Вы находитесь на странице: 1из 12

Nuclear Engineering and Deign 52 (1979) 99-110

g) North-Holland PublishingCompany

OUTLINE OF DYNAMIC ANALYSIS FOR PIPING SYSTEMS

Kent GORDIS
DST ComputerServices, SA, CH.1216 Geneva, Switzerland

Received 11 August 1978

The paper deals, in a coherent mathematical manner, with a number of theoretical and practical problems involvedin
the solution of dynamic structural problems by the modal superposition method. In particular, problems related to piping
system structures subject to s/ngle or multi-levelsupport movement loading are treated.

0. Introduction and summary of main results thematical description of the widely used response
analysis method which is referred to in this paper as
There are a number of theoretical and practical "Classical Response Analysis", and which I show is
problems involved in the solution of dynamic problems valid only for single level problems. In section 16, I
by the modal superposition method. These are: present a precise mathematical description of several
1. How many mass points? methods for solving multi-level problems. Section 17
2. How to concentrate mass? compares the various methods described in this paper.
3. How many modes? The Classical Response Analysis method involves
4. How to account for the effect (which cannot be finding maximum modal responses "relative to the
neglected) of the higher modes? ground motion". In mathematical terms, this means
5. How to account for resonances in the neighbor- that the displacements are represented in a "shifted"
hood of calculated natural frequencies? coordinate system which is moving rigidly with respect
6. How to account for damping? to the support movements. Such a shift is well defined
7. How to construct bound solutions for any kind only if all supports are moving rigidly with respect
of dynamic loads, including multilevel support move- to each other.
ments and applied forces? In this paper I extend the notion of a "shift" from
8. Finally, how to construct modal bound and/or such "ground motion" shifts t o encompass any shift
time history solutions at a reasonable cost? which starts from rest. I define the class of "rigid ~hifts"
I am preparing a paper [4] in which I deal with as shifts for which the coordinates of the points of the
these problems in a coherent and general manner. To structure are moving rigidly with respect to each other.
achieve this objective, I have used a purely mathema. I show that only rigid shifts have the property that all
tical (as opposed to intuitive, physical or experimental) internal member forces due to the shift are zero.
methodology. To illustrate the methodology, I present I examine methods for extending Classical Response
in this paper my results relating to structures subject Analysis to multi-level problems us~g rigid shifts. I
to single or multi-level support movement loading. show that, in general, rigid shift methods result in two
In the first 14 sections, I establish the mathematical types of terms. First, there are terms, similar to terms
notation and, at the same time, I outline results which used in Classical Response Analysis, involving mass
effect practical calculations (such as "rigid mode for- weighted participation factors which multiply accelera-
ces") but which are not specifically related to the ffOn responses. In addition, there are terms involving
mathematical solution of structures subject to support stiffness weighted participation factors which multiply
movement loading. In section 15, I give a precise ma- displacement responses.

99
100 K. Gordis / Dynamic analysis Jor piping s),stems

I show that the terms involving stiffness weighted Non bold faced upper case letters :ire t~sed Io der?otc
participation factors and displacement responses are both 6 X 6 matrices and 6-vectors. Examples are K,.~.
zero for a rigid shift if and only if the problem is a the 6 × 6 stiffness matrix of the member (q, r) at poin~
single level problem with "ground motion" shift. This r, and Xr(t) the 6-vector of displacements corresponding
explains why these terms do not enter Classical Res- to the six degrees-of-freedom at poini r
ponse Analysis calculations, and shows that such terms Non bold faced upper case letters are also used to
must be used to solve multi-level problems by rigid denote the total number of various quantities. For
shift methods. example, J denotes the total number of modes.
The expression (16.27) demonstrates that mass and Non bold faced lower case letters are used to denote
stiffness weighted participation factors are related. scalars. For example, yj(t) denotes the scalar modal
Finally, I examine another method for solving anaplitude for mode ].
multilevel problems, which is based on the "pseudo
static" shift described by Clough and Penzien in [1].
This method, which I call the "Clough and Penzien" 2. Equations of motion
method, is a generalization of Classical Response
Analysis which involves only acceleration responses. The undamped equations of motion of the piping
The Clough and Penzien method has been used by system may be represented in the form
Schmid and Ludwig [2] and Leimbach [3] and by Wu MYc + Kx = f (2.0)
et al. [5] as the basis for a computer solution for multi-
level problems, I show that the Clough and Penzien shift x(0) : o
is not a rigid shift, and hence, that it is necessary to where K is the N X N system stiffness matrix; M is the
include bounds for the internal member forces due to N X N mass matrix; x -= x(t) is the N-vector of unknown
the shift. This requirement is recognized by Clough and displacements; and f - o f ( t ) is the N-vector of applied
Penzien, who refer to this effect as "pseudo static forces. We assume that of(0) = 3~(0) = 0.
stresses" but is not explicitly mentioned by Schmid and
Ludwig.
In (16.23) I have derived an expression for the 3. Natural frequencies and mode shapes
modal response for the Clough and Penzien method.
This expression involves only the 6 X 6 stiffness ma- The natural frequencies co] are scatars, and the mode
trices of support members, rather than the "influence shapes ej are N-vectors f o r j = 1,2 ..... J which satisfy
function matrix" used by both Schmid and Ludwig and Kej = coTMej , (3.0)
by Wu et al. and, therefore, represents an important
eTMe] = t , (3.1)
computational simplification.
where the superscript "T" denotes "transpose". We
1. Notation note that the "dot" product eTMei is a scalar.

The following conventions are usually observed in 4. Modal amplitudes


this paper. Exceptions are obvious from the context.
Bold face upper case letters are used to denote ma- The scalar modal amplitudes Yi(t) are defined by
trices which describe characteristics of the entire struc- the expressions
ture. For example, K denotes the stiffness matrix of t
the structure. sin coj(t - T) eTf(r) d-r. (4.0)
Bold faced lower case letters are used to denote
yj(t) = f
7-=0
vectors which describe characteristics of the entire
structure. For example, x(t) denotes the vector of the We note that the "dot" product e]'.f(7) is a scalar, which
displacements at the degrees-of-freedom of the struc- we may think of as the projection of the vector applied
ture. force on the ]th mode shape.
K. Gordgs / Dynamic analysis for piping systems tO1

5. Solution by modal s u p e r ~ n mode/. Then


t
With this definition of the modal amplitudes, I show
ym/(t) = CO! f sin CO/(t - r)gm(~) d r . (6.2)
in [4] that the solution of the equations of motion
¢=0
(2.0) is given by
And we denote by Pra/the ruth component of the
x(t)= ~]~ elY/(t). (5.0) M-vector in the first parentheses of eq. (6.1), which
all modes
we call the "scalar participation factor" for degree-of-
In [4] I show that for most practical calculations, it freedom m of the independent components of the
is not necessary to use all the modes. Only "non-rigid" loading for mode/. Then
modes need by included, while the effect of the higher
Pml = (UTe/),n • (6.3)
"rigid" modes is reflected by using the "rigid mode
forces" as loads. Then by eq. (6.1)
The integral expression (4.0) for calculating the
modal amplitudes is called a "convolution integral". y/(t) = COl' ~/'n emf/mj(t). (6.4)
DST Computer Services has developed a very efficient
computer routine to evaluate convolution integrals,
which has been implemented in the program EBASCO/
PIPESTRESS. 7. Shift of coordinates

The methods described in the preceding sections


can be generalized by introducing a "shift" of coor-
6. Participation factor/resl~nse method
dinates as follows.
Suppose v -- v(t) is any N-vector such that v(O) -- ~(0) :
Suppose that the applied force f(t) can be repre- O, and define the N-vector u -- u(t) by
sented as a linear combination of M g N independent
components. Then there exists an M-vector g(t) and a u(t) = x C t ) - v(t). (7.0)
time-independent N X M matrix U such that v(t) is the "shift" vector and u(t) is the "shifted" dis-
placement.
f(t) = Vg(t). (6.0)
Then substituting eq. (7.0) into eq. (2.0) we arrive
The components off(t) are the independent compo- at the "shifted" equations of motion
nents of the loading.
Then substituting eq. (6.0) into eq. (4.0) we obtain M~ + Ku -- f - M~ - Xv (7.1)

t u(0) = ~ ( 0 ) = 0 .
yj(t) = col' f sin coj(t - r) eTug(O dr If we denote by
1"=0
f v = f - M~ - x v , (7.2)
t
= CO[2(uTe/)T(CO// Sin CO/(/' -- r ) g C r ) d l " ) . ( 6 . 1 ) then we may think offv(t) as the "shifted" applied
T=o force, and we may define the "shifted" modal ampli-
tudes by
We note that in eq. (6.1) the expressions in the paten.
theses are both M-vectors, while their "dot" product t
is a scalar. In the following, we describe the compo- Yu/(t) = col I / sin co/(t - r) eTfv(r) de. (7.3)
nents of these M-vectors separately. 1-=0
We denote by 7m/(t) the ruth component of the
With these definitions we have
M-vector in the second parentheses of eq. (6.1), which
we call the "scalar response" for degree-of-freedom rn u(t)= ~ ep, v/(t ) . (7.4)
of the independent components of the loading for all modes
102 K. Gordis / Dynamic analysis for piping systems

8. Beam structures of" the member (q, r) at point r; Srq is the 6 X 6 trans-
fer matrix which transfers a 6-vector displacement at
We will apply the preceding results to beam struc- point q to a rigid-body equivalent displacement at
tures. The structural members of beam structures are point r; Xr=-XT(t) is the 6-vector displacement at
beams. Pipes and other types of "beams" have con- r. Xq=-Xq(D the same for point q. At each internal
tinuously distributed mass. The methods described in point r we have
this paper apply to finite dimensional structures.
Newton' Law
Therefore, it is necessary to concentrate the mass of the
structure at a finite number of points which are called
"mass-points". For practical calculations the time for
MrXr + ~ Frq =Fr, (8.1)
all m e m b e r s
solution depends on how many mass-points must be (q,~)
used to accurately model the continuously distributed
where Mr is the 6 × 6 non-negative definite mass ma-
mass. In [4] I describe methods for determining and
trix of inertial mass concentrated at point r. We
locating the minimum number of mass-points, and for
assume that there is no mass cross-coupling. Fr -
efficiently concentrating the mass of the structure at
Fr(t) is the 6-vector applied force acting at point r.
these points. For the purposes of this paper we simply
Substituting eq. (8.0) into eq. (8.t) and separating
assume that the mass is concentrated at the end-points
known and unknown terms, we obtain for each internal
of the beams, which are the "points" of the structure.
point r
We denote by (q, r) the member with end-points q
and r, and we say that the member (q, r) "connects"
Mr2, + KrqXr - ~ KrqSrqXq
the points q and r. all m e m b e r s internal mem-
We assume that the beam structure has six degrees- (q,r) bers (q,r)
of-freedom (three translational and three rotational)
: F~ + ~ KrqSrqXq • (8.2)
at each point. anchors
We assume that there are two types of points, (q,')
"boundary points" where the time-history of the six
The equations (8.2) may be summarized in the form
displacement components is given, and "internal
points" where the six displacement components are M~ + Kx = f , (8.3)
unknowns. (There are, of course, other possible cases, where if there are n internal points then M and K are
such as structures containing partially restrained points, both 6n × 6n and x and f are 6n-vectors.
but these cases are not difficult to treat using general- I show in [4] that for connected structures with at
izations of the methods described in this paper.) least one anchor the positive definiteness of the 6 X 6
An "internal member" is a member which connects member stiffness matrices implies the positive defini-
two internal points, and an "anchor" connects a boun- teness of K. It follows that K is non-singular.
dary point and an internal point. If we assume that both structure and loading are at
We assume that no member intersects itself or con- rest at time t = O, then eq. (8.3) is in the form of eq.
nects two boundary points, and that the structure is (2.0), and represents the equations of motion of the
connected and contains at least one anchor. beam structure.
We further assume that the structure is "elastic". Eqs. (3.0) and (3.1) define the natural frequencies coj
This means that the 6-vector "member forces" Frq(t) and the mode shapes ej of the beam structure. The
satisfy mode shapes ej are 6n-vectors.
Denote by Erj the 6-vector formed from the six com-
Hooke's Law ponents of ej corresponding to internal point r.
Frq (t) = Krq (Xr(t) - SrqXq (t)), (8.0) Then substituting eq. (5.0) into eq. (8.0) it follows
that for internal members (q, r)
where Frq-Frq(t) is the 6-vector "member force" act-
ing at mass-point r which causes the distortion in member Frq(t) = ~ Krq(Erj - SrqEqj) yj(t) . (8.4)
(q, r); Krq is the 6 X 6 positive definite stiffness matrix all m o d e s
K. Gordlx/ Dynamic analy~afor piping arystems 103

(Although eq. (8.4) applies only to internal members, 9. Beam stna:tur~ ~dth shift
in practice, this expression may be used for all mem-
bers of a structure. The "real" anchors can be treated For shift vectorv(t) we denote by V r - Vr(t ) the
as internal members by connecting them to the bound- 6-vector formed from the six components of v(0
ary with "dummy" rigid spings.) corresponding to internal point r, and denote by Ur =-
We may now discuss the computational aspects of Ur(t ) the 6-vector formed from the six components of
this type of time-history analysis. From eq. (5.0) we u(t) corresponding to internal point r.
see that if we have calculated the time-history of the Then substituting eq. (7.0) into eq. (8.0) we ob-
modal amplitudes at various time-steps, then we can tain for internal members (q, r)
construct the instantaneous displacements at "solu-
F,q = K,q (v, - S,q vq)
tion time-steps" using the modal amplitudes and the
mode shapes. (The "solution time-steps" are any sub- + K,q (V, - S,q V q ) . (9.0)
set of the time-steps at which the modal amplitudes
The first term on the right in eq. (9.0) is the force
have been calculated.)
There are two methods which may be used to cal- which causes the distortion of the member relative to
culate the instantanoeus member forces. the shifted coordinate system. The second term, which
The first method consists of using the instanta- we call the "shift force", is the force which causes the
neous displacements to calculate the member forces. member to assume its shifted "shape". Then for inter-
This method requires a "static" solution for each solu- hal mebers (q, r) using eq. (7.4) in eq. (9.0)
tion time-step.
The second method consists of calculating the time- Frq(t) = ~J K,q(E,/- SrqEcd)yui(t )
all modes
independent "modal distortion vectors" which are
defined by + Krq(Yr(t ) - Srq Vq(t)). (9.1)

K,q (E,/ - S,qEqj) (8.s) A shift for which all internal points are moving rigidly
with respect to each other is called a "rigid shift". Any
for each internal member (q, r) for each mode j. rigid shift may be expressed as
This method requires a "static" solution for each
mode. The member forces are then calculated by eq. v,(t) = S, b Vb(t ) , (9.2)
(8.4) for each solution time-step. where Vb - Vb(t) is a 6-vector which we may think of
In [4] I show that for practical calculations only as describing the motion of an arbitrary point b not
the "non-rigid" modes need to be used. The effect necessarily on the structure. The shift force is zero
of the higher "rigid" modes is reflected in the "rigid for all internal members of a connected structure if
mode force" solutions, which require an additional and only if the shift is rigid.
"static" solution for each independent component of I outline the argument which demonstrates this
the loading. important property of shift forces. First, if the shift
From the point of view of cost, the choice between is rigid, then by eq. (9.2)
these two methods depends upon which method
requires fewer "static" solutions. V r - S r q V q = S r b V b - Srq,.,£qbVb = O. (9.3)
Finally, we state several useful properties of trans- Thus, for rigid shifts the shift force is zero for all
fer matrices. internal members.
s,, = 1 (8.6) Conversely, if the shift force is zero for every inter-
nal member, then, since member stiffness matrices are
SrqSqp = Srp. (8.7) positive definite and, hence, non-singular, for any
It follows from eq. (8.6) and eq. (8.7) that Ser3,q = 1 internal member (q, r) we have
and hence that transfer matrices are non-singular and Vr = S r q V q , (9.4)
that
and we see that r is moving rigidly with respect to q.
= sq, . (8.8) Since the structure is connected, using the symmetry
104 K. Gordis/Dynamic analysis,forpiping systems

and transitivity of the relation "moving rigidly with If the shift is a rigid shift, using eq. i9.2
respect to", we infer that the components of the shift
vector corresponding to the internal points are movo ( K v )r = G KrqSrbVb - ~ KrqSrqSqbVb
ing rigidly with respect to each other. all members internal
(q,r) members
In the special case where the shift vector consists (v.~)
of the structure moving rigidly with respect to one of
its boundary points a, we have for every internal ~__a KraSrbG
point r anchors
(~r)
G : &~x~ , (9.5)
(l l.l)
where X a =-Xa (t) is the 6-vector time-history displace-
ment at boundary point a. 0 if no anchor is connected to point r .
We denote by (KBX B - Kv)r the 6-vector formed
from the six components O f K B X B Kv corresponding
10. Boundary movement loading to internal point r.
Tlien for rigid shifts, using eqs. (10.1) and (11. i), we
have
If the loading is entirely due to boundary move-
ments then we may represent the applied force by
Kra(SraXa - SrbVb)
f(t) = KMCB(t) , (10.0) anchors
(~r)
where if there are n internal points and na boundary
(K~xB - Kv)~ = (1 t.2)
points, then K B is the 6n × 6n B boundary stiffness
matrix, x B - x B ( t ) is the 6nB-vector of the boundary 0 if no anchor is connected to
displacements. We assume that xa(0) = An(0) = 0. point r .
We denote by (KaXB)r the 6-vector formed from the
From eq. (11.2) we see that if
six components OfKBX B corresponding to internal
point r. Then from eq. (8.2) SraXa = &bV~ , (1 1.3)
for all anchors a then
KraSraXa
anchors KBx B - Kv = 0 . (1 t.4)
(a,r)
But eq. (11.3) is equivalent to
(KBXB) r = (10.1)
x a = Sa~ G (1 1.S)
0 if no anchor is connected to point r .
for all anchors a, which means that all boundary points
are moving rigidly with respect to point b, which im-
plies that all boundary points are also moving rigidly
11. Boundary movement loading with rigid shift with respect to each other. It follows that eq. (11.4)
holds if all boundary points are moving rigidly with
respect to each other, and if the shifted coordinate
For the shift vector v(t) we denote by (Kv)r the system is moving rigidly with respect to the boundary
6-vector formed from the six components of Kv cor-
points.
responding to internal point r.
Then comparing eqs. (8.3) and (8.2) we see that
12. Bound solutions
(Kv)r = G KrqVr- ~ KrqSrqVct.
all m e m b e r s internal (1 1.0)
(q, r) members The ASME code requires that the solution for
(q,r) dynamic load cases be combined with other cases, such
K. Gordls / Dynamic analy~ for p/p/rigsystems i 05

as dead-weight, thermal expansion, etc. From a practi- greater than "true" bound solutions calculated from
cal point of view, to comply with this requirement it time-history solutions.
is necessary to construct one or more bound solutions Second, we see from eq. (12.0) that the best pos.
for the dynamic load cases. Two types o f methods sible bounds are obtained from eqs. (12.1) or (12.2)
which can be used to construct such bound solutions using
are described below. y~ound = y~aX .

12.1. T~me-history method


13. Maximum modal amplitude method
Using one of the methods available to calculate
the time-history solution of the equations of motion, This method consists of calculating the time.his-
we may construct one or more bound solutions tory of the modal amplitudes using eq. (4.0) for every
from the intantaneous solutions. For example, the time-step for which applied force data is available. The
time-history solutions may be calculated by modal bound for the/th modal amplitude is simply the maxi-
superposition, as explained in this paper. mum absolute value for the corresponding modal
amplitude time-history.
12.2. Modal amplitude bound method I make two observations about this method. First,
this method is identical to a true time-history solution,
Alternatively, we may construct a bound solution up to the step of finding the time-history of the mo-
as follows. Denote by y~OUnda bound for the absolute dal amplitudes. We see from eq. (4.0) that this method
value of the/th modal amplitude, and by y ~ the requires knowledge of both the mode shapes of the
maximum absolute value of the/th modal amplitude. structure and the time-history of the loading. Second,
Then this method yields the best possible bounds for the
modal amplitudes. Therefore, for this method we may
y~ound ~ y~nax= max ly/(t)l, (12.0) write
t
y~ound =y~nax . (13.0)
and we see from eq. (5.0) that we may use as a bound
solution
14. Participation factor/response spectrum method
Ix(t)l< ~ le/ly~ °und , (12.1)
all modes This method is based upon the results of section 6.
Using the notation of section 6, define
and from eq. (8.4) for internal members (q, r)
~' = max l'}'m/(t)l, (14.0)
t
IF,q(t)I, ~ IXrq(F,,/-SrqEql)lYpUnd,(12.2) then by eq. (6.4)
all modes

and from eq. (9.1) for internal members (q, r) in terms ]Yl(t)l< ~o72~ 'Pm/I'T~m~= y~O~d. (14.1)
m
of the shifted modal amplitudes
The curve of 7 ~ x versus frequency is the "response
spectrum" of the ruth independent component of the
lFrq(t)l< ~ IKrq(Er/-SrqEql)lyvl~ und
loarling.
all modes
I make two observations about this method. First, it
+ maxl Krq(Vr(t ) - SrqVq(t))l. (12.3) is clear from eq. (14.1) that, in general, the bounds
obtained from this method will not be the best pos-
Before discussing methods for constructing bounds sible. In most cases, the method of section 13 will
for modal amplitudes, I make two observations. yield better bounds than this method. Second, we see
First, because modal components (and shift forces) from eq. (622) that the maximum response for any
are added "absolute" in the above expressions, the frequency can be calculated without reference to the
bound solutions constructed using eqs. (12.1), (12.2) mode shapes, and is, in this sense, independent of the
or (12.3) are conservative. They are, in general, structure.
106 K. Gordis / Dynamic analysis for piping systems

15. Classical response analysis then there are only three non-zero components of
eTM~, which may be represented as
The method of Classical Response Analysis is a
method for finding bounds for modal amplitudes in
eT Mv = Px "~ax + P yiJ2ay + P z / az " (1 5.7)
the case where the dynamic loading consists of all where Px/, Py/and Pzj are respectively the x, y and z
boundary points translating in parallel. In this case, components of the 6-vector
a rigid shift is made into the coordinate system which
M~L~j, (15.8)
is translating in parallel with the boundary points. internal
Using eqs. (11.4) and (10.0) in eq. (7.2) we have points r

f~ = -M~ (15.0) Pxj, Pyj and Pzj are the familiar x, y and z participa-
and from (7.3) tion factors for mode j. Substituting eq. (15.7) into
eq. (15.1) we obtain
t
yoj(t)=-wf ' f sin ¢oj(t- OefM~(Odr. (15.1) yvj(t) = -col 2 (Pxjraj(t) + pyj~j(t) + pzjraj (t)) (t 5.9)
7-=0 where
Then t
rax/(t)=co]y sinco](t- OXax(r) dr (15.10)
eTMo = ~ E}Mr("r. (15.2) 7-=0
internal
points r with similar definitions for ~ i and ~]. Eq. (15.9)
Using eq. (9.5) and the symmetry of Mr we have shows that we may use
ybound =
uj 6o72([Pxji~x~ax + [pyj[~y~aX + [pzjl~r~ax)
eTMv = ~ (MrErj) T Sra)(a. (15.3) (15.11)
internal
points r rax~aX, r~,}nax,~ a x are the familiar x, y, z acceleration
If we assume that the boundary movement is purely response spectra values for mode j.
translational then Since we have used a rigid shift, the shift forces are
zero for all internal members, and we may use eq.
SraXa = X a . (15.4) (15.11) in eq. (12.3) to obtain bounds for the member
So that forces in internal members.
In this case, we can also arrive at bounds for mem-
eTM~ = ( ~ (MrErj)) T X a . (15.5) ber forces in anchors. Thus, if r is an internal point and
internal a is a boundary point, then using eqs. (7.0), (9.5) and
points r
(7.4) we obtain
eTM~ is represented in eq. (15.5) as the "dot" product F~. = r , . (X, - S~aXa)
of two 6-vectors.
If we denote the components of the purely trans- = Kra(Ur + Vr - SraXa)
lational boundary movement by
= K~a (U, + SraXa - - SraXa)

1X ax
= Kra Ur

Xa = Xazl , (15.6) = ~ KraErlYvj, (15.12)


I all m o d e s
0 and, therefore, we have
i0 Ifra(t)l< ~ g-Lra~r]'.gu]
b" [ "b°und (15.13)
L0 all m o d e s
K. Gordi: / Dynamic analyttsfor piping systems 107

We may summarize the foregoing as follows: if all 16,1 Rigid shift method
the boundary points are moving rigidly with respect
to each other, and if we shift to a coordinate system In this method we shift into a coordinate system
moving rigidly with respect to the boundary points, which is moving rigidly with respect to an arbitrary
then point b not necessarily a point on the structure.
- since we have made a rigid shift, all the members From eqs. (7.3) and (10.0) we see that to calculate
shift forces are zero. This enables us to use the bounds the shifted model amplitudes we need to analyze the
for the shifted modal amplitudes to calculate bounds expression
for member forces;
e~f~ = eT ( K . x . - M~ - gv)
- since KBx n - Kv = 0 the shifted applied force
is -M0, which enables us to calculate simple bounds = e~(KBx B - gv) - e~/W~;. (16.2)
for the shifted modal amplitudes.
Then using eq. (9.2) we have
Note that the foregoing applies even if the rota-
tional components of the boundary motion are non-
zero, and even if there is non-zero rotational inertia.
e~M~= ~
internal
ET/MrVr= (
internal
EM,S,O,o)S, b
- If, in addition, the boundary motion is purely " points r points r (16.3)
translational, and thereis no rotational inertia, then where Io is any one of the levels. We note that the
we arrive at the familiar Classical Response Analysis 6-vector
solution.
Finally, we note that this method depends on the Sa, o# b (16.4)
assumption that the damping is zero or negligible.
describes the acceleration of the shift at point aao.
Using (11.2) and (16.1)
16. Multi-level problems
e~(/CBx B - / c v ) = Z) egx, s,.xo- S,bVb)
anchors
If all the boundary points are not moving rigidly (a, r)
with respect to one another, then we see from eq.
(11.2) that for any rigid shift v(t) we must expect that
levels anchors
S.,bVb) (16.5)
K s x B - Kv ~ O . (16.0) I (a, r) at
level I
Therefore, for any rigid shift applied force will not
We note that the 6-vector
assume the simple form -M~ as was the case for Clas-
sical Response Analysis. On the other hand, if we Xa s - SatbVb (16.6)
require that the shift vector satisfy KBx B -- Kv = O,
then such a shift will not be rigid, and some of the describes the movement of the point a ! relative to the
member shift forces will be non-zero. shift.
These observations lead to two methods, the "rigid The expressions (16.3) and (16.5) suggest that we
shift method" and the "Clough and Penzien method" define the 6-vector participation factors andthe
for extending Classical Response Analysis to the multi. 6-vector responses as follows.
level case. I describe these methods in the following. The 6-vector "mass weighted participation factor"
First, I det'me the "levels" of a structure. A "level" P~t/for level I for mode / is defined by
is a set of boundary points which are moving rigidly
with respect to each other. We assume that the struc- P~ai = ~a sTatMrEr/ (16.7)
internal
ture has L levels, and we choose a "reference point" points •
in each level. We denote by at the reference point for
level l. and the 6-vector "stiffness weighted participation
If boundary point a is at level I then factor" P/~ for level I for mode / is defined by

Xa = SmfYat . (16.1) P~=snchors


~ (~r) 8T~tK,aEr/• (16.8)
at level I
108 K. Gordis / Dynamic analys& for piping ~vstems

The 6-vector "acceleration response" R]bi(t ) of level principle, no more complex to caict~tate thal~ the
l for mode j with respect to the shift Vb is defined by acceleration response spectra, they are not used u;
Classical Response Analysis calculations and, hence,
t
are not part of the "standard" procedure.
R]bi( t) = coi f sin wi(t - r)Satbfib(r) dr (16.9) We complete our discussion of iigid shift methods
7"=0
by examining tile "unshifled" case which .corresponds
and the 6-vector "displacement response" Rab/(t) of to
level I for mode j with respect to the shift Va is defined
by Vb(t ) = 0 {1(i. t 5)

t
in this case the acceleration response is zero, and cq.
(16.11) reduces to
sin ~o/(t - r) ()(at(r) - SalbVb(T)) dr.
7-=0
(16.10) yi(t) = 0o72 ~ PlKTRaad(t) (16.16)
levels
l
Then substituting eqs. (16.3) and (16.5) into eq.
(7.3) and using the definitions (16.7), (16.8), (16.9) where Ra, i(t) is the 6-vector "absolute displacement
and (16.10), we obtain response'r for level l for mode / defined by

y#(t) = co/- 2 [ eloJ


MT a
Rlo~/(t) + ~ KT d /-
levots Rm.(t)]. R ,j(t) = j
o
sin co](t - r)Xat(r) dr, (16.17)
(16.1 1)
l
7-=0
It follows that we may use
and we may use
ybound -2 r,~ ,T~amax ~ pK T/?dmax
oj =6o] II loft l~lobJ + levels --lj *'lbj ]
(16.12)
),bound =COi- 2 ~ [,pKlXRdmax
#' at/ (16.18)
levels
I
If the shifted coordinate system is moving rigidly with
Thus, it is quite simple to calculate bound solutions in
respect to level l o then Xat = SatbV ~ and from eq.
terms of unshifted modal amplitudes. However, if
(16.10) we have
level l is moving approximately (but not exactly) in
phase with level 10, then we may expect the bound
Rdoatol(t) = 0 . (16.13)
solutions calculated from eq. (16.12) to be less con-
It follows from eq. (16.11) that for single level prob- servative (and, therefore, more desirable) than the
lems, if the sfiifted coordinate system is moving bound solutions calculated from eq. (16.18).
rigidly with respect to the supports then
16. 2 Clough and Penzien method
-2 T a t
y oi(t ) = -a~ j t~tol R toatol ( ) . (16.14)
This method is suggested by Clough and Penzien
If there are no boundary rotations and there is no
in section 27-4 of their book "Dynamics of Struc-
rotational inertia, it is easily verified that (16.14)
tures" [11. The method is based on a shift for which
reduces to the Classical Response Analysis expression
the shift vector v(t) satisfies
(15.9).
The evaluation of the 6-vector participation factors Kv = KBXB . (16.19)
presents no computational difficulty. Nor does the
(We can characterize this shift vector as the instanta-
evaluation of the acceleration response spectra, which
neous static equilibrium configuration of the struc-
are simply bounds for the expression given in eq.
ture with respect to the boundary movements.)
(16.9). If there are no rotational components, the
Since K is non-singular, we may solve eq. (16.19) for
acceleration response spectra reduce to the familiar
v and using eq. (7.2) we obtain
values of Classical Response Analysis. On the other
hand, while the displacement response spectra are, in fv = -M~ = -MK-1Kn2B , (16.20)
K. Cordb / Dynamic analysis for piping systems 109

and from the symmetry of M and K [5], the above expressions concern only the "primary"
terms. In [4], I have derived very s~nple bounds for the
eTf v = --eTMK-XK~cB = - ( K - I M e / ) T KB3c a . (16.21)
" s e c o n d a r y " terms.
Using eqs. (3.0), (lO.l) and (16.1)

anchors 17. Conclusion

We compare the different methods described in


=-¢o;= ~-J ( ~ ET KraSrat)~at. (16.22) this paper.
levels anchors (a,r)
! at level i First, for multi-level boundary movement problems,
we compare the Clough and Penzien method to rigid
Then, using eqs. (16.8) and (16.9) in (7.3) we have shift methods. The Clough and Penzien method has
the advantage that it only involves familiar accelera-
yu](t)--co/
_ -4
. ~. P~K T Raad(t)
a
, (16.23) tion response spectra, while rigid shift methods require
levels
l displacement spectra. A further advantage of the Clough
and Penzien method is that it separates "primary" and
and we may use
"secondary" terms.
Next, we compare participation factor/response
y ~ u n d = ¢0;" ~ IP~I T Raa
amax
d • (16.24)
levels spectrum methods of section 14, 15, and 16 (includ-
I ing "Clough and Penzien" and "rigid shift") to the
In [4] I have derived the following interesting identity maximum model amplitude method of section 13 for
for any point b calculating bounds for modal amplitudes.
The advantage of participation factor/responce
~o~ ~ STrbMrEd= ~ ST~KraEr/ (16.25) spectrum methods is that response spectra may be
internal anchors evaluated for "levels", and then used for any struc-
points • (a,•)
tures which contain boundary points at these levels.
(16.25) may be used to establish the following rela- This advantage is relatively minor for problems for
tionship between the mass weighted and the stiffness which the time history of the loading is available. The
weighted participation factors. maximum modal amplitude method requires re-calcula-
tion of the convolution integrals (4.0) for each struc-
T .~
= co/- 2 . ~ . S~ta,:~/. (16.26) ture analyzed. But the evaluation of these integrals can
1@~15
i
be carried out very efficiently. Further this method
yields the best possible bounds, and may be applied to
It follows that for single level problems problems which involve applied forces as well as bound-
ary movements.
= ~72P~o/ (16.27)
Finally, we compare the methods which use bounds
and we see that for single level problems that (16.23) for the modal amplitudes to a true time-history solu-
is identical to (16.14). This, in turn, shows that the tion.
C1ough and Pcnzien method reduces to Classical In unusual cases where there are either very few
Response Analysis for single level problems with no time steps or very many modes, it is possible that
boundary rotations and no rotational inertia. true time-history may require fewer "static" solutions
The calculation of bound solutions for multi-level than modal bound methods. In the typical case, how-
problems using the Clough and Penzien method is ever, true time-history will require the same number
quite simple. The only spectra which are involved are of "static" solutions as modal bound methods, but
the familiar acceleration response spectra. A further will,in addition, require the construction by superpo-
advantage of this method is that it separates "primary" sition of instantaneous solutions at each solution time
(inertial) terms, and "secondary" (support movement) step, and will further requite a method of constructing
terms as foreseen by the ASME code. As in [2] and one or more bound solutions from these instantaneous
110 K. Gordis /Dynamic analysis for piping systems

solutions. Thus, in most cases, true time-history will [2] tt.J. Schmid and J. Ludwig, Design of' Piping Systems
require more calculation than modal b o u n d methods. with the Procedure of Multiple Support Excitation,
(unpublished manuscript).
On the other hand, the true time-history method will
[3] K.R. Leimbach, Multiple Support Excitation, unpublished
usually result in less conservative (and, therefore, more manuscript.
desirable), bound solutions. This " t r a d e - o f f ' should [4] K. Gordis, Modal Superposition Methods for solving
be evaluated on a case-by-case basis. finite Dimensional Dynamic Elastic Structural Problems,
in preparation.
[5] R.W. Wu, F.A. Hussain and L.K. Liu, Nucl. Eng. Des.
References 47 (1978) 273.

[1] R.W. Clough and J. Penzien, Dynamics of Structures


(McGraw Hill, New York, 1975).

Вам также может понравиться