN Y ( Hz)
11 Harmonics
,
2 Harmonics
FIG. 38.
48
56
64
72
80
Response spectra for load case aircraft crash (PWR, standard).
20
_^ 1 6
h*
rr
UJ
_j
UJ
808
<
/
//
04
^v
\y
0
8
16
Dompmg = 0 0 2 %
I l Harmonies
FIG. 39.
"s
24
32
40
C
FREQUEN Y [Hz]
,
48
56
64
72
80
2 Harmonics-
Response spectra for load case aircraft crash (PWR, standard).
374
N. J. KRUTZIK
4.3. Soil Parameters
As a rule, the influence of variation ofthe stipulated soil parameters on

the inherent behaviour of the investigated structures should also be
investigated. The extent of alteration of the inherent behaviour determines
the magnitude ofthe possible scatter ranges ofthe dynamic responses. In
the particular case of an aircraft crash, however, investigations have shown
2 Harmonics
, ' , Harmonics .
" V
Si 12
A
J
'
XV
//
04
^=r
/
0
16
Damping = 0 0 2 %
32
40
48
FREQUENCY(Hz)
-S
64
72
s:
FIG. 40. Response spectra for load case aircraft crash (PWR. standard)
that even relatively wide soil parameter variations (by 25 % and by
+ 50 %, respectively) cause hardly any deviations in the calculated dynamic
response. Variations of the soil stiffness result in shifts in the lowest
eigenfrequencies, but these frequencies are not excited by an aircraft crash
because ofthe short duration ofthe impact. With soft soils, however, the
likelihood arises that the building as a whole may be rocked by the impact.
4.4. Damping Parameters
The dynamic behaviour of structures may be significantly influenced by
their damping capacities. It can be shown that, with higher damping, the
time history acceleration amplitudes are smaller and decay more rapidly.
The response spectrum presented in Figs. 41-43 shows that the maximum
peak ofthe spectral acceleration is lower by about 30-50%. A shift in the
frequency range of the maximum response should also be expected. These
375
1,I.I /
/\
1
1
\ \
16
~--
"
/-
32
24
*-^. ^
40
48
56
64
72
80
FREQUENCY(Hz)
FIG. 41.
Response spectra at the level ofthe foundation plate ( 1 % damping), X\.
u*
28
\ , /
24
E 20
'i
I i
I 1
o
= 16
<
12
<
1
8
i?"C.
4
0
16
24
32
40
48
56
64
72
80
FREQUENCY (Hz)
FIG. 42.
Response spectra at the level ofthe top ofthe building (1 %damping). Xi.
376
N. J. KRUTZIK
7
6
'i
\
\
, -' \ *'r"*
7 \
h'
,~\^ \
FIG. 43.
16
24
32
40
48
FREQUENCY(Hz)
56
64
72
80
Response spectra at the level restraint location RPV (I % damping) X\.
08-
-hfi-lr Li,
- Compued
-Fordesign use
07
06
>l
J05
O
1c
Lu
I
LU
03
02
Ol
02
04
060810
FIG. 44.
20
4 0 6080100
FREQUENCY(Hz)
1
20 0
Broadening the envelope spectra.
I L
040060
377
results were obtained in investigations based on the beam model for the
box-shaped building.
4.5. Corrections and Smoothing Procedures
The extent of the corrections of the response spectra necessary depends
on the extent ofthe possible shifts in the eigenfrequencies ofthe system.
Taking into account the parametric influences which may be expected, the
local deformations, the location-linked ranges of scatter of the eigenfrequencies and the plastification of the impact area, the following
corrections and smoothing procedures should be applied:
(1)
(2)
(3)
(4)
5.
Reduce the acceleration at the first peak by about 20% (plastic

effects).
Widen the acceleration plateau by + 10% of the peak frequency
(Fig. 44).
Envelope with smoothing tangents.
Ignore valleys between consecutive peaks for frequency ratios of
less than 3.
PROPAGATION OF VIBRATIONS FROM ONE BUILDING
TO ANOTHER
11 is known that the dynamic response of coupled oscillating structures may

be enhanced or damped, depending on the ratios of their stiffnesses and
masses. Considering the soil as a connecting medium with a certain
stiffness, two neighbouring buildings should be considered as one
oscillating system.
Investigations of two neighbouring buildings of unequal sizes (reactor
building and secured diesel building; Fig. 45) have shown that:
(1)
(2)
The presence of the smaller building has a damping effect on the

dynamic response of the larger building (the reactor buildings).
Part of the energy is transferred to the smaller buildings (Fig. 48).
An impact on the larger building excites oscillations in the smaller
buildings. The extent of this excitation depends on the ratio of the
masses but. generally speaking, the accelerations may be expected
to be similar to those induced by a small earthquake (Fig. 47).
The dynamic responses of the foundation slabs of the active buildings

(the directly hit building) and of the passive building (the building excited
via the soil) are shown in Fig. 46.
378
N. J. KRUTZIK
il
-ri
SA
MH
ws
MH
DG
FIG. 45. Layout plan of buildings. MH = machine housebuilding. RG = reactor
building, WS = workshop building, SA = switchgear building. DG = diesel
building.
25 00
2000
5 15 00
en
1000
Aclive
5 00
Possivi
000
K>0
...
100 0
FREQUENCY(Hz)
FIG. 46.
Vibration of active and passive building (foundation plate).
379
.SSE
\s
\.
AC C
<*
>*^3
L *"
10
01
8 10
20
50
100
FREQUENCY (Hz)
FIG. 47.
A ccelerations at the level ofthe foundation plate due to earthquake (SSE)

and aircraft crash (A CC) loadings.
25 00
22 50
2 0 00
SINGL
17 50
l\
y 15 0 0
ui
I ^-f
T D--0
12 0 0
g loco
F-
7 50
on
5 00
2 50
^s,
0 00
'0 0
FREQUENCY(Hz)
!0
FIG. 48.
Interaction of adjacent buildings.
100 0
380
N. J. KRUTZIK
20
I
1
C_(,
Atrcrrjf 1 Crosh
Explosion
f\
\\
(K- 8 00
0 00
16 00
^h24 00
Damping, -O 0 2
FIG. 49.
0 00
32 00
40 00
48 00
FREQUENCY (Hz)
56 00
64 00
72 00
80 00
Comparison of response spectra due to external dynamic loads. PWR

reactor building/foundation plate, radial.
800
1600
24 00
32 00
4000
4800
FREQUENCY(Hz)
5600
64 00
7200
8OO0
Damping =0 02
FIG. 50. Comparison of response spectra due to external dynamic loads. PWR
reactor building/foundation plate, vertical.
381
6.
COMPARISONS WITH AN EARTHQUAKE AND AN

EXPLOSION SHOCK WAVE
The excitation functions of all these load cases are of a transient and
stochastic nature. Their effects, however, are quite different, because of
different durations and different frequency contents.
None of these load cases can be considered on its own as representative
for the design of all the structures and equipment components. Figures
0 00
8 00
Damping =0 02
16 00
24 00
32 00
40 0 0
48 00
FREQUENCY(Hz)
56 00
64 0 0
72 CO
8000
reactor building/upper interior cylinder, radial.
49-52, relating to the IRAN reactor building, present comparisons ofthe
frequency-dependent responses (accelerations) of the structure. Although
the effects of an earthquake and an internal explosion can be determined
with the aid of equivalent stick models, the results compared here were all
obtained using the same analytical model (shell model) in order not to affect
the comparisons by disparities in the models.
The calculations were based on the same determinations of the
eigenvalues, except that in the load case of an aircraft crash the analysis was
performed to include eleven harmonics and eigenfrequencies up to 80 Hz,
whereas in the load cases of an earthquake and an external explosion the
analyses were limited to the zero and first harmonics and to eigenfrequencies up to 40 Hz. It can be seen that in the low-frequency range (up to 5 Hz)
382
N. J. KRUTZIK
20
//
// /
/ k* !/
/
12
<
I
I
I
1
Ol
\\
\\
\\
\
\
\
\\
1
t
.//*;
8
16
Damping = 0 2
Aircraft crash
24
32
Safety earthquake Explosion
4C
48
56
64
FREQUENCY (Hz)
72
80
reactor building/upper interior cylinder, vertical.
the load case of an earthquake is governing for the design, whereas in the
high-frequency range (above 10 Hz) the load case of an aircraft crash
definitely governs.
7.
DISCUSSIONS AND CONCLUSIONS
The continuously escalating requirement imposed on building structures

and equipment components relevant from the safety engineering point of
view can be satisfied economically only under certain preconditions. In
designing these structures against external events with relatively low
probabilities of occurrence it is necessary, for reasons of technical and
economic feasibility, to utilise the load-carrying reserves to the utmost
limit.
Thorough experimental proof of the real load-carrying capacity of
reinforced concrete components subjected to impact loads will probably be
delayed for another few years. Nonetheless, it is already possible at present
to arrive at realistic and usable conclusions by analytic means, starting from
realistic assumptions and calculable simplifications. It may be stated
without doubt that the location-linked induced accelerations in characteris
tic structure regions, calculated by taking into account the local elasto-
383
plastic deformations, are much lower than those calculated assuming linear
elastic behaviour and infinite rigidity at the point of impact.
The findings of the investigations carried out so far may be summarised
as follows:
(1 )
(2)
(3)
(4)
(5)
(6)
(7)
(8)
The accelerations of building structures which may be expected in

the event of an aircraft crash reach in some structure regions very
high values compared with those resulting from other load cases of
external origin.
The maximum values of the accelerations lie in a relatively high
frequency range (usually above 20-30 Hz) in which the excitation of
most reactor components and systems is only marginal.
The high acceleration peaks occur only in the immediate vicinity of
the point of impact, and decay very rapidly. With redundant
duplication and physical separation of the components installed in
the building and relevant from the safety point of view, these high
peaks are only of hypothetical interest and are not governing for the
design.
The plastification of the area of impact results in considerable
dissipation of energy and in reduction in the dynamic response of
the structure regions.
The global thrust at the transition boundary from the plastified to
the elastic zone is reduced by about 50%, referred to a wall.
Accordingly, the acceleration peaks in characteristic regions ofthe
structures, referred to the frequency range ofthe highest resonance
peak, are lowered by 30-60%, and in the rigid body domain by
about 50 %.
The frequency range of the highest resonance peak is shifted by
about 10% towards higher frequencies.
The section forces and moments in the governing locations of the
building structure are reduced by about 50%.
In view of these findings, of the several conservative assumptions

incorporated in the load function specified at present, and ofthe specificity
and low probability of occurrence of the load case of an aircraft crash,
calculations taking into account local plastification phenomena constitute
a realistic and economically necessary basis of design.
Until the deformation behaviour of reinforced concrete is better known
on the basis of large-scale experiments, this procedure is the only
technically sound alternative making it possible rationally to solve the
problems in reactor design posed by the load case of an aircraft crash.
384
N. J. KRUTZIK
8.
PROSPECTS
In order to establish a better basis for the dimensioning and analysis of

concrete structures subjected to impact loads, the German Federal
Ministry of Research and Technology initiated in the years 1974-75 some
research projects with the following aims: to determine impact time
histories for impacts from strongly deformable flying objects; to determine
the kinetic load-carrying behaviour of reinforced concrete slabs.
These investigations are being pursued by HOCHTIEF and by the
Federal Agency for Defence Procurements. The programme comprises a
theoretical part and an experimental part. The theoretical investigations
are intended to provide information on: the impact of deformable objects;
the non-linear deformation behaviour of reinforced concrete components
subjected to time-dependent loads; the effects of finite deformations; the
stress-strain state in reinforced concrete slabs in the region of application of
the load.
The experimental part ofthe programme encompasses the procurement,
assembly and trial operation of an accelerator facility, the construction of a
target abutment block and the fabrication of about 24 flying objects and 24
test slabs in reinforced concrete. The theoretical work has been completed.
The experiments are well advanced. The work was expected to be
completed some time in 1979.
In another research project the results of projectile impact experiments
carried out earlier at the Ernst Mach Institute are being analysed (KED)
and compared with each other. The aim of this investigation is qualification
of the calculation methods for impact calculations for the experiments
referred to above.
BIBLIOGRAPHY
1. Biggs, J. (1964). Introduction lo Structural Dynamics, McGraw-Hill. New
York.
2. Chelapati.S. V.and Kennedy. R. P. (1972). "Probabilistic assessment of aircraft
hazard for nuclear power plants', Nucl. Eng. Des., 19, 333-64.
3. Hartmann, R., Schrader, K.-H. and Winkel, G. 'MESYEin Programm zur
Untersuchung von Tragwerken, Konstruktiver Ingenieurbau, Berichte. Heft
22, Ruhr-Universitt Bochum. 1975.
4. Drittler, K. and Grner, P. 'Zur Auslegung kerntechnischer Anlage gegen
Einwirkungen von aussen. Teilaspekt: Berechnung von Kraft-Zeit-Verlufen
beim Aufprall deformierbarer Flugkrper auf eine starre Wand'. Wissenschaftliche Berichte. I1RS-W-14. April. 1975.
385
5. Drittler, K., Grner, P. and Krivy, J. 'Berechnung des Stosses eines

deformierbaren Flugmodells gegen ein deformierbares Hindernis, Wissenschaftliche Berichte IRS-E-SS, April, 1976.
6. Krutzik, N., Semann, R. and Winkel, G. 'Dynamische Berechnungen fr das
reaktorgebude KKK 1, Lastfall Flugzeugabsturz, elastisches Materialverhalten", Techn. Bericht KWU/R 2-3584, December, 1975.
7. Krutzik, N., Winkel, G. and DSS. 'Dynamische Berechnungen von
Hauptstrukturen der Anlage KRB II, Lastfall Flugzeugabsturz', Techn.
Bericht KWU/R 14-36-76. July 1976 (DSS: Ingenieurbro Dynamik spezieller
Strukturen).
8. Krutzik, N.. Winkel, G. and DSS. 'Dynamische Berechnungen des
Reaktorgebudes IRAN bei Belastung infolge Flugzeugabsturz', Techn.
Bericht KWU/R 14-148-77.
9. Lebrenz, H. and Pohl, W. 'Anisotropie-Modell fr PISCES 2 DL". Control
Data GmbH, Frankfurt (Main), November, 1975.
10. Maxwell, D. E. 'Summary of concrete models', Physics International
Company, January, 1970.
11. Nowotny, B. and Daublebsky, O. 'Geschlossbelastung von Beton, Betonmodelle fr PISCES-Berechnungen', Techn. Bericht KED 1/74, Jan. 1974
(KED Ingenieurbro Kerntechnik, Entwicklung, Dynamik).
12. Physics International Company. PISCES 2 DL, Manuals A, B, C; PISCES 2
DELK, Manual and Equations (Version 1) San Leandro, California, 2700
Marced.
13. Riera, J. O. (1968). 'On the stress analysis of structures subjected to aircraft
forces', Nucl. Eng. Des.
14. Semann, R. and Krutzik. N. 'Ermittlung des massgebenden Aufprallpunktes
fr das Reaklorgebudes KKK 1 beim Lastfall Fluzzeugabsturz', Techn.
Bericht KWU/R 2-3609, December 1975.
15. Schalk, M. and Wlfel, H. 'Response of equipment in nuclear power plants to
airplane crash', ELCALAP, Berlin, 1975.
16. Schmidt, H. D. and Winkel, G. 'Lokales Verhalten von Gebudestrukturen
beim Lastfall Flugzeugabsturz unter Bercksichtigung des nichtlinearen
Materialverhaltens von Stahlbeton. Techn. Bericht KWU/R 14-19/77, April,
1977.
17. Schrader, K.-N. 'Die Deformationsmethode als Grundlage einer problemorientierten Sprache", B. I .-Hochschultaschenbcher, Bibliographisches Institut,
Mannheim.
18. Stangenberg. F. 'Berechnung von Stahlbetonbauteilen bei Kernkraftwerken
fr extrem dynamische Beanspruchungen". 2nd SMIRT Conference, Berlin,
September 1973.
19. Kamil,* H.. Krutzik. N.. Kost,* G. and Sharpe,* R. 'An overview of major
aspects of the aircraft impact problem'. 4th SMIRT Conference, San Francisco,
1977.
*Engineering Decision Analysis Company Inc., Palo Alto, California, USA.
20. Winkel, G. and Semann, R. 'Optimierung der Grssen von Plattenelementen
bei dreidimensionaler Idealisierung von Gebudestrukturen", Techn. Bericht
KWU R 11-3608.
21. Zerna, W. and Stangenberg, F. (1974). 'Zur Auslegung von Kernkraftwerken
auf aussergewhnliche dynamische Lastflle. Bauingenieur, No. 49.
386
N. J. KRUTZIK
22. Kaiser, ., Krutzik, . and Schrader, .-. 'Local and global response of
reactor buildings at the load case aircraft impact', 4th S MIRT Conference, San
Francisco, 1977.
23. Schrader, K.-H., Kaiser, A. and Krutzik, N. 'Comparison of the dynamic
response ofthe structures of a typical DWR-reactor building to the load cases
earthquake, aircraft and explosion', Conf. on Structural Analysis Design and
Construction in Nuclear Power Plants, Porto Alegre, Brazil. April, 1978.
12
Pipe Whip Analysis
L. LAZZERI
AMN,
1.
Genova, Italy
INTRODUCTION
Safety requirements force the designer to consider the most severe and
extreme loading conditions for pipes, components and systems, the failure
of which could jeopardise the safety ofthe plant. Leaving to the specialised
literature the description and discussion of the safety problems and the
implications in terms of fracture mechanics, ' " 5 - ' here only the calculation
aspects of the pipe whip problem will be discussed. Then we shall assume
that a break takes place in a given location in a pipe and that it will be
accelerated by the emerging fluid; the calculation ofthe pipe movements
and the problem of sizing the restraints will be discussed.
Particular importance will be given to the different ways of solving the
problems, from the simplified methods for an initial solution to the most
sophisticated.
2.
FORCES C O M P U T A T I O N
In this chapter the thermodynamic aspects of the phenomenon will not be

discussed, as they are considered as input data for the present analysis;
general reference is made to the Moody model. 5 ' 1
The jet force consequent on a complete circumferential crack (force
parallel to the axis) is given by
FSC=CPA
(1)
where is the pressure, A is the outflow area and C i s a coefficient function

of time and geometry flow areas upstream of the postulated break
3X7
388
L. LAZZERI
location. If a longitudinal crack is present, then the force is assumed 4 to be

the sum of the forces of the two possible circumferential cracks, thus:
pA
(la)
where C, is generally equal to the time-dependent coefficient for a

circumferential break on each side of the longitudinal crack. Cs is a
mitigation factor used to account for the possible benefits due to splashing
FIG. 1. (a) Sketch of a longitudinal break, (b) Sketch of a circumferential break.

ofthe fluid, etc. It should be mentioned that the hypotheses made in order
to compute the jet forces according to eqn. (la) are quite pessimistic and
extreme. Various phenomena such as crack propagation and final size, and
fluid flow change of direction around the crack, are physical factors which
could be claimed to mitigate the effects of the forces as computed by eqn.
(la).
In any event, some experimental research appears to be necessary to
make as quantitative as possible such qualitative considerations.
Sketches of possible longitudinal and circumferential cracks are shown in
Fig. 1. The force computation by the Moody method has been automated
and a computer code named G E T T O is available; it computes the forces
due to a circumferential and a longitudinal break for a given pipe line.
Effects due to flow area restriction, change of directions, friction, etc., are
included.
PIPE WHIP ANALYSIS
3.
389
SOLUTION ANALYSIS
In ref. 4 some criteria are given to decide whether restraining ofthe broken
pipe is necessary. No protection is required if any of the following
conditions is fulfilled:
(1)
(2)
(3)
The broken pipe is physically separated from any other component

whose failure could impair the safety of the plant.
No damage can result to any other related component, whichever is
the movement ofthe damaged pipe postulating a hinge at each end.
The kinetic energy of the pipe (due to its size and/or pressure) is not
enough to cause damage to structures or other components
important for the safety of the plant.
If these criteria are not fulfilled the uncontrolled movement of the pipe must
be prevented and restraints (or similar other devices) are necessary.
Basically the desirable characteristics of a restraining unit should be the
following:
(1)
(2)
(3)
(4)
(5)
(6)
No force or restraining action is applied during the normal life of

the plant: thermal movements or other slow movements are
allowed.
Large amounts of energy are dissipated by the restraints, while the
applied force is not too large, in order to make the deceleration
phase as smooth as possible.
If the restraining units are not in contact with the pipe, the kinetic
energy ofthe pipe before impact should be kept as low as possible.
The restraining unit should be as near as possible to the postulated
break point.
A modular solution is desirable for unification purposes.
Dismantling of the restraining units should be possible in order to
make in-service inspection as easy as possible; maintenance (if any)
should be easy.
Solid stainless steel bars working in the plastic range, dampers, friction
elements and other devices have been proposed as restraining units.
It should be further mentioned that the above-listed requirements might
not match at all. The problem of the study of the influence of various
parameters and of dynamic optimisation was discussed by Dini and
390
L. LAZZERI
force
displacement
FIG. 2. Bilinear characteristics ofthe restraint.

Lazzeri7 for the case of a restraint having bilinear stiffness characteristics
(see Fig. 2). It was shown that:
(1)
(2)
(3)
An increase in overhang length (distance ofthe restraint from the

force application point) causes large increases in the pipe strain at
the restraint location (particularly for the case of a circumferential
break).
An increase in the initial clearance between the pipe and the
restraint causes large increases in the pipe strain.
There may exist an optimum solution for the stiffness of the
restraint: overlarge values might cause excessive straining of the
pipe, while if the restraint is too weak, its excessive strain could
result.
All these considerations are clearly shown in Figs. 4-6 for the case depicted
in Fig. 3. It should be mentioned that sometimes the reverse is true:
excessive weakening ofthe restraint could cause large strains ofthe restraint
and of the pipe. The presence of one or another of the two behaviours
depends on the distribution of strain energy in the pipe; the first one is
generally true if the maximum strain is near the restraint location, while the
other tends to be predominant if the hinges are formed near the end. In this
connection, the influence ofthe overhang length is decisive; high values
tend to determine the formation of a hinge near the restraint location and
the first type of behaviour is likely to develop.
It should be further mentioned that particular conditions exist around
391
PIPE W H I P ANALYSIS
11
1 1
10
10
9
WV-| Ri
WH
R8
VW-f
R6
7
7
6
6
5 >
5
4
3 "
2 "
1
4
3
2
-WV-i R4
V W - | R2
I'
FIG. 3.
Model for Figs. 4, 5 and 6.
5
10
15
initial clearance (cm)
FIG. 4.
Influence of clearance.
392
L. LAZZERI
J 10
8.
o-
04
12
20
28
overhang length ( m )
FIG. 5. Influence of overhang length.

the ends ofthe pipe, particularly for the case of short runs, where generally
the restraint cannot work and all the force is directly absorbed by the pipe
end. When the computation ofthe forces on the restraint and ofthe strain
of the pipe has been performed, one should compare the actual with the
allowable ones. The allowable values for the restraining unit should be based
on their maximum load capability. For dampers in the form of hydraulic
snubbers load capabilities as large as 1 IO6 lb have been reported by
restraint flexibility (10" e m/kg)
FIG. 6.
Influence of restraint stillness.
PIPE WHIP A NA LYSIS
393
manufacturers (the allowable values should be some fraction of it). For a

restraining unit of solid 125 in. diameter stainless steel bars, allowable
values of roughly 150 000 lb result if a maximum strain of roughly 25 % is
assumed. For the pipe maximum capabilities should be considered in order
to assess the allowable; it should be mentioned that the straining on the pipe
could be quite high and possible ovalisation and even flattening instability
phenomena could be present during the transition.
4.
ENERGETIC PRELIMINA RY SOLUTION
A preliminary solution which tentatively fixes the position and characteris

tics ofthe restraining units is highly desirable for the designer, particularly
as there is often an optimum solution, for which no excessive straining both
ofthe pipe and ofthe restraint takes place, as discussed in Section 3. Such
preliminary analysis is performed by GETTO 2 code 10 under the assumption
that the restraints are as follows: modular units have bilinear characteristics
(as in Fig. 2); and modular units have different characteristics as far as the
direction of impact is concerned. So each unit will have two different sets of
properties (maximum load, stiffness, initial gap), one for the frontal and
one for the lateral direction.
The output of the G E T T O 1 program containing the applied jet forcesat
each break location is directly used by the G E T T O 2 program. For each
location in which a restraint is placed the number of units which are wanted
by any postulated break are computed. The following information must be
given to the GETTO 2 program: frontal and lateral characteristics ofthe
restraint; and the break forces by which each restraint is loaded.
A rough energy balance between the external work done by the jet forces
and the energy dissipated by the restraint and the pipes is performed under
the assumption that a predetermined amount ofthe energy is absorbed by
the restraint. The tentative number of modular units, N, is consequently
computed by
/ = \
E,
i=
where IF is the external work, E the maximum energy per restraint unit, N
the number of units per restraint, i the number of restraints and av the
correction coefficient (percentage of total energy in the restraints).
394
L. LAZZERI
5.
FINITE ELEMENTS ANALYSIS
Once a preliminary solution has been obtained, the final appraisal can be
performed by means of a more refined finite elements code. In these
paragraphs some indication will be given of some of the finite elements
available, with particular reference to the codes developed by the author
and his colleagues.
5.1. Analysis by Means of Finite Elements Codes
As an example one may quote the cases of the A D I N A " and the
MARC 3 0 codes; these general finite elements codes allow a complete
simulation of pipeworks in the plastic field. ADINA's straight beam
element can be used for a finite elements representation, while some
difficulties may be found in simulating elbows. The MARC code has a
special element made ofthe shell's plastic element (according to Sanders 12 )
which can be used to simulate elbow elements. However, the simulation of
the pipework by means of a general finite element code may be very heavy,
so that ad hoc codes may be used with some advantage.
5.2. The FRUSTA Model ofthe Pipe
The aim of this chapter is the description of the computation models and
algorithms necessary to make a detailed analysis of the transients
consequent on the application of the jet forces by means of the FRUSTA
code. 1 3 Early calculations 7 1 3 performed with a preliminary version ofthe
FRUSTA code made it clear that only the part of the pipe immediately
adjacent to the break was absolutely necessary in order to compute the
forces, stresses and strains during the transient. Consequently it was
FIG. 7. Typical pipe run.
PIPE WHIP ANALYSIS
395
decided to make a full dynamic simulation only of the pipe spans adjacent
to the break.
With reference to the piping sketch (Fig. 7), if a longitudinal break occurs
at location 1, then only the two spans 01 and 12 will be fully simulated
(statically and dynamically), while the remaining part of the pipe will be
simulated only statically by means of suitable dummy elements. The
author 7 has already shown that in many cases even the static simulation of
(a
(b)
FIG. 8. (a) Cantilever geometry, (b) Double built-in geometry.

the pipe part far from the break is not important in the dynamic response of
the structures. The simulation ofthe pipe is performed by the use of four
basic elements: (1) a pipe element with complete elasto-plastic behaviour;
(2) a dummy element with indefinite elastic response ; (3) a restraint element
with bilinear stiffness characteristics; and (4) a damper element whose
restraint force is proportional to the velocity ofthe damper location node.
Two different geometries can be studied: (1) the cantilever geometry of
Fig. 8(a) (for circumferential breaks); and (2) the double fixed geometry of
Fig. 8(b) (for longitudinal breaks).
The general behaviour of these elements and the integration schemes and
techniques will be discussed below.
5.2.1. Pipe Element
General hypotheses. The following hypotheses are made to simulate the
pipe behaviour:
(1)
(2)
(3)
Plane sections remain plane.

Displacements are due to moment only.
In the elastic domain Hooke's law holds: = (where a is the
stress, is the strain and E is Young's modulus).
396
(4)
(5)
(6)
(7)
L. LAZZERl
In the plastic domain : = ' (where A is the plastic constant and

m the plastic exponent).
Ovalisation occurs if the maximum strain in the section is larger
than a predetermined amount.
Instability occurs when the specific energy overcomes a pre
determined value.
Crushing of the pipe is possible when it impacts a hard flat surface.
moment,.
elastic unloading
-instability
\
ovalisation
E3
elastic plastic
rotation
FIG. 9. Schematic drawing of the behaviour of the pipe subelement.

Pipe subelement. Each element is subdivided into a number of
subelements (maximum 20) and the energy accumulated by each
subelement is the main parameter to determine pipe behaviour. Three
energetic parameters are used to define it : , (elastic limit), E2 (plastic limit)
and E3 (ovalisation). Let "se be strain energy for each subelement; then the
behaviour is as follows, where the dot denotes derivative with respect to
time:
se<l
> 0
, < s c
2<Sse<
&3 <
<2
*.,
elastic
plastic
ovalisation
instability and flattened behaviour
r5se < 0 elastic, if E < E3, i.e, elastic unloading is followed. Furthermore,
once unloading has occurred, elastic behaviour is assumed until the energy
for which unloading first happened is overcome. In Fig. 9 the behaviour just
described is schematically drawn, The values of E,E2 and 3 are
calculated as follows:
Ex is the energy value for which the elastic and plastic moments
computed according to the plastic law are the same.
397
E2 is the energy value which corresponds to the value of the moment for
which the maximum strain is equal to , i.e. the value of the strain for which
ovalisation occurs.
3 is assumed to be some multiple ofthe E2 value; in the present version it
is assumed that
E3 = nE2
=3
(2)
Any different value for could easily be assumed to match the result of
experimental data; however, this problem does not appear to be decisive, as
will be discussed later. The value of is computed by means of the
equation 9 , 1 4
= 06yt/R
(3)
where t is the thickness, R is the radius and is a factor which typically
ranges from 1 to 2.
For each subelement a constant value ofthe bending moment is assumed,
so that if M a n d Fare the moment and shear at one end ofthe subelement
and / is its length, the average moment which is considered is given by
3 = |05/
(4)
An incremental approach is followed and increments of displacement W

and rotation at each step at the subelement level are given by
cW
AW = Â T
cW
+ ^ A M
CM
(5)
n
0 = + A M
where the coefficients of the tangent flexibility matrix can be computed by

the following formulae directly in terms ofthe strain energy. In the plastic
range it is
cW
2 /
~df
dW
M
1
mEIm
\ 2 n i ; l +m /
\1 m'l+m
+1
1 m/1 + m
2 \ mEIm
+1
1 ml 1 + m
>
M
Elm =
2m/ 1 + m /
R\2
A R2t
1 m 1 + m
1 4
,mElm
m
sin
l mi 1 + m
'OdO
(6)
398
L. LA ZZERl
In the elastic range the flexibility matrix is given by eqns. (7):

cW _
l3 W _ cO _
~df~ AElcM
I2 OL
2EIJM
I
El
(7)
where / is the elastic inertia moment. In the ovalisation range eqns. (7) are
again used but the elastic modulus is drastically reduced in order to obtain a
characteristic as flat as possible. The displacement and rotations at the
subelement level are internally condensed and the final flexibility matrix for
each element is assembled.
moment
not crushing
'conoc < f l,r

contact ;> In
rotation
FIG. 10. Schematic drawing of the crushing behaviour of the pipe.

It should be stressed that the flexibility matrix at the element level is a
function of the flexibility of each subelement and consequently of the strain
energy distribution on each of them, and that propagation of plasticisation
and ovalisation is therefore automatically considered.
If on any of the subelements the energy F se overcomes the 3 value,
instability occurs and unloading happens until complete annihilation ofthe
element moment is reached. This unloading is computationally handled by
application of fictitious forces at each node, so that redistribution ofthe
moment pattern is achieved. A hinge is consequently assumed on each node
where instability occurred. Furthermore, if the pipe impacts on a restraint
where crushing may occur, the absorption of energy is simulated by means
of a'crushing restraint' (see Section 3): when the contact force between the
pipe and the fictitious restrain overcomes a specified value, i.e. Fconlacl > F,,m.
no further absorption of strain energy is allowed on the pipe, the pipe
is unloaded and the hinge is assumed to be present (see Fig. 10).
Total pipe stiffness matrix and general performance. To show the
performance of the pipe element the example in Fig. 11 is given. A pipe
PIPE WHIP ANALYSIS
ace (10 m / s 2 )
displ(10"' m )
mom (10
kg m )
force I elastic
IT
plastic
ovalisation
.fe
at instability
moment (10
kg.m)
5~
instabihty
displacement ( 1 0 " ' m )
FIG. 11.
Performance ofthe pipe element.
399
400
L. LAZZERI
segment of 2 m length, 610 mm diameter and 25 mm thickness is loaded as a

cantilever by a force of 350 tons; 20 subelements have been considered to
simulate the transient and the pipe behaviour. It should be mentioned that
the ovalisation phenomenon tends to be extremely localised ; in this case the
ovalisation can arrive only at the second element from the end before a
hinge is formed at the pipe end. This fact accounts for the small absorption
of energy after ovalisation began to take place. In fact, even if the
absorption of energy in the ovalisation range is comparatively large, the
zone where ovalisation takes place is extremely localised, so that the total
energy increment at the element level is quite limited. The abrupt fall in the
moment obviously causes an abrupt increase in the acceleration, as the
equilibrium to the applied force is then simply due to d'Alembert force of
the mass at the pipe end. The elastic, plastic, ovalisation and instability
phases are clearly depicted in Fig. 11. Owing to extreme localisation of the
ovalisation phase, the actual choice ofthe value ofthe n parameter is not
critical and even large changes should not strongly affect the general
response ofthe pipe element, because once the ovalisation occurs, if other
parts do not contribute to the overall resistance, rapid collapse of the pipe
occurs without any further large absorption of energy.
5.2.2. Dummy Element
A dummy element can be introduced which has an indefinite elastic
behaviour; its flexibility matrix formulation is obvious. The use of a dummy
element allows the simulation of particular end conditions for the pipe.
5.2.3. Restraint Element
A restraint element can be introduced and a sketch describing its
behaviour is depicted in Fig. 12. The stiffness ofthe restraint element is
force
elastic unloading
displacement
FIG. 12. Restraint properties.
PIPE WHIP ANALYSIS
401
added directly to the pipe element at each step for the computation ofthe
overall stiffness matrix:
K, = 0,
if X < XQ
K = FK0,
if XG < X < XE
K=FK
if X > AV
(8)
where XG is the initial gap, XE the elastic limit and X the displacement ofthe
node where restraint is applied.
Unloading is supposed to be elastic, so (if impact has occurred) one has
X < 0,
KT = FK0
Once unloading has occurred, the stiffness is given by eqn. (1 l),even if ->0,
until overcoming of the value of displacement for which unloading first
takes place. Obviously no traction by the restraint is possible, so that the
stiffness is equal to zero if contact is lost between pipe and restraint, the Xc
value is consequently changed to 2 (i.e. the value of displacement for
which contact is lost).
This model of restraint is a relatively simple and crude one; however, the
author believes that it can be satisfactorily used for practical situations.
Furthermore, a maximum force for the restraint can be specified. Should
the actual force overcome it, the restraint is assumed to have been broken
and no force can be given by it.
In practice this is done by assuming a flat characteristic for the stiffness
and by applying a continuous force opposite to maximum.
Pipe crushing is simulated by means of a proper restraint (see subsection
2.2), which simulates the pipe distortion (see Fig. 13) during the impact; a
value of Ehm must be specified at the restraint level. It could be discussed
force
pipecrushing displacement
FIG. 13. Crushing behaviour of the pipe.
402
L. LAZZERI
further when a 'crushing' or 'not crushing' restraint is present; this

obviously depends on the characteristics of the pipe and of the restraint.
However, one could consider that if the restraint is represented by a hard
flat surface, crushing will be highly likely, while if the restraint is a
comparatively soft one and prevents lateral deformations ofthe pipe, such a
possibility will be much less likely.
5.2.4. Damper Element
A damper element can be used and a restraint force proportional to the
node velocity can be introduced, thereby simulating the effect of a damper.
As the force is generally continuously directed in one direction, a damping
force is not generally sufficient to stop the running ofthe pipe, but only to
limit its velocity. The pipe may stop then only if the applied force drops
enough to allow the pipe itself to sustain the applied loads. However, if it is
so, an element of restraint can be added to the damper element. Then the
applied force is given by
F r d l = CX ; Fr2 = KT(X - Xg)
Fr = Ftdl + Frd2
if X > X
(9)
5.2.5. Friction Element

A friction element is specified where
F=0,
X < Xc
and
F = (velocity)
if X > XG
5.2.6. Masses
The mass of the system can be given as both a distributed and a
concentrated quantity directly lumped on each node. Even if the mass is
distributed on each element, it will be directly lumped on the nodes by the
program. The masses are consequently assembled as diagonal matrix.
5.2.7. Forces
The jet force can be applied at each node, but no more than one node can
be loaded (this is due to the hypothesis made to simulate the action of the jet
leaving the broken pipe, while no difficulty obviously exists at the program
level should a different simulation become necessary). The jet force is
assumed to change with time in a stepwise manner, 5 i.e.
F=Fy
F= F.
t<t,
/,,_, < / < !
= ,
(10)
PIPE WHIP ANALYSIS
403
5.2.8. Time Integration

The general equation to be integrated is
[K]\X\ + [C]\X\ + [M]X = \F(t)\
(11)
where [K] is the total matrix, [K] = [Kp] + [Kr]; [Kv] is the pipe stiffness
matrix; [KT] is the restraint stiffness matrix; [C] is the damping diagonal
matrix; [M] is the mass diagonal matrix; and |F(/)| is the applied force
vector (in this version only one term is different from zero). An incremental
approach is followed and the 'constant average acceleration' scheme is
used:
IA-,1 = |JT0| + 0-5A/|JTo + A-, |
| * , | = \X0\ + At\X0\ + 0-25|l 0 + X,\At
(12)
Here the subscripts 0 and 1 denote the value ofthe selected quantities at the
beginning and end of each time step. I f eqn. ( 12) is substituted into eqn. (11),
an incremental integration scheme results.
The stiffness matrix (and possibly the applied force vector) are reevaluated at each step, because the pipe and restraint stiffnesses change
during the integration phase. However, the effect of change of geometry is
not included.
A check on the actual moment, as compared with the theoretical
moment, is performed and then the maxima (positive and negative) ofthe
parameter are given by the program. A check is also performed on the
actual energy accumulated by the restraint as compared with the theoretical
one. Provided the time step is small enough, the discrepancies between
'calculated' and 'true' values are quite limited and negligible. Finally, a
check is performed on the overall energy (some discrepancies could arise as
a consequence of the abrupt change of properties during the integration);
again optimum convergence was generally achieved.
5.3. A Complete Three-dimensional Simulation of the Pipe
The relative merits and defects of the FRUSTA simulation have been
described; however, cases exist where a more detailed three-dimensional
simulation is necessary. Hence, the development of a new finite element
code with a complete pipe simulation began in the author's organisation in
early 1978. The code will have an ///W-library, which will be introduced in
a non-linear finite element code, possibly ADINA. 1 1
A description of the elements fully developed is given here.
404
L. LA ZZERI
5.3.1. PA PS Model ofthe Pipe

The pipe section mode/.2128 Let r and be radius and thickness of a pipe
in the initial undeformed condition and let us assume that the following
generalised correlation exists between octahedral stress and equivalent
strain :p:
=/()
(13)
as obtained from some uniaxial test, as, for example, the RambergOsgood
correlation for total stress and strain, i.e.
= ,+ (/) 1 "'
(13a)
where A, m = plastic parameters and E = modulus of elasticity.

Let us assume now that a generalised load vector \F] (three forces, three
moments and pressure) is present and that this vector is correlated to the
generalised strain vector |,|, through a correlation in the incremental force
of this type:
|A f: ,| = [ ] " , , | A F l
(13b)
where [] is a generalised stiffness matrix. To formulate the expression of

the flexibility matrix, the following hypotheses will be made.
(1)
The local equilibrium equations are waived, while the total

equilibrium equations are maintained, assuming that the only
components of the stress tensor different from zero are:
. =
a =
r. r =
r. =
(2)
axial stress
hoop stress
radial shear stress
circumferential shear stress
As suggested by Hodge. 15 hypotheses are formulated about the

strain distribution in the section as substitutes for the local
equilibrium equations.
Let us assume that the axial component of strain is
. = ! cos + 2 sin + 3
( 13c)
i= 1.2.3 being functions of Z. Similarly if no warping or

deformation of the section takes place, one has (where y denotes
shear strain)
)'_. = y, cos + 2 sin + 3
>'.-r = i'2 c o s + y sin
(13d)
(13e)
405
,, i = 1,2.3 being functions of Z. In this model all the unknowns

are strain components (viz. ,, y,, ; = 1,3 and the hoop strain r;h), and
this is the reason why this model has been named the strain model.
However, one could postulate crosssection warping occurring
during the plastic flow stage (experimental evidence would really
suggest so), so that one could consider a more complicated strain
field: as an alternative, hypotheses on the shear stress distribution
have been made in place of eqn. (13):
. = 0:
(3)
,0 = , cos f? 4 2 sin + 3
(14)
With this model (named 'mixed model' by analogy with the stress
and strain field hypotheses) one can immediately compute ,, 2 , 3
as functions of F2, F3, F 4 (or their increments).
The relationships between stress and strain tensor components are
formulated in terms of PrandtlReuss equations; 17 then for the
present case,
a,,, z being the components of the stress tensor
being the components of the strain tensor
y,j being the engineering shear strain
E, G being elastic constants
being equivalent stress and strains (plastic components):
di:: = (a. 0-50) + \E(da: ^0)""
0 = (0-
0-5.)/ + \/(0 - da.)
(15)
dy-o = 3. 0 / + dT_0/G
dy rr = 3. / + dx J G
(4)
von Mises equations are used to compute the equivalent stress. This
equation is formulated as:
2 = a2 + a2 - 0: + 3(xl, + 20)
(16)
Together with the above assumptions about the strain-hardening rule,

these hypotheses do completely define the pipe section model. However, as
far as the material strain-hardening model is concerned, one may note that
the general formula ( 13) can be further processed to account for the effect of
elastic unloading, by specifying that:
=
if < 0
or
a < a
where the dot indicates differentiation with respect to time and ays is the
406
L. LA ZZERI
TABLE I
H.
H:r
H
0;
g(0)
l/(0)
l/%(0)
1/G=O(0)
1/G.r(0)
=
=
=
=
=
=
=
=
cajcap
cijcap
czJdav
l/apdf/ap
(a. 0-5aO)g(0)IH.(0) + \E
(<r, 05a.)g(0)IH~.'B) v/E
3g(6)T:0/H.0(9) + 1/G
3*(0)r_ r /#,(0) + 1/G
prior m a x i m u m stress. This model c o r r e s p o n d s to the usual strain

hardening isotropic a s s u m p t i o n ; however, no difficulty arises if the
kinematic h a r d e n i n g rule is taken into consideration.
In addition, one should note that, if the mixed model is resorted t o ,
changes of shear strain would occur in the section according to eqns. (15)
and that an averaging technique is to be considered; in this case we shall
have:
~\
d)',, =
rdy.gde,
rdO
dy s 2 =
rdy.0cos(id0
dy s 3 =
fdy. 0 sin OdO
(17)
rdO
>
d0
J
where dy ( 1 , d y s 2 , d y s 3 are the average shear strains a r o u n d the axial and
towards the two horizontal axes.
Leaving the details o f t h e calculation to ref. 28, it is e n o u g h to say that for
the strain model the following equation is found for each section:
,
[b,k]-l\AF,\
(18)
}',
\AF\ being the increment of generalised load in each section. T h e meaning
o f t h e b;k p a r a m e t e r s is shown in Tables I and II. In the case o f t h e mixed
model, while the u p p e r portion of the equation still holds, the A y,
parameters are related to the increments of loads by means of
\Ay,\l
= u3
= {a,k]\AFk\k=2,4
(19)
TABLE 11
riv.
b)) =
rt cos :.()
12 =
smrt E.JO) dll
>13 =
b2 =
sin cos 0(G:r(U) + G ; (0))d
/>,_, =
,/(G. r (0)cos 2 + G .(0) sin2 0) d
b2 =
riE::()dO
r(G.(0) sin d
rT[C7rr()sin2 + G\ () ()cos 2 ]d
34 =
F, 5 =
2"
Jo
, =
/) 45 =
/,2,sinG..tf()d
/>4
E ..(0) dll
bi2=
ri E:JU)cosU sin UdU
bbl=
s.
rtE,JO)cosUdU
> s ., =
Jo
"0
fin
11 sinU cos OE:J0)d0
>
>
<
rln
'2
ricos2
riG.()d
2it
b5l =
"
rtGJO)cosOdU
0
r2tcosOG:U(0)dO
r-2n
r/ sino cos 0(G:r(0) + G:l(0))dU
-c
m
-0
^2
rz t E.JO) sin2 U du
63
r2iE:JU) sind
408
L. LA ZZERl
TABLE III
' 2 "cos 2 0d0
2nr 2 f
rir.
a212 == Inrh o
'il
fl,, = 2nrt
GJ)
coso sin 0 d
1
2nr2t
Gzti(t)
' 2 " sin 0 coso d0

GJO)
0
~ 2 "sin 2 0d0
o
24
Grfrf)
f 2,, sind
22 o GJ)
1 2" coso d
2nr2t o G:U(0)
'cosdt
1
4^J/J0
r2n
sinodi
1
3
~ 4nr t o G;(0)
1
= 4nr 3 ;
'34
G:ti(0)
" d
Glfl(0
where the a,k terms are shown in Table III. In any case the section
p a r a m e t e r s , Ay, will be related to the increments of loads in any section
by m e a n s of an expression of the type
,
7;
(20)
= [,*]'lA F*,
Then if the local loads \Fk\ are related to the loads applied at the end o f t h e
element \Nk\, a n d if the correlation between local strains and element
displacements |vt'| is expressed by means of the o p e r a t o r Lik, one o b t a i n s
\AWk\=L,k([;ki][n,k])\ANk\
where the matrix [rk] and the o p e r a t o r L are defined in Tables IV and V;
then the local stiffness matrix is defined by
[K,krl
= i rt ([,j'[/*])
(21)
Integrals as defined in Table V are appraised by digital p r o c e d u r e and

integrands are calculated in a predetermined n u m b e r of control (or Gauss)
points, thus ensuring u t m o s t flexibility, as the n u m b e r of integration points
does not interfere with the formulation itself. A stepbystep procedure is
used, with evaluation of the element stiffness at each step as follows:
(1)
A t each step evaluation of the equivalent moduli by m e a n s of Table

1 and c o m p u t a t i o n at each G a u s s point.
TABLE IV
All [i]ik] terms are equal to zero, excluding
= 1,
i = 1,5;
5 3 = 'h,2 =0
409
PIPE WHIP ANALYSIS
TABLE V
,,,}',) =
.4(,,,) =
L5(ABAy,)= \ ,/rdz
2(}>,) =
Lb(AsAy,) = f ,/rdz
,/rdr +
>', dr
(2)
(3)
(4)
Ay3/rdz
.3(>',)= A ejrdz + \ A y2dz

Jojo
Jo
Evaluation of matrices [bik] and/or [a,k] for each'control' section by

means of Tables II and III.
Calculation ofthe local stiffness matrix ofthe element to be used in
the formulation of the total stiffness matrix of the structure.
Evaluation of the load increments for the element and evaluation of
stresses. A ll the pertinent elements are stored on tapes and re
evaluated at each step, so that use of central processing unit areas is
kept to a fairly low level. This procedure is followed by the PA PS
T2G program. It should be mentioned, in addition, that while the
program uses the RambergOsgood material model with elastic
unloading according to eqn. (13), the formulation as presented in
the foregoing is absolutely general.
Pamelspepial elbow element. In order to completely simulate pipeworks,

an elbow element is necessary. A s mentioned earlier, a simulation of the
elbow by means of finite elements may be obtained if the MA RC code is
used. However, a special finite element is preferable in terms of simplicity of
application. The main hypotheses will be briefly reviewed. 20
(1)
(2)
(3)
(4)
Vlasov's thin shells theory is used. 21

Ilyushin's yield rule 2 2 in terms of stress resultants is used to avoid
the necessity of an integration through the thickness.
The normality principle is used, formulating the flow rule in terms
of stress resultants and generalised strains, as suggested by
Kachanov. 2 3
The displacement field is represented by means of a complete
Fourier expansion; this is somewhat different from the hypotheses
of Spence, 24 as a consequence of the fact that the complete set of
loads (three forces, three moments) is considered by this chapter.
410
(5)
(6)
L. LAZZERI
An energy approach has been used, rather than formulating the

equilibrium equations.
The influence of large displacements on geometry and orientation
matrix is considered.
6.
APPLICATIONS
6.1. A Computation Example

A typical pipe geometry is given in Fig. 14. Even if the line considered is a
relatively simple one, it has some features which are typical of the problems
one finds during normal analyses.
The jet forces consequent on the various breaks have been computed by
means ofthe GETTO 2 code and are reported in Table VI. A preliminary
8 5
FIG. 14. A typical pipe geometry.

analysis has been performed in order to make a predimension of the
restraint model number; the characteristics of each model are reported in
Table VII (the type is the one depicted in Fig. 2). The characteristics of the
impact for each break in each specified restraint (frontal or lateral) are
specified in Table Vili and the resulting number of moduli for each
restraint is given in Table IX together with the critical break (i.e. the one
requiring the maximum number of moduli).
PIPE WHIP ANALYSIS
41 1
TABLE VI
JET FORCES
Break code
1
2
3
4
7
8
Key:
CV
CT
L
CV
CT
L
CV
CT
L
CV
CT
L
CV
CT
L
CV
CT
L
C)
(ms)
U)
(ms)
148
148
266
148
148
266
148
148
266
148
148
266
148
148
266
148
148
266
0
16
1
16
26
1
26
10
1
10
0
1
0
6
I
6
6
1
103
103
186
103
103
186
103
103
186
103
103
186
103
103
186
103
103
186
0
104
104
32
72
72
84
20
84
104
0
104
0
24
24
12
12
12
Fi
^3
(0
186
115
271
144
115
233
120
115
212
114
115
205
115
154
242
115
170
256
CV = circumferential breakage vessel side;

CT = circumferential breakage turbine side;
L = longitudinal breakage;
F, = force after time /,;
F2 = force after time ',;
F3 = force after time t2.
A certain number of breaks have been studied by means ofthe FRUSTA

code; the main results are given in Table X. For each break the following
data have been reported:
(1)
(2)
(3)
Conventional stop timei.e. the maximum time for which the

phenomenon has been studied; it is generally assumed to be the
time for which inversion of the velocity has been reached both for
the node to which the jet force has been applied and for the nodes on
which the loaded restraints have been placed.
Distribution of energy; the percentages of kinetic, pipe and
restraint energy have been reported at the 'conventional time'.
Restraint propertiesfor each break the loaded restraints are
specified with the indication ofthe maximum force and the 'usage
412
L. LAZZERl
TABLE VII
RESTRAINT PROPERTIES
Xc= 0-15m
XE= 0151 5m
F A O = 2 - 9 x 10 7 kg/m
KKI = 1-3 x 10 5 kg/m
frontal
frontal
for each unit
for each unit
0-48 m
0-481 5 m
lateral
lateral
TABLE VIII
RESTRAINT CONDITIONS
No.
Circumferential
Turbine side
Vessel side
I
2
3
4
7
8
2F
418F
8F
81-
IF
3F
3F
3F
Longitudinal
IL
3L
5L
5L
7L
2L
4L
6L
6L
8L
TABLE IX
TENTATIVE DIMENSIONING OF RESTRAINT
Resi taint
No. modular
units
Critical
break
3
4
5
5
5
6
3
3
7
7
factor' (i.e. the ratio ofthe actual to the maximum force for the total
number of moduli) is given. However, it should be mentioned that a
fully elastic response corresponds to a usage factor of roughly 0-63.
The allowable values are determined under the assumption of a
maximum strain of 25 %.
It is fairly obvious that in terms of energy the margins are somewhat
higher than those which can be derived from the consideration ofthe 'force
TABLE
MAIN RESULTS OF DETAILED F R U S T A A NA LYSIS
Break
identification
Conventional
situiti
Resinimi
Enei gy total ( %)
Pipe max.
time
( 10Fv;
(xl0~2)
2L..3
40
163
12
2C..J
3L 2 . 8
29, 8
53
125
189
2
14
5L3.8
88
6 91
5C 8 9
No.
No.
moduli
Max. force
(xiO5 kg)
38
50
26
35
72
51
(11
707
292
439
oval.
0
0
91
745
9
255
1
2
8
3
4
5
6
8
5
6
5
6
7
8
5
6
5
6
6
6
4"
5
5
6
6
6
6
6
6
6
6
6
5
5
5
5
342
335
256
309
290
355
0
268
282
370
0
0
355
2 61
313
82
81
92
90
82
86
0
65
68
90
(1
0
86
63
925
23
33
68
96
Dformt ion
6L
3.8
95
98
8L
3,9
75
oval.
01
776
223
5132, 8
924
719
71
2')
63 8
> 100
oval.
745
255
JL
Usage
(%)
Restraint
Kinetical
"0
>
>
r
<
j.
w.
" This value was used because such a number was required by break 2C.
U)
414
35
05
-o-
15
05 0 5
f S
(d)
FIG. 15. (a) Geometry of case 2L13. (b) Geometry of case 2C12. (c) Geometry of
case 3L28. (d) Geometry of case 5L38. (e) Geometry of case 5C89. (f) Geometry of
case 6L38. (g) Geometry of case 8L39.
usage factor'; in fact, a usage factor of 0-9 corresponds to an 'energy usage
factor' of approximately 0-6.
In all the cases the predimensioning resulted in a somewhat conservative
evaluation ofthe number of moduli for each restraint; this was due to the
fact that in all the cases it was deliberately supposed that most ofthe energy
is absorbed by the restraint. This hypothesis is obviously untrue whenever
the break is quite near a fixed end, and tends to be verified for the breaks far
apart from them.
415
ZT
5
3^VVVf^
3 5
fr
0 5
15
0 5 0 5
8 ^ 9
- O
co
o
(f)
(e)
08
11
11
(g)
FIG. 15conici.
It should also be mentioned that in all these cases the maximum strains
are localised near the ends, so that a decrease in the number could cause an
increment both in the pipe strain and in the restraint strain. This is shown by
cases 5L38 and 6D38, where the run was repeated assuming five moduli for
both restraints; an increase in the pipe strain resulted in these cases.
In both cases a reduction ofthe restraints moduli is clearly possible (see
Tables IX and X); so from the tentative solution a reevaluation is made on
the basis of results of the FRUSTA B3 code, and the final solution is
obtained with the maximum number of restraints which are compatible
416
L. LAZZERI
TABLE XI
GENERAL CHARACTERISTICS FOR CASE A
Pipe diameter: 670mm

Pipe thickness: 29 mm
Total pipe length: 14-500mm
Anomalies: element 2 represents a valve (rigid)
(10 t weight)
Restraints
1 on nodes 4-5 FK0 = 0-85 x 105 F) = 0-80 x 104 kg/cm

2 on node 9 FK0 = 0-85 x 105 FK\ = 1 x 104 kg/cm
* g = 78mm X,,= 125 mm
Forces
1 < 2-6 ms, force = 4701
2-6 < < 4 9 , force = 430t
49 < t, force = 3501
with the applied loads and the allowable values of restraint force and strain
on the pipe.
As far as the residual kinetic energy is concerned, it is generally quite low,
apart from the 3L28 case, where the residual kinetic energy is relatively high
and further consideration of the phenomenon is necessary. This was true
also for case 8L39, where the phenomenon was followed until practically no
kinetic energy was present in the system. However, the usual criterion
previously described is generally satisfactory and rebound phenomena are
not usually very important; the somewhat irregular behaviour of these two
cases is due to the large inertia of a valve which retains large kinetic energy
even at low velocities.
The models used in the FRUSTA code analysis are shown in Figs.
15(a)-15(g), while the detailed results for each model are shown in Tables
VIIIX. Figures 16(a)-16(c) show the main features of case 8L39; the
energy percentage, node displacements and restraint force are given as a
function of time.
6.2. Further FRUSTA Analysis
6.2.1. Example A
As a first example that given in Fig. 17 is presented. The break is assumed
to take place at the end of a large valve placed on a 660 mm diameter, 29 mm
thick pipe. In the model the pipe is restrained at three locations on one side
of the valve; no room was available on the other side to place another
restraint. The data on pipe and restraint characteristics are given in Table
El kinetic energy
E2 pipe energy
E3 restraint energy
7 node displacement
(10"1m)
(b)
5.5
7.5
time ( 1 0 - 2 sec )
FIG. 16. (a) Energy for case 8L39. (b) Node displacements for case 8L39.
418
L. LAZZERl
force
(10 5 kg)
le)
J_
-L.
7
time U0~ 3 sec)
FIG. 16. (c) Restraint forces for case 8L39.

XI ; it should be mentioned that the restraint located near the break has been
lumpedon two successive nodes nearby. This was done to take into account
the relatively large axial size of each restraint unit.
The time sequence of events is as follows:
(1)
After 16-3ms contact is reached on the restraint located at node 4

and a further contact is reached after 18-2ms on node 5. Contact
takes place at slightly different times on the different units of the
restraint; it should be noted that large yielding is present on
elements 1 and 2. The displacements after 20ms (i.e. just after
contact) are shown in Fig. 18(a).
4
io
10
FIG. 17. Geometrical configuration for example A.
419
PIPE WHIP ANALYSIS

3
10
(a)
125 mm
FIG. 18. Displacements of the pipe at various times for case A.

(2)
(3)
After 32-8 ms ovalisation begins at one end (node 0) and propagates

for a few subelements until a plastic hinge forms on the same node
after 47-2 ms. Unloading rapidly occurs and the moments on 1 and
2 rapidly decrease an order of magnitude; the displacement
pattern is shown in Fig. 18(b) after 50ms, i.e. a few ms after the
hinge formation time.
The transient ends after 85 ms, when inversion of the velocity is
obtained on the node where the force is applied and on the nodes of
the restraint. The geometrical situation is depicted in Fig. 18(c).
Finally, one should observe that:

(a) The total energy given to the system is distributed as follows : 47-5 %
on the restraint; 51-5% on the pipe; and 1 % residual kinetic energy.
Furthermore, the first element where a hinge is formed absorbs
roughly 30 % of the total strain energy, while the remaining part is
absorbed in the restraint zone. It is then obvious that the correct
simulation of energy during and after the development ofthe hinge
is a decisive factor in assessing the total behaviour and that
simplified hypotheses (a hinge from the beginning) would yield
incorrect results.
420
L. LAZZERI
(b)
(c)
The behaviour ofthe system is such that the absorption of energy is

different on the two restraints located on nodes 4 and 5; in fact,
more than 70 % of the total restraint energy is given to the restraint
closer to the break location. It reaches a maximum force of 83 % of
the allowable, while the other one is roughly 62 %. This makes clear
that a very refined model may be necessary in the predimensioning
phase.
The difference in the displacement pattern between Figs. 18(a) and
18(c) in the zone around mode 0 (the tangent changes inclination
from horizontal) clearly shows the hinge formation mechanisms.
6.2.2. Example
The case depicted in Fig. 19 is examined. A flat surface is supposed to be
present at node 2 and a further restraint is applied split between nodes 4 and
3
I4
^JL^.
FIG. 19. Geometrical configuration for example B.

5 (a total of six modular units was supposed; three were lumped at each
end); furthermore, a valve is present and is simulated by element 7.
The pipe is 609-6mm in diameter and 25 mm thick, and a critical
crushing force of 34-4 ton was assumed (this value is quite irrelevant and its
only use is to show the performance of the program). Typical data are given
in Table XII.
TABLE XII
GENERAL CHARACTERISTICS FOR CASE
Pipe diameter: 609-6mm

Pipe thickness: 25mm
Total Pipe length: 13-60m
Anomalies: element 7 is a valve (rigid) (0-41 weight)
Restraint 2 simulates the crushing behaviour of the pipe 1 on nodes 4 and 5
Forces
t 1ms, 2651
,72 ms, 1861
r72ms, 233 t
421
PIPE WHIP ANAL YSIS
20 m sees
4 0 msecs
54msecs
1 m1
JlOOr
FIG. 20. Displacement of the pipe at various times for case B.

The time sequence is as follows:
(1)
(2)
(3)
After 18-9ms impact on node 2 occurs and the crushing force is

overcome in a few ms ; a hinge is conservatively assumed to form on
node 2; the geometrical situation after 20 ms is shown in Fig. 20(a).
Impact on restraint 4/5 takes place after 325-36 ms, while an
inversion of velocity on node 2 causes disengagement between pipe
and restraint ; the geometrical situation is depicted in Fig. 20(b) and
refers to a time of 40 ms after impact.
The transient is supposed to stop after 54 ms, when the situation is
the one depicted in Fig. 20(c). In this case the formation of a hinge
on node 2 is evident.
The following conclusions may be drawn:

(a)
The total energy given to the system is 59-5 % on the restraint, 37 %

on the pipe and 5 % residual kinetic energy. Most of the energy is
RUNS
Type
tJ
io
Load P ercentage
o
.
"
simili /led
no
pressure
K3/emi
150 ru//cmi
Cp
eo -
R- OOO m
t m
r
>
Ky
Sx f<2
f > o.rtx*e>
m ,
-1
so
30
Displacement
FIG. 21.
(cm)
Runs type influence of pressure on defleclion.
N
N
rr
7:
PIPE WHIP ANALYSIS
(b)
423
given directly to the restraints on nodes 4 and 5 (65 % on restraint

on node 4 and 33 % on restraint on node 5), while energy absorbed
during the crushing is (possibly conservatively) very low.
It should further be considered that after the hinge on one half is
created, the total force is Fjet m3X3. and is directly absorbed by
the pipe part where crushing did not occur.
6.3. PAPS T2G Analysis

Three types of runs have been performed in order to study some
particular effects. They are:
Runs to study the influence ofthe pressure on the lateral deflection
characteristics.
Runs to study the particular behaviour of the pipe part overhanging
from the restraint.
Runs Mto study the influence of the model on the deflection
characteristics.
In all the cases the geometry which has been considered is a cantilever beam,
with various end loads. Obviously the PAPS T2G code can handle much
more complicated situations; however, a comparatively simple one has
been selected in order to enable comparison with simpler models and for an
understanding of the phenomenon.
Some runs have been depicted in Figs. 21 and 22. The general results can
be summarised as follows:
(1)
(2)
(3)
(4)
The pressure does not seem a decisive parameter, at least if it

produces hoop stresses as usual in the pipe design.
The shear stress may be responsible for some deformations and
displacement of the pipe, particularly in the overhang portion,
where a shear hinge might develop, even if in rather extreme cases.
The simplified model generally produces a reasonable, somewhat
underestimated evaluation of the pipe deflections and strains.
The use of a more refined element should be confined (for obvious
cost reasons) to cases where either geometry or load would produce
an abnormal situation (large pressure, large shears, large torque, a
shear torque, a shear force changing its direction, etc.), so that the
use of both the simplified and the complete elements would be
profitable.
IO
RUNS
TV/SC
Case P rete
-I
2
V 3
load Percent \
Shear Torque Bend
tD
O.IS
tO G /3
tO
O.IB
o simplified (pure
I*
v*
OSO
O
O
Oto
bending)
i
? - O. das m
t O.O m
iC3}t **
fi0.43*10*fy/m'
m o.ie>
axial
/
.--.
-~
mixed model
Di*picement (cm)
FIG. 22.
Runs type influence of secondary loads on deflection.
>
N
N
PIPE WHIP ANALYSIS

7.
425
CONCLUSIONS
A general discussion of the pipe whip problem has been given. Basically
three types of codes are available:
(1)
(2)
(3)
Very simple ones, generally based on an energetic a p p r o a c h , or

similar concepts, for a preliminary evaluation of the restraint types
a n d geometry.
Special codes, based on an ad hoc formulation.
G e n e r a l finite elements codes.
T h e use of o n e or other type of code d e p e n d s strongly on the type of

problem one is facing; of course, the use of a m o r e sophisticated code
produces larger costs in terms of personnel and c o m p u t i n g time.
REFERENCES
1. Irwin, G. ( 1957). 'Analysis of stress and strain near the end of a crack traversing
a plate", J. Appi. Mech., 24(3).
2. ASME 3, Nuclear Components, 1976.
2. Wessel, Clark and Pryle. 'Fracture mechanics technology to heavy section steel
structures', 2nd Conf. Fracture Mechanics, Brighton, Paper 72/VI.
4. AEC Regulatory Guide 1.46, US Regulatory Commission.
5. Moody, L. F. 'Prediction of blowdown thrusts and jet forces', ASME, paper
69-HT-31, 1969.
6. Esswein, G. A. et al. 'Systems criteria for protection against pipe break for the
mark III containment", N E D O 10990, GEC, September, 1973.
7. Dini, D. and Lazzeri, L.'An analysis ofthe dynamic elasto plastic behaviour of
a pipe during the pipe whip accident". 2nd SMIRT Conf., Berlin 1973, paper
F5/4.
8. Palusamy. S., Patrick, W. L. and Cloud, R. L. (1974). 'Dynamic analysis of
non-linear pipe whip restraints', Nuil. Eng., 31, 106.
9. Gerard, G. Handbook of Stability, NACA-TN-3783.
10. Bisconti, N.. Lazzeri. L. and Strona, P. (1976). 'Pipe whip analysis for nuclear
reactor applications', Nucl. Eng. Des., 37, 347.
11. Bathe. J. K. and Adira, M. 'A finite element program for automatic dynamic
incremental non-linear analysis', Rep. 884499-1, MIT.
12. Sanders, L. 'An improved first-approximation theory for thin shells', NASA TR
R-24.
13. Dini. D. and Lazzeri, L. (1973). 'Descrizione del programma di calcolo
FRUSTA', CNEN Rep. RT/PROT, 15.
14. Flgge, W. (1932). 'Die Stabilitt der Kreiszylindershaie", Ing. Arch., 3.
15. Hodge, P. G. (1959). Plastic Analysis of Structures, McGraw-Hill, New York.
426
L. LAZZERI
16. Vlasov, V. Z. and Leontiev, Y. (1960). 'Balki, Pliiie Obolochli na Uprugom

Osiovanii', Gosudarsvenoe isdatelisto-fisiko-matematicheskoi literaturii,
Moscow.
17. Johnson, W. and Mellor, P. B. (1975). Engineering Plasticity, Van Nostrand
Reinhold.
18. Timoshenko, S. P. and Gere, J. M. (1961). Theory of Elastic Stability,
McGraw-Hill, New York.
19. Larson. L. D., Stokey, W. F. and Franzen, W. E. (1975). 'An approximate
model for an elastic-plastic pipe element under combined loading'. Trans.
ASME, 97, Series 1.
20. Lazzeri, L. 'The analysis of elbows in the elastic and plastic field' (to be
published).
21. Vlasov, V. Z. 'General theory of shells and its applications in engineering',
NASA 64-19883.
22. Ilyushine, A. A. (1956). Plasticit, Eyrolles, Paris.
23. Kachanov, L. M. Foundations of the Theory of Plasticity, North Holland,
Amsterdam.
24. Spence, J. 'The creep bending of short radius pipe bends', 3rd SMIRT Conf,
London, paper F6/2.
25. Lazzeri. L. 'On the influence of multiple loadings on plastic pipe elements
applications to the pipe whip problem', 4th SMIRT Conf., San Francisco,
paper F3/2.
26. Lazzeri. L.. Giuliano, V.. Strona, P. and Scala, M. 'Analysis of piping in the
nuclear industry". Nuclex 78. Basel. Sect. D4.
27. Lazzeri, L. (1979). 'Applications of the plastic analysis to the pipe whip
problem". Int. J. Pressure Vessel Piping, 7(3).
28. Lazzeri. L. 'A model for a strain hardening pipe element', ASME paper 77PVP-47.
29. Dini. D. and Lazzeri. L. (1976). 'Modelling technique for the pipe whip
analysis', Nucl. Eng. Des., 37, 361-72.
30. 'Applications of MARC to pipe whip analysis', paper by MARC Corporation.
Minneapolis.
13
Material Behaviour and Modelling in Transient
Dynamic Situations
C. ALBERTINI, J. P. HALLEUX and
M.
MONTAGNANI

Italy
The present chapter deals with the experimental characterisation of

material response in transient dynamic situations and with the development
of numerical models to simulate strain-rate-dependent elastic-plastic
behaviour. The experimental techniques developed at the JRC-Ispra to
investigate the response of structural materials under high strain rate are
described in Part 1 by C.A. and M. M. On the basis of experimental
indications, numerical models are being developed to introduce the effect of
strain rate in the elastic-plastic constitutive relations employed in computer
codes. Part 2 (by J . P . H . ) describes such a relation and suggests an
extension to strain-rate dependence for fast transient problems.
PART 1 : Experimental Techniques to Characterise

Behaviour at High Strain Rates
1.
Material
INTRODUCTION
It has long been known that the mechanical properties of metals and
composite materials, such as plain, reinforced and prestressed concrete,
depend on the strain rate. Therefore, the computer codes for structures
subjected to dynamic loading require the knowledge of material
constitutive relationships which correlate the stress increments with the
strain and strain rate increments.
In order to give the experimental basis to the formulation of constitutive
relationships for metals, we have developed some techniques based on the
principle ofthe split Hopkinson-Davies bar 1 which permit the deformation
at high strain rate of a short specimen of the material.
427
428
C. ALBERTINI, J. P. HALLEUX AND M. MONTAGNANI
Theoretically the constitutive equation should be able to describe the

behaviour ofthe material under very different dynamic loading conditions.
Considering this requirement, the techniques developed allow us to
determine experimentally the response of the material subjected to the
following loading conditions: tests at constant strain rate; tests at constant
stress with rapidly changing strain rate; interrupted loading tests; and
incremental loading tests.
Once the material constitutive equation has been formulated on the basis
ofthe experimental results, its validity must be verified. To this end a special
experimental device has been developed 2 which permits the measurement
of the elastic-plastic wavefront as it travels along a long specimen of the
material studied. Comparison ofthe measured wavefront parameters with
the calculated ones gives a measure of the validity of the formulated
constitutive relationships.
With the devices described uniaxial tensile tests at strain rates ranging
between 10~ 2 /s and 103/s can be performed.
The applied dynamic loading will often create a multiaxial state of stress
in the structures. To cope with this situation, the above-mentioned devices
were further developed in order to carry out dynamic biaxial tensile tests on
cruciform specimens.
The previously mentioned test conditions can be independently realised
along the two loading directions of the biaxial testing devices, and the
nature of loading can be independently chosen tobe tensile or compressive.
2.
DYNAMIC UNIAXIAL TENSILE TESTS
2.1. Hydropneumatic Machine for Tests at Medium Strain Rates

A hydropneumatic machine for carrying out tests at strain rates ranging
between 10"' and 102/s has been developed, which can be set to perform
either constant-strain-rate tests or tests at constant stress with rapidly
changing strain rate.
For tests at a constant strain rate (Fig. 1) water is filled into the upper
tank, and both chambers are loaded at an equal pressure. After the
membrane has been punctured, the liquid flows out through a calibrated
orifice at a constant rate, thus imposing a strain on the specimen with
a constant strain rate. In fact, the pressure acting on the piston governs
the displacement velocity, because the load resulting from it is much greater
than the force needed for breaking the specimen. Stress measurements are
made with a strain gauge attached to the lower fixed bar holding the
MATERIAL BEHAVIOUR A N D MODELLING
CALIBRATED
ORIFICE
MEMBRANE
429
VALVE 1
^<<%^
HIGH
PRESSURE
GAS LINE
DISPLACEMENT
TRANSDUCER
STRAIN GAUGES
i:''"'///-.
FIG. 1.
ÊAR_TH
Schematic of hydraulic-pneumatic machine for dynamic tensile tests at

constant strain rate.
Membrane
Valve i.
^-j<
ûfjftW
Valve 1
BS r*
:..,..,..
FIG. 2.
Schematic of pneumatic rapidloading machine for dynamic constant

stress tensile tests.
430
C. A LBERTINI, J. P. HA LLEUX A ND M. MONTA GNA N1
AL 1mm/sq.
M&SHHK
*- 5 m s/sq.
116 MN/m2
FIG. 3. Hydraulicpneumatic machine record of a test at constant strain rate.
Higher trace: displacement record; lower trace: stress record, = 15s~'.
specimen, and strain measurements are accomplished with a transducer
measuring the displacement of a plate which moves with the piston.
This machine, when operated without the water control, allows tests to be
carried out at a constant stress where strain rates change rapidly. In this
case (Fig. 2) the upper tank can be filled with air and a much larger
membrane is used, so that the pressure above the piston drops very quickly.
This causes the specimen to be loaded by the constant force due to the gas
pressure below the piston and thereby to undergo strains at a constant
stress. In fact, during a test the increase in the volume ofthe air tank, the
pressure of which moves the piston, is negligible compared with the total
volume of the tank, so that the gas pressure can be maintained constant.
420 r^
b
-
LO
LU
cc
LO
5- - i n
Ê(t)
?Rn
s- u- -OR
z
<i*
210
IAO
6- -12
(0
IRO
CE
-td
J(t)
70
-0.6
80
120
TIME ( m sec)
LU
tr
J2- -0.4 <

CE
1--0.2
AO
LO
160
trfi Yield delay time

FIG. 4. Stress, strain, strain rate history in a test at constant stress on a mild steel
specimen. /d = yield delay time.
MATERIAL BEHAVIOUR AND MODELLING
431
The recording of a testat a constant strain rate is shown in Fig. 3; one can
observe that during plastic deformation the slope ofthe displacement-time
curve is nearly constant; therefore the strain rate is also constant.
Figure 4 shows the recording of a test at constant stress. In this case, when
the applied stress is lower than the ultimate tensile strength ofthe specimen,
it is possible to observe also the effects of interrupted loading.
Incremental loading tests can also be carried out, connecting the elastic
stress bar of the hydropneumatic machine to the moving cross-head of a
slow tensometer. In this case the specimen is deformed initially at a low
strain rate (10 _ 4 -10~ 2 /s), which is suddenly increased to a value of the
order of 102/s by superimposing the action ofthe gas piston on the action of
the slow cross-head.
2.2. Modified Hopkinson Davies Bar with a Prestressed Bar-loading Device
for Tests at High Strain Rates (10 2 -10 4 s _1 )
2.2.1. Principle ofthe Hopkinson-Davies Bar
The original Hopkinson-Davies bar is schematically shown in Fig. 5. It
essentially consists of two half-bars with the specimen introduced in
between.
A projectile strikes the first half-bar and generates a pulse having a
constant amplitude or duration, depending on the length ofthe projectile,
the pulse being transmitted along the first half-bar. On the assumption that
the pulse wavelength is long compared with the bar diameter, and further
that the pulse amplitude does not exceed the yield strength ofthe bar, it can
be assumed that the pulse will propagate along the half-bar at the velocity
C0 ofthe elastic wave, its shape remaining constant and in accordance with
the theory of a one-dimensional wave propagation in circular-cylindrical
rods.
When, after passing through the first half-bar, the incident pulse ,
reaches the specimen, part of it, , will be reflected by the interface ofthe
bar, whereas part will pass through the specimen, propagating itself into
the second half-bar. The relative amplitudes of the incident, reflected and
transmitted pulses will depend on the physical properties ofthe specimen.
Since the incident pulse is of long duration compared with the time of
passing through the specimen, numerous reflections of elastic-plastic waves
take place inside the specimen, rapidly permitting the creation of a state of
stress equilibrium along the specimen. It is on such a hypothesis that the
validity of the method is based.
Two strain gauge stations applied to each ofthe two half-bars are used to
measure the elastic deformations ,, and as functions of time, which
K)
BAR ELEMENT
>
II HALF-BAR TRANSMITTED
PULSE E T
r
OS
IT
>
r
r
rr
C
X
>
/.
S
STRIKER BAR
>
>
STRAIN GAUGES
FIG. 5.
Schematic of the split HopkinsonDavies bar.
433
were created in both half-bars by the incident, reflected and transmitted

pulses, respectively. The strain gauges are placed at a certain distance from
the end of the two half-bars, the distance being longer than the incident
projectile, so as to record the pulses without any interference caused by
subsequent reflections.
From the theory of elastic wave propagation in a bar under onedimensional stress we can deduce the relationships to obtain the stress,
strain and strain rate, using the measurements taken by the two strain gauge
stations. 3
If l0 and A0 are the initial length and cross-sectional area ofthe specimen,
the average strain . strain rate , and stress rjs in the specimen are:
Co f'
2C 0
T. ^
( -e, , - e- R - e"T. ) d F ^ -- ,
Jo
C
2C
4 = -~-(]- f:R - ) ~ 9-,,
'o
'o
, A
A
,= {E (, + 4- ) = E
%dF
(1)
II
(2)
(3)
The + signs differentiate compression or tension test, compression being

taken as positive. The approximations on the right-hand side of eqns.
( 1 )-(3) are based on the approximate equivalence of the forces on both faces
ofthe specimen, i.e. of stress equilibrium across the specimen. This requires
that
t +
ER
&r
(4)
This approximation can be experimentally verified by summing the

indicated signals. It is usually valid to within experimental accuracy for
metal specimens of small dimensions (/0 < 4 in.). 3
2.2.2. Description of the Modified Bar
As noted in subsection 2.2.1. the standard split Hopkinson-Davies bar
was developed for research in compression with the use of a projectile as a
loading device.
It was ascertained that tensional waves are developed in a vessel wall
struck by an internal explosive accident. Besides, the strain rates measured
on reactor models during explosion tests were found to be in the range
100-500/s. A dynamic test performed at a strain rate of 200/s, up to a
deformation of 50%, would require a pulse lasting 2-5 10 3 s. To
434

Earth
Loading mechanism by
hydraulic piston
Gripping jaws
Earth
Bar I
Strain gauge I
Strain gauge 2
Bar 2
FIG. 6.
Modified Hopkinson bar with prestressed bar-loading device.
generate a pulse of a similar duration, the standard Hopkinson bar would

have to be struck by a projectile 7 m long, which is practically impossible.
To solve this problem the device described in Fig. 6 was developed. 4
where the energy needed to produce the deformation of the specimen is
stored in the front ofthe first half-bar, which has been tensioned within the
elastic limit by a hydraulic piston. A robust bar-connected locking tooth
acting on a notched bridge of brittle material and connected to grounded
supporting pieces is adopted in order for tension to be created solely in the
front part ofthe first half-bar. This brittle bridge reaches its breaking point
without any appreciable previous deformation and therefore permits the
release of energy in an extremely short time lapse (25 ps). The brittle bridge
is shaped in such a manner as to facilitate the fracture along the centre-line
and the ejection of fragments.
As soon as the tension transmitted to the brittle bridge has caused it to
break, a rectangular-tensile wave will propagate, at a rising time of
approximately 25 ps. along the first half-bar; it will then reach the specimen
435
screwed to the two split bars, causing deformation and being partly
reflected and partly transmitted from there, as shown in Fig. 7. Strain gauge
1 records the incident wave , and reflected wave , whereas strain gauge 2
will record the transmitted wave . The device generates impulses having a
2-5 10~ 3 s duration; it is consequently suitable for strain rates of
100-1000/s. The two propagating split bars have the same length as the
prestressed tract, thus affording a wave train in the specimen for 2-5
10" 3 s without any superimpositions.
In order to vary the strain rate it is sufficient to vary the incident pulse
amplitude by changing the stress in the tension bar and, consequently, the
resistance of the brittle notched bridge used as a lock.
Each of the two strain stations is made of two strain gauges connected in
series and mounted at 180 from each other on the bar in order to cancel all
signals following bending ofthe bar, which cannot be avoided completely.
Both strain gauge stations are statically calibrated by the following
method: the two split bars are connected by a strong specimen; the
complete device is then brought into tension, which is measured by a
calibrated dynamometer.
2.2.3. Analysis of Results
The signals transmitted by the two pairs of strain gauges are recorded on
a double-trace oscilloscope (Fig. 7a). The higher signal is the recording of
the first strain gauge and is proportional to the deformation provoked by
the incident and reflected wave.
in fact, the first length in the figure corresponds to the incident wave, and
the second length, after the step, corresponds to the algebraic sum of the
incident and reflected wave. In our measurements the strain gauges are
usually at 75 cm from the test piece. The lower trace of Fig. 7(a) corresponds
to the signal of the second strain gauge and is proportional to the
deformation induced in the second half-bar by the transmitted wave.
By means of an electronic integrator the signal of the first trace is
integrated on a second oscilloscope (Fig. 7b). The first length, descending
from left to right, corresponds to the integral ofthe area determined by the
signal of the incident wave. The second length, with a lower slope,
corresponds to the integral of the sum of the incident and reflected waves.
By extending the first length of the curve (see the broken line) it is easy to
determine a length on the y axis for any given moment, proportional to the
deformation ofthe specimen, calculated on the basis of eqn. ( 1 ). From eqns.
(2) and (3) the strain rate and the stress can be also determined as a function
of time. The active length, l0, appearing in these formulas is corrected to
436
C. A LBERTIN!. J. P. HA LLEUX A ND M. MONTA GNA NI
61
,_.
Ak 100-10-6/sq.
200-10 _6 s/sq.
200 MN/m2
b)
200 KT 6 s/sq.
0-173/sq.
FIG. 7. Modified Hopkinson bar with prestressed barloading device, (a) Record
of test on specimen UQH 2 DIN 4306 steel, = 320s" 1 ; (b) integration of ,
and , cR.
SPECIMEN
437
Scale 4M
FIG.
8.
take into account the contribution ofthe two connections to the threaded
extremities of the specimen (Fig. 8). This correction is carried out by
checking the deformation of the specimen with a lattice printed on the
specimen and measured by microscope before and after testing. For the
correction the following formula was used, based on the conservation of the
volume ofthe deformed part ofthe specimen:
ih = i0 +
A! - /
/
(5)
where / = length including the two connections; /' = corrected active

length; l0 = nominal active length; and / 0 , Al = increments ofthe lengths
l0 and / caused by testing.
The signals transmitted by the two pairs of strain gauges were also
recorded on two transient recorders, digitised and stored on a digital tape.
The evaluation of the results was then performed with the help of a
computer.
2.2.4. Feasible Test Conditions
The modified Hopkinson bar, described above, is used to perform tests at
constant strain rate. Tests at constant stress with rapidly changing strain
rate are carried out by lowering the amplitude ofthe incident stress wave to
values of the order of the ultimate tensile strength of the specimen.
Interrupted loading tests are performed by fixing the length of the
prestressed bar so that the duration of the incident pulse deforms the
specimen only up to a fixed percentage; the introduction of a pulse shaper
438

EARTH
'/////////
LOADING MECHANISM BY
DIFFERENTIAL SCREW
FIG. 9. Modified Hopkinson bar with prestressed bar-loading device and with
pulse shaper for very short rise time.
(Fig. 9), which transmits only tensional waves, avoids reflected compressive
waves, successive to the incident tensile pulse, acting on the specimen.
Incremental loading tests are also possible connecting the second halfbar to the moving cross-head of a slow tensometer; in this case the specimen
is deformed initially at a low strain rate (10~ 4 -10~ 2 /s) which is suddenly
increased t o a value of the order of 103/s by superimposing the action ofthe
prestressed bar on the action of the slow cross-head.
2.3. Results
Tests at constant strain rate were performed on various austenitic
stainless steels (AISI 316, AISI 304, AISI 321, AISI 347) in virgin, welded
439
MATERIAL BEHA VIOUR A ND MODELLING
and irradiated conditions, using the devices described before. Eliminating

the time from the recordings of () and e(t), engineering stressengineering
strainstrainrate curves were obtained for deformations up to rupture. The
active length used for the calculation of strain and strain rate is that given by
eqn. (5). Considering that strain rate has no influence on the modulus of
elasticity, 1,6 ' 7 the various slopes of the elastic part of the stressstrain
diagram obtained with the different devices (Instron, hydropneumatic,
STRAIN
G = E A L = E'
AU
L
L
CORRECTION FOR STRAIN VALUES
k'
L
AL
L "
m)
FIG. 10. Correction of stressstrain curves to the slope of static elastic modulus.
Hopkinson bar) were then corrected to the static elastic modulus. This
means a correction of the values of strain, performed as shown in Fig. 10, in
function of stress. This correction takes into account the contribution to
deformation measurements from the connection pieces between specimen
and test machine. A s mentioned in subsection 2.2.1, the experimental
elastic modulus determined with the Hopkinson bar is not exact, because of
the finite time the specimen takes to arrive at stress equilibrium along its
length, through several wave reflections.
The results obtained on virgin AISI 316L tested at ambient temperature,
are reported in Fig. 11, where one observes that with increasing strain rate
the flow stress at a given strain increases and fracture strain decreases.
Some tests at constant stress and rapidly changing strain rate were
performed at ambient temperature on the same steel, in order to study the
influence of continuous change of strain rate during one loading cycle on
the stressstrainstrainrate relationships. Taking instantaneous values of
440
eoo
600
/ "',--
f/.-
'
400
TEST TEMPERATURE 20 C
CURVE STRAIN RATE
200
sec-1
m-2
n .
Ut.
420
n
30
45
60
75
90
ENGINEER
FIG. 11.
STRAIN'/.
Stress-strain curves for AISI 316 L stainless steel, virgin collaboration

Euratom-C.N.E.N.
20
FIG. 12.
30
40
50
60
70
ENGINEER. STRAIN 7.
Stress-strain curves for AISI 316 L stainless steel, virein.
MATERIAL BEHAVIOUR A N D M O D E L L I N G
441
stress, strain and strain rate from the recordings of these tests (Fig. 4), we
have constructed part of the flow curves of AI SI 316L for three strain rates.
From Fig. 12 it is possible to compare the short flow curves obtained with
tests at constant stress and the flow curves obtained with tests at constant
strain rate.
Making the comparison at strain rates of the same magnitude, one
generally observes a good agreement. This means that imposing two
different dynamic deformation histories on the virgin material yields the
same response.
2.4. Verification of a Material Constitutive Law
Knowledge of the response of the material to the various dynamic
loading conditions enumerated in the introduction might permit the
formulation of a material constitutive law in the general form
=/(,)
In order to verify whether the material constitutive law is reasonably
formulated, a special device based on immediate measurements of the
elastic-plastic wavefront, as it travels along a long specimen, has been
developed. This device (Fig. 13) in its actual form consists of a prestressed
steel strip with high yield strength, similar to the bar described in subsection
2.2.2. Once the brittle intermediate piece is broken, the stress wave
propagates into another strip made of the material to be tested and brazed
on to the former.
Along the test strip several strain gauge stations are positioned to
measure stress and strain. Further measurements of strain are performed
by a fast camera, the test strip having been previously marked. The strip is
pretensioned by a small load to keep it straight, without any fold or bend.
TESTING STEEL RIBBON
INTERM PIECE |
BRAZING
230-750_p
HIGH STEEL RIBBON [ i
SG2
S60 SBIj t s g 3 SG4/SG5 SG6 5G7 Oj .FIXATI0N
BRITTLE
HYDRAULIC
PISTON
4580
170
670 1
900 _|
3070__
5710...
11150
12660
13660
FIG. 13. Testing device for long specimens.
442
SG2 1 1 ^ 5 0 - 1 0 ~ 6 A
SGi.
500 yusec/sq
-SG6
-^SG7
FIG. 14. Test results on long specimens of AISI 304 steel.

The end ofthe test strip is fixed to a metal block which reflects the incident
wave with the same sign.
Some results are presented in Fig. 14, where strain gauge 2 gives the
incident pulse on the impacted end and strain gauges 6 and 7 show the
doubling of the stress due to the reflection at the fixed end of the strip.
Hence, the experiment is suited to measuring the deformation wave
propagating along the specimen.
The propagation of the deformation wave can also be calculated by
solving the set of equations:
cax
ox
dv
cx
dv
=
at
ce
et
/(,)
,
.
(equation of motion)
(equation of continuity)
(constitutive equation)
where .vis the Lagrangian coordinate measured on the unstressed specimen.

The boundary conditions are measured experimentally at the impacted end
of the strip; the conditions for a one-dimensional wave propagation are
realised.
Comparison ofthe calculated deformation wave and the measured one
gives a measure of the validity of the proposed material constitutive
equation.
The measurement ofthe first elastic-plastic wavefront as it travels along a
long specimen enables us also to better determine the first part of the
dynamic stress-strain curve corresponding to the transition from elastic to
plastic flow than with the Hopkinson bar. In fact, in the Hopkinson bar the
process of wave reflections, which produces a uniform stress distribution in
443
Section A - B
52x52
46
40
16.7
0.7
0.7
10
7Z&Z
22
Wa
r=3,
1A-rSlllIL
FK. 15.
Biaxial specimen for dynamic tensile tests.
- CM
r-
2 B ^
444
the short specimen, takes a short finite time which affects by a degree of
uncertainty the first part of the flow curves.
3.
DYNAMIC BIAXIAL TESTS
The stress, strain and strain-rate diagram under dynamic biaxial loading is
obtained by a machine which is capable of loading the four arms of a
DISPLACEMENT TRANSDUCER
WATER
PISTON
DETONATOR
MEMBRANE
ORIFICE
FIG. 16.
Biaxial dynamic tensile testing machine.
445
cruciform specimen (Fig. 15) simultaneously, up to rupture, along two

orthogonal axes. On each axis two gas pistons, the displacements of which
are controlled by the flow of a liquid through an orifice, exert the loading in
tension or compression (Fig. 16).
The problem of synchronisation of incident stress waves along the two
directions is solved by explosive detonators. The strain rate, the magnitude
of loading and the tension or compression state can independently be
chosen along the two axes. The machine can be set to perform tests at
constant strain rates ranging between 10" ' and 10 2 /s. Furthermore, it can
be used to perform tests at constant stress with rapidly changing strain rate,
interrupted loading tests, incremental loading test's, by adopting techniques
LOADING
PRESTRESSED
SHOCK
DEVICE
CABLE
ABSORBER
FIG. 17. High load dynamic biaxial testing machine. Load = 3MN; pulse
duration = 0-04 s; rise time = -200 10"6 s.
446
similar to those described in subsection 2.1 for the uniaxial hydropneumatic

machine.
For non-homogeneous materials it is required to perform tests on larger
test sections. This is needed for complex structures, such as the reactor core
internals, and also for composite materials, such as plain, reinforced or
prestressed concrete.
Also, for materials which can, in general, be considered as homogeneous
(e.g. steel) it is in some cases required to perform tests on large test sections,
when the original material (e.g. the plate for the construction of a vessel) is
to be tested without alterations due to machine tool working for the
construction of the reduced size specimen, or when it is required to test a
welded plate of the original dimensions.
For the performance of tests on such large specimens a high-load biaxial
machine (Fig. 17) is now in an advanced phase of design. It is based on the
principle of the Hopkinson bar. with a prestressed bar-loading device,
characterised by four arms at 90 in which each arm has a length of 100 m. It
is intended that this machine will allow impulse loads to be obtained in
tension or compression with a maximum force of 3 MN and a duration of
0 0 4 s, with a rise time of about 200 ps. Thus, the machine should produce
strain rates of up to 102/s on large specimens.
Using such an apparatus, we intend to verify computer codes for
dynamically loaded structures by performing simple model tests which
enhance the influence of material constitutive equation, wave propagation
effects and overloads due to stress concentration caused by wave reflections.
PART 2:
Numerical Models to Simulate

Strain-RateDependent Elastic-plastic Behaviour
4.
INTRODUCTION
Structural transient dynamic behaviour is generally produced when

structures are submitted to impact loadings (crashes, explosions,
earthquakes, etc.). Often in such environments large deformations occur:
large displacements, large rotations, inelastic material behaviour, large
strains. Important inertia effects have also to be taken into account. A
multitude of difficulties are associated with numerical simulations of these
phenomena. However, since about 10 years ago, investigators have
attempted to design computer programs able to predict, through numerous
447
simplifications, what happens in an impact-loaded structure. Efficiency of

the various available programs depends very closely on the specific problem
they are used for and, in particular, on the time scales of physical
phenomena.
We are not going to explore here the transient structural simulation
methods (for a proper classification see other parts of this course, or ref. 25),
but to try to establish a formulation ofthe structural dynamics equations
which poses the problem of the determination of constitutive relations in an
easily understandable way for common engineering sense and which can be
used as a conceptual basis for most of the present computational methods.
Thus we would like to trace the logic which brings the structural engineer
from the problem posed in physical terms to a sensible system of partial
differential equations which he can then solve by a method suiting his
experience and his taste. We shall insist on the material modelling and to
this end we shall describe with some detail how to treat metallic materials,
often encountered in practice.
It is well known that matter is a highly discontinuous assembly of
energies governed by complex fundamental laws studied in the framework
of quantum mechanics, statistical mechanics, particle theory, etc. One
could think of trying to establish macroscopic laws by suitable integration
of these microscopic laws. This is still illusory, for various reasons: the
mathematical complications involved seem insuperable; some fundamental
laws themselves and especially their interconnection are still a subject of
discussion; and tentatives on simple configurations often fail in
representing experimental results obtained on a macroscopic basis. So one
chooses a more realistic approach: continuum mechanics. 1 2 1 3 It is based
on a field concept : matter is indefinitely divisible without losing its defining
properties. Continuum theories represent 'direct experience' without trying
to explain them by the previously mentioned fundamental laws. Such
theories are called phenomenological. The aim is to set up a mathematical
model which represents in an approximate but suitable way the response of
structures (geometrically defined pieces of known materials) to loading (e.g.
forces acting on specific portions of the structure). This mathematical
model will contain general field equations based on principles of structural
mechanics and specific constitutive equations defining the materials used to
build the structure. It might seem at first glance that any form of
constitutive relation could fit the theory: this is, of course, not true and a
careful analysis shows even that it is not at all easy to construct a 'consistent'
constitutive relation.
In the range of very small strains metallic structures exhibit a linear
448
behaviour governed by the infinitesimal theory of elasticity. The classical

concepts underlying this theory and the definition of deformation, strain,
stress, etc., are well known and can serve as a basis for small (moderate) and
large strain theories. By small strains we intend to refer to those problems
where large rotations and non-linear material behaviour (plasticity) occur
but where the strains remain bounded. Full large-strain plasticity
formulations 28 still present serious difficulties, mainly due to the lack of
reliable experimental results. Key, 2 6 'Only a small handful of papers
actually contain working finite strain plasticity models.' (ref. 26, p. 263) As
it is not our intention to fall into ' . . . a number of articles which, while
appearing profound, are based on precious little...' (ref. 26, p. 263), we
choose a small strain, large rotation theory. In practice this theory works
well for strains of up to 5 % (sometimes 10-15 %), which is already largely in
the plastic regime. It should be pointed out, however, that for the type of
problems we intend to treat (impacts) this 'small strain hypothesis' is often
too restrictive and it is of importance at least to check its validity a
posteriori.
We shall restrict our attention to metallic materials which exhibit an
elastic-plastic behaviour (steel, for example). Impact loadings produce not
only large strains but also other complications: cyclic loading, strain-rate
effects, eventual temperature gradients, etc. General theories involving
these effects present considerable difficulties: ' . . . such theories exist only in
the minds of the most incurable optimists'(ref. 21, p. 59). We shall try here
to include some strain-rate effects. Another difficulty lies in the frequent
anisotropy ofthe materials: here we shall only consider isotropic material
properties.
For review, discussion and bibliography of these points, consult ref. 20.
As compared with statics, the dynamic equations of equilibrium have to
take into account forces needed to produce the motions of and in the body.
There are two main types:
(1 )
(2)
Inertial forces which are proportional to density and accelerations

and which produce no dissipation.
Damping forces which are usually of a viscous type that is
proportional to the gradients of the velocity field and various
parameters. These forces produce dissipation (mechanical energy
converted into thermal energy).
In transient structural dynamics the inertial forces are often far more
important than the damping forces. One means by this that during the part
of interest ofthe problem treated (e.g. analysis of structural resistance of
449
reactor components to accidental power excursions) the damping terms are

negligible as compared with the other terms of the equilibrium equations
and we shall not consider them.
5.
FIELD EQUATIONS
5.1. Kinematics
When a continuum undergoes deformation, the particles (material
points) of the continuum move along various paths in space. This motion
may be expressed by equations of the form
x = x(X,f)
(6)
which give the current positions of the particles as functions of the

positions X in a reference state: a material point is carried from a spatial
position X in a reference state at time t' to a spatial position at time t. X are
the material coordinates and the spatial coordinates. This formulation,
where X are the independent variables, is called material description (often
referred to as Lagrangian description). X and are usually measured in a
rectangular system of axes. We shall suppose that the Lagrangian
coordinates X are the spatial coordinates ofthe undeformed state (reference
state = undeformed state, and t' = t0); this is referred to as total or strict
Lagrangian description.
A unique displacement vector can be defined for each particle:
u(X, ) = x(X, t) - X
(7)
It can be shown that local deformation is completely described by Green's

symmetric deformation tensor:
i-a = i(ui.j + uj.i + uk.iuk.j)
(8)
where , holds for derivative with respect to Xm.

The quadratic term of eqn. (8) contains the large-strain information and
eliminates the rigid body modes due to rotation. As we limit our analysis to
small strains, this term vanishes if we explicitly apply an orthogonal
transformation of coordinates (we change the axes), eliminating rotation:
this is called corotational formulation (sometimes referred to as convected
coordinates formulation). It is of great interest in non-linear small-strains
transient structural dynamics. 1 6 2 4
450
C. A LBERTINI, J. P. HA LLEUX A ND M. MONTA GNA NI
5.2. Kinetics
The establishment in the Lagrangian description of the equilibrium
equations is rather awkward. This is due to the 'Eulerian' character ofthe
stresses.
If xis the spatial position where the stress is experienced and we choose
as independent variable (spatial or Eulerian description), one easily shows
that the equilibrium equations are:
cj.j+f=0
cu = cM
and
(9)
where c,} are the true stresses (Cauchy stress tensor);/,' are the internal
volumetric forces (including intertiai forces); and _} designates the derivative
with respect to x. Now, transcribing these equations to a Lagrangian
description involves the Jacobian matrix ofthe transformation between
and X and requires a choice for the stress tensor. Depending on the stress
definition, various forms of equilibrium equations are obtained.
Unfortunately, none of them is simple. The most convenient for our
purpose seems to be
(o,j + akjuLk) j + F, ~ 0
and
,} =
(10)
where , is the second PiolaKirchhoff stress tensor, obtained through

ij
i,j.n
nin
F, = Jf are the internal volumetric forces (including inertial forces);

y = \xk.J is the Jacobian of the transformation between and X; and
j denotes the derivative with respect to X.
Remembering that strains are assumed to be small, one can interpret the
Jacobian matrix as a pure rotation, and the second PiolaKirchhoff stress
tensor appears as a simple rotation of the true Cauchy stress tensor: it is a
measure for true stress, expressed in the material description. This has a
great advantage: it links the second PiolaKirchhoff tensor to Green's
deformation tensor in the same way as stress and deformation are linked in
the infinitesimal theory. 27
6.
PROBLEM DEFINITION
As mentioned before, we do not consider discontinuities of the field

functions (shocks) and all field equations are differential. No 'jump'
conditions are needed. Boundary conditions (and initial conditions) are
451
supposed to be given in a suitable manner and for our purpose do not

need further comments.
Physical phenomena occurring in a purely mechanical continuum obey
conservation laws: linear momentum, angular momentum and mass. The
equilibrium equations (10) represent linear momentum and angular
momentum conservations. The definition of the deformation tensor (8)
implies some continuity assumptions independently of conservation laws.
Mass conservation has not been considered yet; the reason is simple: the
form in which the equations have been written does not explicitly show a
time derivative, and if one forgets the inertial forces in which the time
derivatives are hidden, the above formulation is immediately applicable to a
problem where no time dependence exists (static formulation). So far,
therefore, our considerations are valid, as well for static as for dynamic
problems. However, eqns. (10) deserve some critical attention.
Let us focus on timedependent behaviour. Expliciting the inertial term,
eqns. (10) become:
pJ, = (a,j + ajku,k)j + F?; a,s =
(11)
where is the current density; and " denotes the second time derivative and
Ff the internal volumetric forces. Mass conservation is expressed by
pJ = Po
where p0 is the initial density (in the reference state) and finally
p0 = (a,j + aJkuLk)_j + Ff
,1 =
(12)
(13)
and, as expected in a material description, mass conservation is a trivial

matter, u appears still as a fundamental dependent variable but requires the
determination ofthe acceleration followed by a double integration in time
(this is particularly evident in the framework of a finite element
displacement formulation with explicit time marching).
This shows that, if from an engineering point of view it seems that
dynamic behaviour is simply obtained by adding some suited forcing terms
to the static equilibrium equations, the fundamental aspects of the
structural problems treated are drastically different and require specific
methods of solution.
We have 6 independent equations, (10) in the static case, (13) in the
dynamic case and 12 unknown variables (u,a,7). Deformations are
explicitly defined by eqn. (8) and density does not appear as an unknown.
Six equations which can no longer be derived from pure field concepts (we
exhausted them already) are missing for a proper definition of the problem.
Until now no distinction has been made between various types of materials.
452
The characteristics of the material will be brought into the formulation

through appropriate constitutive equations relating local effects (defor
mation) to local causes (stress). Considering the structure of the field
equations, a suitable form for the constitutive equations is
ij = Hijklakl
(14)
This relation is inconvenient in the classical 'displacement' formulation

widely used in finite element methods, where one needs the inverted
relation:
ii = H7jk\zk,
(15)
Though some problems about the equivalence of eqns. (14) and (15) can
arise owing to singularities, we suppose that H~ ' exists. When the material
has a non-linear behaviour, one prefers to write
dau = Pijkldekl
(16)
where d stands for a small increment between two following configurations

(at times t and t + dl in a dynamic problem).
Noting the symmetry in zi} and , P,jkl can be constructed symmetrically
so that eqn. ( 16) contains six independent equations, exactly what is needed.
Summarising:
(1)
(2)
In the Lagrangian description and with the choice made for the
stress tensor, an observer follows always the same material points
and is able to measure in an understandable way both stress and
strain in a structure subjected to loading ( = a macroscopic
experiment): he is able to describe the behaviour ofthe material he
is 'sitting on'that is, to derive P,Jkl.
We have a complete system of equations to tackle non-linear
material behaviour and arbitrarily large rotations in a structural
problem provided that the strains remain small: if one knows the
expression of PiJkl for the material used to build a geometrically
defined structure, one is able to calculate the stresses and strains
arising in the structure when it undergoes loading.
7.
CONSTITUTIVE RELATIONS
7.1. General Comments

The basic laws of structural mechanics (conservation of mass, linear
momentum, angular momentum) are valid for all types of materials and
453
lead to the field equations : they characterise the structure, not the material.
As a matter of fact, it is clear that for these relations also a certain
constitutive choice has been made concerning the material. For instance, it
was mentioned that the material must be continuously divisible without
losing its defining properties. However, one prefers to look at the field
equations as the fundamental mathematically derived equations of a
theoretical discipline (structural mechanics) dealing with the description of
problems involving various subclasses of'structural" materials, each of them
being defined by its constitutive equation. This explains that constitutive
relations have to fulfil various conditions in order to be consistent with the
general theory they fit.
In order to ensure reasonable solutions to correctly posed problems,
some mathematical principles have been studied for a long time. ' 2 1 3 They
can be used to help in formulating proper constitutive equations, but until
now no exhaustive reliable procedure for 'constititutive relations
construction' exists. It is, of course, not our purpose here to enter into the
axiomatic foundations of constitutive equations.
A constitutive equation defines an ideal material and therefore is a
mathematical model for a particular class of materials encountered in
nature rather than an exact representation for a specific unique material.
Thus, in practice, to simulate a real material the analyst first chooses an
analytical model suited to some general features (those he is interested in) of
the material and then tunes the parameters ofthe model in order to fit the
real material as accurately as possible.
We only considered purely mechanical aspects until now. It is most
probable that a separation between mechanical and thermal aspects is
unrealistic in the description of material behaviour in structural problems.
Other physical phenomena, such as electromagnetics, for instance, are
also correlated to mechanics, but are negligible in most cases. A better
structural model can be obtained by introducing thermal variables through
energy conservation, which involves thermomechanics.
Clearly, a constitutive relation should not depend on the actual structure
one studies: it is, by definition, common to all structures made out ofthe
same material. This suggests designing test structures for which it is easily
possible, by means of forces, torques, pressure, and so on, to produce some
desired stress patterns; adequate strain measures (by strain gauges) permit
then to find stress-strain relations that can be used to build up a constitutive
equation. For example, a common macroscopic experimental configuration used in static plasticity is the combined traction, torsion and
internal pressure tubular test specimen with which it is possible to realise a
454
large range of stress states, with the further advantage that the stress state is
about homogeneous in the specimen. 18
In transient dynamic behaviour large strain rates can be generated. If
many metallic materials exhibit only limited rate dependence when low
strain rates occur, this is not the case when strain rates go up to, say, 103/s.
This is particularly true for austenitic steels used in nuclear reactors.
Therefore, the constitutive relations obtained in static experiments have to
be adapted in order to simulate also the rate dependence. If the study'of
non-linear material behaviour in quasi-static situations is obviously a
complicated subject, what about dynamic situations?
The experimenter testing to determine the mechanical properties of'
materials under dynamic loading conditions encounters immediately at
least two difficulties : first, the obligation to measure in real time phenomena
of very short duration; secondly, the fact that disturbances due to wave
propagation effects will appear. The first will render the experiment rather
cumbersome to perform; the second will complicate the design of the
experiment.
Some practical difficulties are as follows: friction effects in the joints
present in the experiment ; unwanted reflection of waves on boundaries ; and
inertial effects due to non-continuous geometry (mainly concentrated
masses) which produce oscillations in the stress patterns.
It is therefore evident that the combined traction torsion pressure test
tube cannot be easily used in dynamics. Researchers have therefore tried to
design and study very simple configurations capable of producing onedimensional dynamic stress-strain curves (simple traction or compression).
Apart from the precursor work of Hopkinson (father 8 and son 9 ), extensive
work on dynamic properties of materials has only been done during the last
30 years. A survey of the work performed in Europe in the first part of this
period (until the 1960s) can be found in a paper by Cristescu; 1 ' it is also in
this period that Kolsky's famous book on stress waves in solids appeared. 1 0
Although it does not seem that much progress has been made in the
second period, 17 at least some more reliable reproducible results now exist.
Also, very helpful numerical experiments 23 have been performed with the
tools offered by the developing computing business. At present bidimensional tests are also designed, but their use is at an early stage. More
about experimental aspects can be found in the first part of the chapter.
7.2. Constitutive Relation for Static Elastic plastic Behaviour
7.2.1. Uniaxial Behaviour
Within a limited strain range (0-5 %) a simple traction or compression
455
test performed on a metallic material can be idealised as shown in Fig. 18:

from O to A the material has a linear elastic behaviour (A is the initial yield
point); from A to D the material hardens following a monotonically
increasing irreversible curve ABD (elastic-plastic behaviour); if unloading
occurs between A and D (. for instance), the behaviour is again linearly
elastic and is now the new yield point.
FIG. 18. - curve of a simple traction test.

A detailed analysis of the test results would show slightly more
complicated behaviour; in particular, the significance of point A, which is
defined here as the transition point between linear elastic and elastic-plastic
behaviours, could be discussed. 18
This simple test curve is of great importance, because, as we shall see
later, multiaxial states can be linked to it through the concept of
comparison value. Moreover, it already shows that there are two different
regimes of behaviour that we shall have to face.
Our constitutive relation, eqn. (16),
da.
Pijki de fc/
will thus have two distinct forms, depending on what regime it applies in:
(1)
(2)
Linear elastic form Pijkl = Dijkl, the classical elastic matrix

depending only on material constants: E the elastic modulus,
Poisson's ratio. This form applies to OA or BC, for instance;
An elastic-plastic form where F depends on material constants but
also on the current stress state. This type of behaviour is called
456
hypoelasticity : the stress increment is obtained by the knowledge of

current values only, but the overall response depends on the
deformation history.
7.2.2. Yielding Criterion
The determination of point A (Fig. 18) in a multiaxial case involves a
yield criterion explicitly defining the limit between elastic and elastic-plastic
behaviour. This criterion will be expressed in the stress space where it
represents a plasticity surface:
/'K) = 0
(17)
When < 0, the regime is elastic ; when = 0, the regime is elastic-plastic or

purely plastic.
As the material properties are isotropic, the yield criterion can be
expressed in terms of principal stresses or in terms of the invariants of the
stress tensor. Moreover, the yield function should be symmetric in the
principal stresses.
Experimentally, one notes that no plastic deformation occurs when the
hydrostatic stress in the material changes (by adding pressure, for instance).
This indicates that the yield function should not depend on the first
invariant of the stress tensor () or, to put it in other words, that the yield
criterion only depends on the deviatoric tensor of stress:
S
ij = ij -
immSij
(18)
One tries now to construct a yield function satisfying the properties

mentioned above and keeping as close as possible to the experimental
results.
The most widely used criterion is probably von Mises' criterion, ' 8 which
is expressed by
71^-^=0
(19)
where cy represents the monodimensional yield point (A or in Fig. 18),

and
= y/frjSij
(20)
is the comparison stress, which can also be expressed in principal stresses:

a = "TzVVi )2 + (Pi <F,)2 + (. ,) 2
(20a)
Compared with other criteria (for instance. Tresca 18 ), the von Mises'
457
criterion has the advantage of being a smooth function, which permits an

easy analytical treatment.
In order to remain consistent with the uniaxial stressstrain curve, it
should be possible to define also a comparison strain. Taking a form similar
to eqn. (20a) and imposing compatibility, one obtains
2
1
=
X/M 2) + (2 3 ) 2 + 3 J
(21)
>/2(1 + )
where is Poisson's ratio.
There are some difficulties in handling eqn. (21) in the elasticplastic
range; fortunately, the constitutive relation that we shall establish makes no
use of it. However, we shall have to introduce it for the strainrate
dependence.
7.2.3. Hardening Rule

A hardening rule defines the loading function describing the evolution of
the plasticity surfaces. We choose a hardening rule conserving isotropy: the
plasticity surface grows uniformly and keeps its original shape. So a certain
elasticplastic stress state always corresponds to the same plasticity surface
independently of the deformation history.
From an experimental point of view this is realistic as long as no reversed
loading occurs: in this case one has to take into account the Bauschinger
effect (apparent reduction ofthe yield point if loading is reversed after some
plastic deformation has occurred). This phenomenon destroys the initial
isotropy of the material.
Some theories try to take the Bauschinger effect into account. For
instance, Prager's kinematic hardening takes a kind of ideal Bauschinger
effect into account: the plasticity surface moves undeformed through the
stress space. It seems, however, that only poor results are obtained. It is
probably best to use the overlay model by which a Bauschinger effect can be
constructed by suitable superposition of isotropically hardened material
sublayers. For a discussion of current hardening models, see refs. 21, 22.
7.2.4. Description of Plastic Flow
The elasticplastic deformations are split into two parts (elastic and
plastic):
dr.;j = diTjj + di^
(22)
so that:
,} = D,]mn dr.l,n = Dijmn(di:mn
- dr*,,,)
(23)
458
C. ALBERTINI, J. P. HALLEUX AND M. MONTAGNAN1
Put in this form the problem consists in the determination of df.pm. This is
obtained through the law of normality:
=-^-
(24)
This relation, based on the postulate of Drucker and confirmed by the

experiments, is also compatible from a mathematical point of view if the
plasticity surface/ in eqn. ( 17) is a convex function (which is the case for von
Mises' criterion). Experiments also show that plastic deformations are
incompressible ( = 0-5).
7.2.5. Incremental Stress-strain Relationship
After some mathematical manipulation it is now possible to derive the
constitutive relation for elastic-plastic material behaviour (for a complete
demonstration, see ref. 15):
n,j = (Du
~Ks,jS
)d,
(25)
-v-
(26)
where
9G 2
K=
G = F/[2(1 4- v)] is the shear modulus.

Alternative interesting forms can be obtained, ' 4 but they seem to present
less generality. 19
7.3. Strain-rate-dependent Constitutive Relation
Monodimensional dynamic tests on metals (see Part 1) often show
the following features as compared with quasi-static tests: increase ofthe
yield point, the material seeming to be elastic over a larger range of
deformations; increase of the ultimate stress; diminished strain at rupture:
and dependence of hardening on strain rate. These effects normally increase
with the strain rate. As we are interested in a small strain region, only the
increase in the yield point and the dependence of hardening on strain rate
will be considered.
Roughly, it seems that strain rate plays a role only when plastic
deformations occur.
Owing to the lack of experimental information on two- and threedimensional results, one is obliged to extrapolate monodimensional results
in order to estimate material behaviour under multiaxial loading: clearly.
459
this is unsatisfactory and one can only hope that the extrapolation has a
sense in real life.
Two-dimensional experiments have not delivered sufficient information
yet to be used in deriving a constitutive relation, but can be used in the
framework of numerical experiments performed with extrapolated
monodimensional constitutive relations so as to confirm the extrapolation :
this is current research.
Thus, in our view, at this stage it is best to try to keep out as much as
possible of the characteristics of static elastic-plasticity, trying to adapt
only some quantitative parameters: a strain-rate-dependent elastic-plastic
material appears, then, as belonging to the subclass of elastic-plastic
materials. From a computational point of view it would certainly be nice to
use relation (25) where only K depends on the strain rate.
The test features mentioned above suggest the consideration of the
strain-rate effect as a simple additional hardening; let us write a formal
constitutive relation, capable of describing the monodimensional tests
during plastic loading:
da = + '
(27)
where da is the stress increment; d is the strain increment; dt; is the strain
rate increment; and Ep and E' are, respectively, the hardening at constant
strain rate and the hardening at constant strain, both depending on the
actual state of the material.
In order to simulate an increment of stress by using an adapted 'static'
relation, one needs the form
da = Epdt:
(28)
where E'p becomes an equivalent plastic modulus. Now,

d
E'p = Ep + E'
dt:
(29)
is the expression of a corrected hardening explicitly depending on strain

rate.
If we transpose this to a general threedimensional case, keeping all the
elements ofthe static elasticplastic relation, it results in a simple change of
the coefficient in eqn. (26). However, a problem appears: how to
characterise strainrate dependence in the threedimensional case? E' is a
scalar whose calculation depends on the strainrate tensor and the strain
tensor. If it is common for stress or strain to extrapolate from one
dimensional to threedimensional by simply using the comparison values,
460
there is no evidence that strain rate should follow the same rule. A more
attractive alternative missing, we arbitrarily choose to represent strain rate
as we represent strain (eqn. 21). Owing to the different Poisson's ratios in
elastic (0-3 for steel) and plastic (0-5) deformations, it is more convenient to
write:
3 = a v /(d8 1 - 2) + (2 - 3)2 + (3 - derf
(30)
During elastic-plastic deformations the elastic part (eqn. 1 22) is often

very small compared with plastic deformation (plastic modulus <sc elastic
modulus), and one can use:
a =
^2(1 + v)
in the elastic range
_/2
= -^-
, .
in the elastic-plastic range
is then obtained through integration of eqn. (30). When the total plastic
deformation becomes very large compared with the total elastic
deformations, it is possible to use eqn. (21):
3-
V(E> - )2 + (2 - 3 ) 2 + (3 - .) 2
(31)
Strain-rate comparison values are obtained by simply replacing e by ' in

eqns. (30) and (31).
Of course, more extended expressions for dynamic constitutive relations
where the strain rate dependence appears explicitly in all components can
be derived; in particular, this is easily done for Prager's kinematic
hardening model by simply adding strain-rate dependence to Prager's
hardening rule. The fundamental difficulty encountered above remains: it is
not known how strain-rate effects in various directions combine, and
therefore the specific coefficients of the strain rate components cannot be
correctly derived.
8.
C O M P U T A T I O N A L ASPECTS
It is not intended to enter into the structure of the algorithms used to

describe elastic-plastic rate-dependent material in a transient structural
dynamic program. Nevertheless, it is of interest here to show a typical flow
chart that can be encountered in practice.
For the sake of clarity we choose a completely explicit procedure in which
461
no iteration is needed; this can be used, for instance, in the framework of a

corotational (rotated coordinates) lumped (mass) explicit (time integration)
formulation 1 6 1 9 , 2 4 , where the natural time step limitation ensures small
strain increments. Note, however, that in frequent situations (for instance,
when sharp stress shapes due to fast impact loading are present) this might
be insufficient.
One has
current yield point of the previous step

stress state at time r_ {
rf;1
". deformation states at times r_j, t
fc
lJ 'J-u current strain rate of the previous step
old
, ', the material parameters
>
Calculate the comparison stress a"
Estimate new strain rate , and new comparison strain

Calculate yield point increase due to strain rate: '(, ) (enlarged
elastic regime)
-YESNO
YES
NO
d(T
ij=Dijkl^kl
Calculate equivalent
plastic modulus E'p
o,j = P,jk, dekl
_n
_n 1
+ da,
462
C. ALBERTINI. J. P. HALLEUX AND M. MONTAGNANI
FIG. 19.
Piecewise linearised stress-strain curves.
In order to carry out the calculations of a'y and E'p, it is convenient to w o r k

on piecewise linearised uniaxial stress-strain curves. Practically o n e takes
experimental data for various strain rates and fits them as shown in Fig. 19.
a'y and E'p(Ep and E') are then obtained by linear interpolations between
the plotted curves which can be easily memorised in a c o m p u t e r .
REFERENCES
1. Davies. R. M. (1948). 'A critical study ofthe Hopkinson pressure bar". Phil.
Trans. Roy. Soc, London. Ser. A. 240, 375.
2. Albertini. C. and Montagnani, M. (1976). 'Wave propagation effects in
dynamic loading'. Nue!. Eng. Des.. 37, 115-24.
3. Lindholm. Ulric S. (1971). "High strain rates tests', in Techniques of Metals
Research. Vol. 5. Part 1. Wiley. New York.
4. Albertini. C. and Montagnani, M. 'Testing techniques based on the split
Hopkinson bar". Inst, of Phys.. Conf. Ser. No. 21. London, 1974; Oxford. 2-4
April. 1974.
5. Kolsky. H. (1953). Stress Waves in Solids. Clarendon Press. Oxford.
6. Campbell. J. D. (1973). 'Dynamic plasticity: macroscopic and microscopic
aspects'. Mat. Sei. Eng.. 12, 3-21.
7. Lindholm. U. S. and Yeakley. L. M. (1968). 'High strain rate testing: tension
and compression', J. Exp. Mech.. January.
8. Hopkinson. J. (1872). Collected Science Papers. Vol. 2. p. 316.
9. Hopkinson, B. (1905). 'The effects of momentary stress in metals'. Proc. Roy.
Soc. ., 74, 488-506.
10. Kolsky. H. (1953). Stress Waves in Solids, Clarendon Press, Oxford.
MATERIAL BEHAVIOUR AND MOD ELLING
463
1 1. Cristescu, N. (1960). 'European contributions to dynamic loading and plastic

waves', in Plasticity Proceedings oj the 2nd Symposium on Naval Structural
Mechanics (Ed. E. H. Lee and P. S. Symonds), Pergamon. Oxford, pp. 385 442.
12. Truesdell, S. C. and Toupin. R. A. (1960). 'The classical field theories', in
Encyclopedia of Physics, Vol. Ill (Ed. S. Flgge), Pergamon, Oxford, p. I.
13. Cemal Eringen. A. (1967). Mechanics of Continua. Wiley, New York.
14. Hartzman. M. and Hutchinson, J. R. (1972). 'Nonlinear dynamics of solids by
the finite element methods'. Comp. Sinici.. 2, 47-77.
15. Frey. F. ( 1973).'Le Calcul Elasto-Plastique des Structures par la Mthode des
Elments Finis et son Application l'Etat Plan de Contrainte'.rapport no. 33,
Laboratoire de Mcanique des Matriaux et de Statique des Constructions,
Universit de Lige.
16. Belytschko. T. and Hsieh. B. J. (1973). 'Nonlinear transient finite element
analysis with convecled coordinates". I nt. J. Num. Meth. Eng., 7, 255-7!.
17. Campbell. J. D . (1973). Review paper: 'D ynamic plasticity: macroscopic and
microscopic aspects*. Mai. Sei. Eng., 12, 3-21.
18. Massonnet, Ch. ( 1974). Rsistance des Matriaux. Vol. 2, 2nd edn. Sciences et
Lettres. Lige.
19. Halleux. J. P. (1975). 'Analyse de structures parois minces en rgime
dynamique non linaire, l'aide d'un lment fini de plaque'. Mmoire
d'Ingnieur Physicien, University of Lige.
20. Saczalski. K. J. and Park, K. C. 'Comments on inelastic constitutive relations
for prediction of structural crash-impact response'. Workshop on Inelastic
Constitutive Equations for Metals. Rensselaer Polytechnic Institute (April
1975).
21. Armen. H. 'Plasticity in general purpose software', Workshop on Inelastic
Constitutive Equations for Metals. Rensselaer Polytechnic Institute (April
1975).
22. Stricklin. J. ., Hunsaker, . and Vaughan. D . K. 'A comparison of current
work hardening models used in the analysis of plastic deformations". Workshop
on Inelastic Constitutive Equations for Metals. Rensselaer Polytechnic
Institute (April 1975).
23. Bertholf, L. D.and Karnes. C. H. (1975).'Two-dimensional analysis of the split
Hopkinson pressure bar system', J. Mech. Phys. Solids. 23, 1-19.
24. D ona. J., Giuliani, S. and Halleux. J. P. (1976). 'Theoretical aspects ofthe
EURDYN computer programs for non-linear transient dynamic analysis of
structural components', EUR report 5473e.
25. Belytschko.T. (1976).'A survey of numerical methods and computer programs
for dynamic structural analysis', Nucl. Eng. Des., 37, 23-4.
26. Key, S. (1978). 'Concepts underlying FEM for structural analysis', Nucl. Eng.
Des., 48, 259-68.
27. Frey. F. (1978). 'L'Analyse Statique Non-Linaire des Structures par la
Mthode des Elments Finis et son Application la Construction Mtallique',
Thse de D octorat, Universit de Lige.
28. Rice.J.R., McMecking. R. M., Parks, D. M. andSorensen, E. P.,'Recent finite
element studies in plasticity and fracture mechanics', FENOMECH ISD .
Stuttgart (September 1978).
Index
Acceleration response, 66
Acceleration-time history, 65
Accelerations, 125
Acoustical analysis, 319-22
Acoustical source characteristics, 320
Aircraft impact problems, 337-86
corrections and smoothing
procedures, 377
damping parameters, 374
deformable wall calculations, 357
displacements of deformable
concrete structure, 344
dynamic response of typical
structures, 356-65
dynamic structural investigations
for characteristic buildings,
349-65
influence of model and parameter
variations on validity of
results, 368-77
load case characteristics, 338
load functions, 338-49
analysis, 347-9
with assumption of rigid wall,
340
with elasto-plastic deformation
behaviour, 341
model representation, 368
response spectra, 365
selection of governing points of
impact, 354
soil parameters, 374
time histories of accelerations, 357
Aircraft impact problemscontd.

undeformable wall calculations, 357
vibration propagation from one
building to another, 377
Almansi strain tensor, 140
Amplification factor, 209
Antihour-glass viscosity, 277
Arbitrary Lagrangian Eulerian (ALE)
formulation, 257-68
ARES, 241
Artificial viscosity, 277
ASTARTE, 240
Bar impact problem, 92-4
Bauschinger effect, 150
Bending
equation, 74, 80
moments, 37, 224
Blowdown suppression system
fluid-structural dynamics in,
291-314
structural dynamics, 294
Boundary integral equation (BIE)
method, xvi
Brunsbiittel reactor, 291, 308
Buildings, dynamic structural
investigations, 349-65
Canonical principle, 13
Cantilever beam, 46
Cauchy stress, 99, 140
465
466
INDEX
Central difference time integration,

103-10
Characteristics method, 216
Compatibility equations, 29
Complementary energy principle, 16,
19
Compressibility, 207-16
Compressor blade eigenfrequencies, 39
Computer methods, xi
Conditionally stable methods, 125
Consistency, 209
Consistent mass matrix, 24
Constant-pressure elements, 268, 274
Constitutive relations
static elastic-plastic behaviour,
454-8
strain-rate-dependent, 458-60
elastic-plastic behaviour
simulation models, 452-60
Continuum formulation, 112
Convergence, 209
properties of equilibrium models,
21
Corotational formulations, 111
Courant-Friedrichs- Lewy condition
(CFL), 106-8, 214
Courant's minimax principle, 19
COVA. 245
Crack
extension. 187
initiation, 1 56
propagation, 153, 156
energy balance criterion for.
165-9
in pressurised pipelines. 169-73,
188
models, 179
stress criterion of, 165, 171
velocity, 187
Crank-Nicolson rule, 147
Cylindrical shell
analysis, 36-40
immersed in fluid. 1 14
d'Alembert
force, 119
principle, 100
Damping
characteristics, 87
coefficients. 331
factor, 90
matrices, 162
parameter, 82
Deflection, large, 161
Deformation, large, problems of, 140-2
Degrees of freedom, 43-51
Differential equations, 97
Digital computers, 123
Digital programmers, xi
Direct integration methods, 124, 125
Discontinuities, 219
Discrete elastic system, 7-9
Discrete equations, 78
Discretisation
of shell equations, 227-31
technique, 255
Disjoint partition, 116, 121
Displacement, 4, 125
discretisation, 24
in elastodynamics, 10-18
increments, 148
variational principle, 11, 15
vector, 258
Divergence operator, 212
Double-cantilever beam (DCB). 135,
156. 157, 164-9
Dynamic crack propagation. 153, 169
Dynamic energy release rate, 155
Dynamic equilibrium equations, 125
Dynamic fracture mechanics, 154-6
Dynamic fracture problems, 156
Dynamic load factor, 349
Dynamic stress determination. 313
Dynamic stress intensity factor, 155
Dynamic transient analysis of solids.
123-52
Dynamic transient finite element
analyses, 153
Dynamic transient situations, 123
Earthquake
excitation, 59-65
loads, 381
resistance. 349
INDEX
Eigenvalues
derivation of, 51-3
recursive characterisation, 10
Eigenvectors, derivation of, 51-3
Elastic matrix, 162
Elastic-plastic matrix, 162
Elastodynamics
displacements in, 10-18
finite element implementation of
variational principles of, 18-31
Elastoplastic incremental stress-strain
relations, 144
Elastoplasticity, 142-4
Elastoviscoplastic response, 177
Elasto-viscoplasticity, 175
Energy
balance criterion for crack
propagation, 165-9
equations, 267, 274
usage factor, 414
Entropy condition, 204
Equations of motion, fluid-structural
dynamics, 304-6
Equilibrium
approach, 27, 33
models, 21, 34
Euler-Bernoulli beam element, 108
Euler-Bernoulli dynamics, 109
Euler equations, 4, 14, 17, 23, 25,
202, 204, 206
Euler process, 147
Eulerian codes, 248
Eulerian description, 257, 258
Eulerian fluid cells, 235
Eulerian formulations, xvi, 205-7
Eulerian mesh, 248, 249
Eulerian methods, 212
EURDYN-1M, 246, 249, 282
Explicit-explicit partitions, 120-2
Explicit-implicit operator, 89-93
Explicit-implicit partitions, 120-2
Explicit Lagrangian calculation, 272
Explicit linear multistep formula, 117
Explicit methods, 124, 158
Explicit procedures, 78, 79
Explicit time integration, 97-122,
158-64
basic equations, 98-103
467
Explicit time integrationconld.

flow chart, 112
implementation, 110-14
Explosion shock wave, 381
External forces, 160
Finite difference method, xv, 124,
208, 342
Finite element
implementation of variational
principles of elastodynamics,
18-31
method, xi, 1,43, 124, 126, 217,
255-90, 394-410
nodal point displacements, 135
Flat shell elements, 37
Floor response spectra, 67
Flow-induced vibrations in piping
systems, 315-36
Fluid analysis algorithm, 256-78
Fluid dynamics, 202-21
Fluid flow characteristics, 196
Fluid-in-cell (FLIC) method, 213
Fluid-structural dynamics
coupling between, 302-4
equations of motion, 304-6
in blowdown suppression systems,
291-314
Fluid-structure
coupling, 231-6, 279-82
interaction problem, 89
Force usage factor, 412-14
Fourier methods, 104
Fracture mechanics problems, 153-90
Free vibration analysis, 7-9, 36-40
Frequency-by-frequency comparison,
78
Frequency distortion, 87
Friedrichs transformations, 13
FRUSTA
code, 411, 415-23
model. 394
Galerkin method, 100
General eigenvalue problem, 134
GETTO 2 code, 410
468
INDEX
Green-Gauss theorem, 277

Green-Lagrange strains, 141
Green strain tensor, 12
Ground response spectra, 67
Hamilton's principle, 2-5, 11, 19
operating on displacements and
velocities, 4
variational derivatives of, 13
Hamilton's rule, 23, 25
Hardening, 150
rule, 457
Hexagonal tube under internal
pressure pulse, 282
Hopkinson-Davies bar, 427, 431-41
Houbolt method, 128-9, 137-8,
149-50
Hydrodynamic equations, 201, 208
Hydropneumatic machine, 428
Hypothetical core disruptive accident
(HCDA), 191
containment of, 193
sequence of events, 195
ICE (Implicit Continuous-fluid
Eulerian) method, 215
Impact analyses, 88
Implicit Lagrangian calculation, 272
Implicit methods, 78, 81-87, 124, 136
Implicit time integration methods,
124, 128-33
Implicit trapezoidal scheme, 147
Inertial problems, 126
Initial value problem, 205
Integration methods, 216-19
Internal energy equation, 268
Internal force vector, 104
Internal forces, 160
IRAN reactor building, 381
Isoparametric element configurations.
127
Isotropic hardening, 150
Jacobian determinant, 259
Jointed partition, 115
Kinematic approach, 22
Kinematically consistent
displacements, 100
Kinetic energy, 45, 86
Kirchhoff hypothesis, 99
Kirchhoff-Piola stress, 113
Kirchhoff-Trefftz tensor, 12, 14
Lagrange multipliers, 4
Lagrange's principle, 116
Lagrangian boundary, 235-6
Lagrangian codes, 239, 240
Lagrangian coordinate system, 141
Lagrangian derivative, 205
Lagrangian description, 257, 259
Lagrangian difference mesh, 211
Lagrangian finite difference program.
117
Lagrangian formulations, xvi, 114,
174, 178, 205-7
Lagrangian mesh, 240, 248
Lagrangian methods, 210, 218. 231
Lagrangian shell nodes, 235
Lagrangian surface tractions, 14, 17
Laplace's equation, 74
Lax scheme, 213
Legendre transformation, 5, 6
Linear displacement field, 106
Linear elastic dynamic transient
problems, 124
Linear elastic fracture mechanics, 154
Linear wave-propagation-type
problems, xv
LMFBR, 193
containment loading, 195
Load distribution in threedimensional models, 356
Lumped mass, 162
matrix, 24, 106
MAC method. 215
Mass equations, 274
Material
behaviour at high strain rates,
427^*6
constitutive law, 441-4
coordinates, 258
INDEX
Mean flow parameter analysis, 317

Membrane elements, 37, 39
Mesh
formulations, xvi
partitions, 114-22
Minimax principle, 9, 10
Modal analysis of piping systems.
327-9
Modal characteristics calculation of
piping systems, 322
Modal damping
of coupled oscillating structures,
353
parameter, 134
Modal method
for transient response, 43-69
use of term, 53
Modal superposition techniques. 124.
133-5
Momentum
conservation, 266
equation, 266, 276
Moody model, 387
Multi-degree-of-freedom system,
43-51
Natural frequencies, 68
Navier-Stokes equations, 201, 202
Newmark method. 83. 84. 132-3,
139-40, 149
Newmark method, 82, 86-7, 90
Nodal displacement components, 124
Nodal forces, 102
Non-linear solution algorithms.
135-40
Non-linear solution procedures, 125
Non-linear transient dynamic
analysis, 173-9
Nuclear power plants, 349
Nuclear reactor buildings as
axisymmetric structures, 350
Nuclear reactors, xii
postulated blowdown, 291
safety aspects, 191-253
Pamel special elbow element, 409
469
PAPS model, 404

PAPS T2G analysis, 423
Partial differential equations, 97, 100
Participation factor, 64
Particle-in-cell (PIC) method, 213
Piola-Kirchhoff stresses, 141
Pipe whip analysis, 387-426
applications, 410-23
energetic preliminary solution, 393
finite element analysis, 394-410
forces computation, 387-8
solution analysis, 389-93
three-dimensional simulation, 403
Pipeline fracture problems, semiloof
element for, 179-87
Pipework
automatically selected degrees of
freedom, 48
fundamental mode of, 49
Piping systems
flow-induced vibrations in, 315-36
model analysis of, 327-9
modal characteristics calculation of,
322
PISCES-II DELK program system,
342
Plastic
deformation, 142, 177
flow, 144, 153, 187
strain, 143
Plasticity effects, 161
Plate bending
confirming model, 31
equilibrium triangle, 32
mixed model, 31
Plate equilibrium models, 3640
Poisson equation, 216
Potential energy, 12
Power spectral densities, 316-17
Pressure
fluctuations induced by flow, 315
suppression systems, 291, 306-16
Pressurised pipelines, crack
propagation in, 169-73, 188
Primary containment model response,
283-8
Principle of virtual power, 266
Pseudo-force formulation, 136
470
INDEX
Rankine-Hugoniot conditions, 219

Rayleigh damping, 59
Rayleigh quotient, 9-10, 26, 105
Rayleigh-Ritz methods, xv
Reaction force vector, 60
Recursive characterisation of
eigenvalues, 10
Reduced stiffness matrix, 45
Response spectrum method of seismic
analysis, 65-8
Rezoning techniques, 238, 273
Seismic analyses, 88
response spectrum method of, 65-8
Seismic loading situations, 135
Semi-analytical techniques, 217
Semidiscretisation, 97, 102
Semiloof element for pipeline fracture
problems, 179-87
Semiloof shell element, xvii, 173-9
SEURBNUK-2, 240-5
Shell equations, discretisation of,
227-31
Shell material, stress-strain
relationship for. 226
Single-degree-of-freedom system. 66.67
SING-S computer program, 306-16
Skew cantilever plate analysis, 31-6
Slave displacements, 50
Slide lines, 239, 240
Software market, xii
Spatial coordinates, 258
Spatial discretisation, xv
Spatially discrete models, 72-8
Stability, 80, 209, 210
Stillness matrix, 137
Strain energy density, 12
Strain energy release rate. 155
Strain-hardening parameter. 161
Strain rate, 145, 147
Strain-rate-dependent elastic-plastic
behaviour simulation models.
446-62
computational aspects, 460-2
constitutive relations, 452-60
field equations, 449
problem definition, 450
Strength of materials, xiii

Stress criterion of crack propagation.
165. 171
Stress increments, 148
Stress intensity factor. 1 54
Stress-strain matrix, 100
Stress-strain relationship
for shell material, 226
incremental. 458
Stresses, variational principle
operating on. 15. 16
Structural dynamics problems, xi-xviii
Structural stiffness matrix, 124
Tangent stiffness method, 137
Tangential stiffness matrix, 148
Taylor series expansion, 147
TEDEL computing code. 315, 324,
326-9. 331
Tensile tests
dynamic biaxial, 444-6
dynamic uniaxial, 428
Theory of elasticity, xiii
Thin shell theory, 222-6
Time integration
operators, 78-87
schemes, 87
transient response by, 71-95
Time integrators. 87-9
Time step stability limits. 177
Time-stepping scheme. 147. 159
Time steps required for stability, 163
Timoshenko beams. 46
Toupin's functional, discretisation. 28
Toupin's principle. 5-7
Transfer functions, calculation of, 325
Transfer matrix method, xv
Transient dynamic analyses, 88
Transient dynamic finite element
analysis, 158-64
Transient dynamic fluid-structure
interaction, 255-90
fluid-structure coupling, 279-82
structural elements, 278-9
Transient dynamic numerical
methods, 153
Transient dynamic response, 72
INDEX
Transient dynamic situations, 427-63

Transient excitation. 66
Transient fluid-structure problems,
191-253
characteristics of, 196-8
containment codes, 236-46
current status and future trends.
236-50
fluid dynamics, 202-21
fluid flow characteristics, 196
fluid-structure interaction, 198
popular simplifications. 198-200
structural elements, 197
structures, 221-2
Transient response
analysis, 123
by time integration, 71-95
modal method for. 43-69
Translation modes of rigid bent tube
with open extremities and
containing quasiincompressible fluid, 332-5
TRICO computer code, 328
Truncation error, 209
Unconditionally stable methods, 125
US NRC spectra, 67, 68
Variational equation, 25
Variational form of continuity, 237
Variational methods, 1-41
numerical applications. 31-40
Variational principle, 1
based on complementary energy, 19
for small and large displacements in
elastodynamics. 10-18
of elastodynamics, finite element
implementation of, 18-31
471
Variational principleconici.
special-purpose two-field. 17
two-field. 15
Variational theorems for discrete
elastic systems, 2-10
Vector
function. 204
iteration scheme, 27
Velocities, 4. 125
discretisation, 24
variational principle operating on,
15. 16
Velocity strain-Cauchy stress
formulations, 113
Velocity-strain continuum description,
99
VIBRAPHONE computer code, 321
Vibrations, xiii
in piping systems, flow-induced,
315-36
Virtual displacement system, 128
Viscoelastic-plastic material response,
146-9
Viscoplastic flow rule. 146
Viscoplastic strain rate, 175
Viscoplastic theory, 144
Viscous forces, 104
von Mises criterion, 226
von Mises yield surface, 176
Wall pressure fluctuations, 316-17
Wave
equations, 72, 80
propagation problems, 126
Wilson 0 method, 129-32, 139-40,
149, 150
Yield criterion, 175
CONTENTS
1. Variational methods of structural
dynamics and their finite element
implementation.
2. The modal method for transient
response and its application to
seismic analysis.
3. Transient response by time integration: review of implicit and explicit
operators.
4. Explicit time integration of structuremechanical systems.
5. Implicit finite element methods for
the dynamic transient analysis of
solids with particular reference to
non-linear situations.
6. Application of transient dynamic
numerical methods to problems in
fracture mechanics.
7. Survey of computational methods
for transient fluid-structure problems in reactor safety.
8. Finite element analysis of transient
dynamic fluid-structure interaction.
9. Coupled fluid-structural dynamics
in blowdown suppression systems:
numerical schemes and applications.
10. Computation of flow-induced vibrations in piping systems.
11. Analysis of aircraft impact problems.
12. Pipe whip analysis.
13. Material behaviour and modelling
in transient dynamic situations.
Index.
TITLES OF INTEREST
DEVELOPMENTS IN FRACTURE MECHANICS1
Edited by G. G. Cheli
6 9", + 310 pp., 105 illus.
STABILITY PROBLEMS IN ENGINEERING

STRUCTURES A N D COMPONENTS
Edited by T. H. Richards and P. Stanley
6 9", xiii + 428 pp., 219 illus.
DEVELOPMENTS IN PRESSURE VESSEL

TECHNOLOGY1
Edited by R. W. Nichols
6 x 9", xii + 230 pp., 81 illus.

DEVELOPMENTS IN PRESSURE VESSEL
TECHNOLOGY2
6 9", xii + 242 pp., 84 illus.
DEVELOPMENTS IN STRESS ANALYSIS1

Edited by G. S. Holister
6 x 9", ix 4- 197 pp., 80 illus.
POST-YIELD FRACTURE MECHANICS

Edited by D. G. H. Latzko
6 9", xv + 349 pp., 165 illus.
NON-LINEAR PROBLEMS IN STRESS ANALYSIS

Edited by P. Stanley
6 x 9', xiii + 472 pp., 197 illus.

MATERIALS A N D COATINGS TO RESIST HIGH
TEMPERATURE CORROSION
Edited by D. R. Holmes and A. Rahmel
6 x 9", xvi + 406 pp., 265 illus.
DEVELOPMENTS IN STRESS ANALYSIS FOR

PRESSURISED COMPONENTS
6 9", + 210 pp., 79 illus.
PHYSICAL METALLURGY A N D THE DESIGN OF

STEELS
F. B. Pickering
6 x 9', xil 4- 275 pp., 168 illus.

Cdna06693enc 001

Загружено:

Сведения о документе

Оригинальное название

Авторское право

Доступные форматы

Поделиться этим документом

Поделиться или встроить документ

Параметры публикации

Этот документ был вам полезен?

Это неприемлемый материал?

Авторское право:

Доступные форматы

Cdna06693enc 001

Загружено:

Авторское право:

Доступные форматы

Advanced

APPLIED SCIENCE PUBLISHERS

This book presents a review of

ADVANCED STRUCTURAL DYNAMICS

APPLIED SCIENCE PUBLISHERS LTD

British Library Cataloguing in Publication Data

Introduction und General Overview oj Structural Dynamics

1 Variational Methods of Structural Dynamics and their Finite

2 The Modal Method for Transient Response and its A pplication

3 Transient Response by Time Integration: Review of Implicit

4 Explicit Time Integration of StructureMechanical Systems

5 Implicit Finite Element Methods for the Dynamic Transient

6 Application of Transient D y n a m i c Numerical

7 Survey of C o m p u t a t i o n a l M e t h o d s for Transient Fluid-Structure

8 Finite Element Analysis of Transient D y n a m i c F l u i d - S t r u c t u r e

9 Coupled F l u i d - S t r u c t u r a l D y n a m i c s in Blowdown Suppression

10 C o m p u t a t i o n of Flow-induced Vibrations in Piping Systems

11 Analysis of Aircraft Impact Problems

13 Material Behaviour and

Commission of the European Communities. Joint Research

Department of Civil and Nuclear Engineering, The Technological

Commission of the European Communities. Joint Research

Commission oj the European Communities. Joint Research

Central Electricity Generating Board, Research Division,

Laboratoire de Techniques Aronautique s et Spatiales, Universit de

Commissariat l'Energie Atomique, Division des Etudes Mcaniques

Commission of the European Communities, Joint Research

Commission of the European Communities, Joint Research

Kernforschungszentrum Karlsruhe GmbH, Institut fr Reaktorent

Kraftwerk Union A .G.. Berliner Str. 295299, D6050 OJjenbach

Via P. Pesce, 1 16151 Genova Sampierdarena,

Commission of the European Communities. Joint Research

Department oj Civil Engineering, University of Wales, Singleton Park,

Commission of the European Communities. Joint Research

Introduction and General Overview of Structural

Commission of the European Communities, Joint Research Centre, lspra,

relations) with occasional small and clean experiments on semi-finished

INTRODUCTION AND GENERAL OVERVIEW

their computer implementations, but also include many examples of case

CLASSIFICATION OF SOLUTIONS FOR DYNAMIC

Until fairly recently, standard university courses in structural mechanics

At present we are in a completely different situation. For most of the

INTRODUCTION AND GENERAL OVERVIEW

conservation of mass, momentum and energy, interrelated by constitutive

characteristics in vogue up to some years agoin particular for fluid

The choice is mostly dictated by the physics of the problem. Even if

INTRODUCTION AND GENERAL OVERVIEW

interfaces in a Eulerian meshfor problems in fluid mechanics with large

An attempt has been made to highlight the role of modern computerised

Various types of approach to dynamic problems as they will be dealt with

Universit de Lige, Belgium

Certain numerical methods, such as finite differences, apply most naturally

elastodynamics, Toupin's dual formulation for describing the dynamics of

VARIATIONAL THEOREMS FOR DISCRETE ELASTIC

2.1. Hamilton's Principle

VARIATIONAL METHODS OF STRUCTURA L DYNA MICS

For conciseness, let us adopt a matrix formalism. Equation (1) can be

where L isa Boolean matrix of dimension 3N( l)/2 3V. If theelastic

with the alternative definitions of the stillness matrix

The kinetic energy of the system is that of the individual particles

with the mass matrix of the system M = diagfm!,).

over a time interval subject to the boundary conditions

Performing the variation of eqn. (6) and integrating by parts in time