Академический Документы
Профессиональный Документы
Культура Документы
Giovanni Garcea, Antonio Madeo, Raaele Casciaro Dipartimento di Modellistica per l'Ingegneria, Universit della Calabria, 87030 Rende (Cosenza), Italy
Abstract
Such we call Implicit Corotational Method, is proposed as a tool to obtain geometrically exact nonlinear models for structural elements, such as beams or shells, undergoing nite rotations and small smooth strains starting from the basic solutions for the 3D Cauchy continuum used in the corresponding linear modelings. The idea is to use a local corotational description to decompose the deformation gradient in a stretch part followed by a nite rigid rotation. Referring to this corotational frame and using standard change in the observer algebra we can derive, from the linear stress tensor and the deformation gradient provided by the linear theory, an accurate approximation for the nonlinear Biot stress and strain tensors which implicitly assure the frame invariance of the description. The stress and strain elds recovered in this way are then entered in a mixed variational formulation to obtain, nally, a nonlinear modeling in terms of standard generalized stress and strain parameters, in a form directly suitable for FEM implementations. The great potential of the method lies in its ability to recover objective nonlinear structural models by fully reusing information gained from its linear counterpart and so exploit in a quite automatic way the great experience and many results already available from linear theories. The applications regard the construction of 3D beam and plate nonlinear models starting from the Saint Vennt rod and Kirchho and MindlinReissner plate linear theories, respectively. A nite element implementation of these models, suitable for both pathfollowing and asymptotic postbuckling analysis, is reported, showing the eectiveness of the proposed approach for obtaining numerical solutions in nonlinear analysis. Dierent aspects of the FEM modeling are discussed in detail, including the numerical handling of nite rotations, interpolation strategies and the equation formats. Two mixed nite elements are presented, suitable for nonlinear analysis: a 3D beam element, based on interpolation of both the kinematic and static elds, and a rotation free thin plate element, based on a biquadratic spline interpolation of the displacement and piecewise constant interpolation of stress. Both are frame invariant and free from nonlinear locking. A numerical investigation has been performed, also comparing beam and plate solutions in the case of thinwalled beams. The good agreement between the recovered results with available theoretical solutions and/or numerical benchmarks, clearly shows the correctness and robustness of the proposed approach as a general strategy for numerical implementations.
Contents
1 The implicit corotational method
1.1 1.2 From linear to nonlinear analysis . . . . . . . . . . . . . . . . The Implicit Corotational Method . . . . . . . . . . . . . . . 1.2.1 Kinematical Preliminaries . . . . . . . . . . . . . . . . 1.2.2 Kinematics of a 3D Cauchy body . . . . . . . . . . . . 1.2.3 Constitutive equation for a 3D Cauchy body . . . . . 1.2.4 Use of linear solutions to set up a nonlinear modeling . 1.2.5 Basic elements of the ICM . . . . . . . . . . . . . . . . Some tutorial implementations of ICM . . . . . . . . . . . . . 1.3.1 Planar beam with rectangular cross section . . . . . . 1.3.2 Thin walled beam under axial force and torsion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 10 10 12 13 14 16 18 18 23
1.3
2.2
2.3
Nonlinear beam model based on Saint Vennt general rod theory 2.1.1 Obtain SV linear solution - statics . . . . . . . . . . . . . 2.1.2 Obtain SV linear solution - kinematics . . . . . . . . . . . 2.1.3 Apply corotational kinematics . . . . . . . . . . . . . . . . Nonlinear plate model based on MindlinReissner plate theory . . 2.2.1 Obtain linear solution - static . . . . . . . . . . . . . . . . 2.2.2 Recover nonlinear strains and constitutive equations . . . 2.2.3 Apply corotational kinematics . . . . . . . . . . . . . . . . 2.2.4 Relation with nonlinear plate modeling by Simo . . . . . . 2.2.5 Rotation free modeling for Kirchho thin plate . . . . . . Further comments and remarks . . . . . . . . . . . . . . . . . . . 2.3.1 Linear or quadratic recovery of Biot strains . . . . . . . . 2.3.2 Errors inherited from linear theory . . . . . . . . . . . . . 2.3.3 Linear assumptions in nonlinear analysis . . . . . . . . . . 2.3.4 Mixed versus compatible formulation . . . . . . . . . . . . 2.3.5 Obtaining more than quadratic local accuracy . . . . . . .
27
27 27 29 33 34 34 36 37 37 38 38 39 39 39 40 40 42 43
42
3.2
The asymptotic analysis . . . . . . . . . . . . . . . implementation of nonlinear ICM models . . . . . . Element interpolation in linear analysis . . . . . . Element interpolation in nonlinear analysis . . . . Updated Lagrangian or Corotational interpolation Mixed or compatible format . . . . . . . . . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . .
45 47 48 49 51 52
4.2
A nonlinear beam element based on Saint Vennt general theory 4.1.1 Some preliminaries on rotation algebra . . . . . . . . . . . 4.1.2 Handling the nonlinear beam model . . . . . . . . . . . . 4.1.3 Mixed nite beam element . . . . . . . . . . . . . . . . . . 4.1.4 Setting of the element frame . . . . . . . . . . . . . . . . . 4.1.5 Stress interpolation . . . . . . . . . . . . . . . . . . . . . . 4.1.6 Displacement interpolation . . . . . . . . . . . . . . . . . 4.1.7 Expression for the complementary energy . . . . . . . . . 4.1.8 Expression for the strain work . . . . . . . . . . . . . . . . 4.1.9 Energy variations . . . . . . . . . . . . . . . . . . . . . . . 4.1.10 Global handling of the beams assemblage . . . . . . . . . 4.1.11 Further remarks . . . . . . . . . . . . . . . . . . . . . . . A nonlinear plate element based on Kirchho theory . . . . . . . 4.2.1 Handling nonlinear thin-plate model . . . . . . . . . . . . 4.2.2 Mixed plate nite element . . . . . . . . . . . . . . . . . . Numerical validation . . . . . . . . . . . . . . . . . . . . . . . 5.1.1 Accuracy of ICM models . . . . . . . . . . . . . . . . . 5.1.2 The eect of postbuckling stress redistribution . . . . 5.1.3 Saint Vennt beam vs Kirchho plate modeling . . . . 5.1.4 Modal interaction test: Channel beam in compression . . . . . . . . . .
55
55 55 56 56 57 58 58 58 59 59 60 60 61 61 62 64 65 68 71 78
5 Numerical validation
5.1
64
6 Conclusion
82
Introduction
The nonlinear analysis of slender elastic structures requires appropriate nonlinear modeling to achieve reliable and accurate solutions. The use of a frameinvariant (or objective) structural model, where stress and strain elds are unaected by nite rigid motions of the body, is a necessary requirement for the analysis [1, 3]. While it is quite easy to satisfy frame-invariance for 3D bodies using the Cauchy continuum, e.g. by referring to GreenLagrange strain tensors, it can be dicult to obtain a coherent, simple enough modeling, for slender structural elements, such as beams or shells, which are more conveniently described as one or two-dimensional bred continua characterized by 3D displacements and rotations. For this reason in the last 40 years great eort has been made to develop nonlinear models of beams, plates and shells and their nite element implementation. The number of papers on this topic is impressive and this is also due, in the authors' opinion, to the complexity of the nonlinear modeling process and to the at times unsatisfactory results obtained by its FEM discretization. The great majority of beam and shell models are based on the socalled geometric exact theories such as those developed by Cosserat [4], Reissner [5], Antmann [7], Simo [8] and Wriggers and Gruttmann [11]. These are generally based on a direct assumption of constitutive laws in terms of stress/strain resultants. Models so generated are geometrically exact, that is exactly frameindependent, but are generally unable to describe important details of the corresponding linear modeling. This is evident, for example, in the classical AntmanSimo nonlinear beam model where the assumed simplied constitutive law lacks the shear/torsional coupling manifested by the 3D Saint Vennt linear solution [13] and more subtle nonlinear couplings associated to the section warping, such as the axial-torsional 2ndorder coupling recognized by Wagner [14]. A Galerkin reduction of the 3D nonlinear continuum equations, using the same simplifying assumption as the linear theory, could allow a more detailed modeling, at least in principle. This approach was followed, for instance, by Kim [40], Pai and Nayfeh [26, 27, 28], Petrov and Geradin [15, 16], Bradford [41], and Lin [42]. However the nonlinear models obtained by entering the displacement provided by the linear theory into a compatible variational formulation, like the nonlinear beam model by Petrov and Geradin [15, 16], appear somewhat overcomplex and also require ad hoc simplications in order to eliminate spurious locking. Models obtained by the use of problemdependent engineering nonlinear strain measures, like the beam and 3
shell models by Nayfeh and Pai (see [28]), are only aimed at an essential simplied modeling. On the other hand the availability of linear structural models for bred continua developed in the frame of the small displacements hypothesis is notable. These models came from well consolidated theories derived from 3D Cauchy equations through appropriate simplied hypotheses on the statics and kinematics of the body. The possibility of reusing available linear models, as a basis for generating appropriate nonlinear ones is then attractive due to the possibility of recovering all the eort spent in developing linear theories, without needing a separate ad-hoc derivation. The aim of this paper actually is to exploit this possibility through the use, in the continuum description, of the corotational approach initially proposed in a FEM context by Wemper [17], Belytschko [18] and Rankin [19, 20], and used by Argyris and Sharpf in their paper on the natural modes technique at the end of the '60's [43]. We will show that, by transferring this idea from the element to the continuum, we can derive a standard methodology to obtain a frameindierent nonlinear modeling which maintains all the richness of the embedded linear theory. More precisely, the method proposed here, which we call Implicit Corotational Method or simply ICM, is an automatic tool able to recover objective nonlinear structural models for structures undergoing nite rotations and small strains, starting from the corresponding linear one. From this point of view the method behaves at the continuum level, exactly like the standard corotational formulation at the nite element level. That is, while standard corotational formulation recovers a nonlinear nite element starting from a linear one, the ICM method recovers a nonlinear structural model starting from its corresponding linear one. The main idea is to associate a corotational frame (observer) to each point of the 3D continuum so allowing the motion in the neighbor of the point to be split in a pure stretch followed by a pure rotation, according to the decomposition theorem [1, 2]. It will be shown that, using the small strain hypothesis and rotation algebra, the linear stress and linear strain solution elds, when viewed in this corotational frame, can provide accurate approximations for the Biot nonlinear stress and strain tensors elds. Once the corotational rotation is appropriately dened, the local statics and kinematics of the model are recovered from the linear solution as a function of the stress/displacement resultants. Stress and strain elds are then introduced within a mixed variational principle in order to obtain the constitutive laws directly in terms of stress/strain resultants. This completes the ICM denition of the nonlinear model. Note that the nonlinear model so obtained maintains all the details of the linear solution, while its objectivity is guaranteed by the use of a change in observer algebra. Furthermore, the use of a mixed approximation in deriving its constitutive laws allows locking eects, which can come from small (highorder) incoherences between the stress and strain elds estimates to be avoided. It is also worth mentioning that the proposed approach does not require any adhoc assumption about the structural model at hand, nor depends on any particular parametrization of the rotation tensor, but actually behaves as a blackbox tool able to translate known linear modelings into the corresponding nonlinear ones. Moreover, the direct use 4
of a mixed (stress/strain) description provides an automatic and implicitly coherent methodology for generating models free of nonlinear locking eects [51, 52] in a format directly suitable for use in FEM implementations. To better illustrate the ICM features and to show its potential as a tool able to provide a nonlinear generalization of existing linear theories, the method will be implemented here in order to obtain a nonlinear 3D beam model, derived from the general Saint Vennt linear rod theory, and two nonlinear plate models, derived from the Kirchho and Mindlin plate theories. We will also show that from these new structural models, through simplifying assumptions, models already available in literature can be derived. Implicit Corotational Method is essentially based on two main steps: the general solution provided by the linear theory is exploited to recover a description of the nonlinear stress and strain elds, as viewed by a corotational local observer moving with the material neighbor; then this description is transferred to a xed global frame directly by using standard changeintheobserver algebra. In this way we obtain the following advantages: i) a general procedure is derived which can be implemented in a simple and automatic way in dierent contexts; ii) the recovered nonlinear structural models implicitly satisfy objectivity requirements (in other words, they are geometrically exact); iii) they also maintain all the details of the 3D linear solutions, without any deterioration; iv) nally, they appear in a form similar to that of their linear forerunners which is also convenient for numerical implementations. Since linear structural theories are widely available in literature, it is easy to obtain a large number of possible nonlinear models suitable for use in FEM analysis. Three ICM models have been derived, as the rst implementation examples of the method: a 3D beam nonlinear model based on Saint Vennt general rod theory and two plate/shell nonlinear models based on MindlinReissner and Kirchho plate theories, also discussing variants deriving from dierent assumptions or simplications. The method however can be easily applied to other structural contexts, once the corresponding linear solutions are available. The FEM implementation part has the purpose to show the potential of the method in dierent analysis strategies (pathfollowing or asymptotic analysis, and to prove its accuracy and robustness through numerical validation. We refer to 3D assemblages of beams and plates, using the beam and Kirchho plate models recovered. The paper is organized as follows: section 1.1 gives a brief overview of nonlinear analysis and explains the motivation for the present research; section 1.2 introduces our proposal of exploiting the corotational description for reusing linear solutions within a nonlinear context and provides the main algebra which is involved in the proposed method; section 1.3 presents and discusses some tutorial implementations to underline its main features and to show its eectiveness and reliability in recovering known noteworthy results from literature; sections 2.1 and 2.2 develop a nonlinear 3D beam model, based on general Saint Vennt rod theory, and a nonlinear plate model, based on the MindlinReissner linear plate theory, also giving a comparison with similar proposals available in literature; section 2.3 contains some comments 5
and further insights; In section 3.1 we present, the two analysis contexts which have been investigated, that is the pathfollowing analysis, according to Riks' incremental iterative strategy [48], and the asymptotic analysis, according to the Koiterlike approach described in [49]. The latter, because of its exacting requirements in terms of geometrical coherence of the modeling, provides a rigorous check on the accuracy obtained with the ICM approach. The general aspects of ICM implementation in FEM analysis will be examined in section . We will also consider dierent choices in the element description (i.e. Lagrangian or Corotational description) and in the interpolation of its internal elds. Frame independence plays an important role in element set up and the lack of it can result in spurious locking, we called nonlinear interpolation locking in [49] (also see [50]) which can signicantly aect the numerical solution. The equation format (i.e. mixed or purely compatible format) also plays an important role in the convergence of the iterative solution processes and in the accuracy of the extrapolation results (see [51, 52]). These topics will be discussed and the use of mixed FEM discretizations and a mixed format is suggested. Sections 4.1 and 4.2 will present and describe in detail two nite elements derived by the beam and Kirchho plate ICM models, which appear suitable for the analysis of general 3D assemblages. Both use a mixed format and a separate interpolation of the displacement and stress elds. The beam element is based on a quite standard interpolation, but paying attention to obtain a frame invariant description for the internal stresses and displacements. The plate element is based on the biquadratic spline interpolation, already used in [54, 55, 56, 57], which allows a description of its kinematics in terms of nodal displacements alone (without needing rotations) so noticeably simplifying the element algebra. The availability of analytical results or known reference benchmarks and the possibility of comparing, in the case of thinwalled beams, the results obtained by two dierent modelings (as a onedimensional beam or as an assemblage of two dimensional plates) allow, in section 5.1, a detailed discussion about the performance of the proposed elements and their eectiveness in test cases of technical interest. Finally section 6 summarizes the results obtained and introduces possible extensions and future implementations.
+D=FJAH
[u , ] u = 0 ,
u U , u T
(1.1)
where U is the manifold of the admissible congurations, T its tangent space and the prime stands for Frecht dierentiation with respect to u . Usually (and conveniently) the conguration is described making U a linear manifold, so T becomes independent from u. Condition (1.1) states a relationship between and u describing a curve in the {u , } space. The goal of the analysis is to determine this curve, the so called 7
equilibrium path, with particular accuracy in the evaluation of the maximum value c of the load multiplier. The potential energy can be split into two separate terms, the rst expressing the internal strain energy and the second the external load work, bilinear in p and u: [u , ] := [u ] p[] u (1.2)
We also know (see [51, 52]) that, with an appropriate choice of conguration variables u := {, d } and the corresponding choice p := { , q }, these terms could be conveniently written in mixed form as
[u ] := [ ] + [d ] ,
p[] u := q [] d + []
(1.3)
and d being the stress and displacement elds, q and the external forces and distortions, [d ] the compatible strain eld associated to d through kinematics, and [ ] the so called complementary strain energy which is usually assumed as quadratic in : 1 (1.4) [ ] := C 2
2
C being the positive denite bilinear compliance operator. The [d] function, expressing the kinematical relationship between displacements and strains, plays an important role in the analysis. Assuming the [d] relationship is linear, we obtain a linear formulation. So we can translate from a linear to a geometrically nonlinear formulation simply by referring to a proper nonlinear expression for the geometrical relation [d], which also implicitly denes the stress as workassociated with . The invariance from superposed rigid body motions, that is objectivity, is an essential prerequisite for function [u]. This requirement is easily satised if referring to the 3D Cauchy body, simply by taking u as the displacement eld u[X ] and as the GreenLagrange strain tensor eld g [X ], X R3 being the point material position (by this choice, will be the second PiolaKirchho stress tensor eld g , workassociated with g ). In this way, the main dierence between the linear and the nonlinear modeling lies in the use of g in spite of the linear strain tensor L . Tensor g , while ensuring objectivity, is a simple quadratic function of the displacement gradient u[X ], so the setup of the nonlinear model does not present particular diculties. In many cases of practical relevance, such as in the case of beam or plate elements, there is, however, the need to refer to a more articulated but reduced description which denes the body motion through control quantities dened over a one dimensional (the beam axis) or two dimensional (the median plane of the plate) domain. In linear analysis, this reduction is directly derived, through appropriate simplifying assumptions, from Cauchy basic solutions. In this way, the original threedimensional problem is converted to a twoor onedimensional problem, which retains all the signicant aspects of the 3D behavior but leads to simpler equations, much more easily managed by FEM technology. As a result this modeling approach is really
convenient and generally used in linear analysis, but its extension to nonlinear analysis is, however, not so obvious and current proposals for nonlinear models available in literature have disadvantages. Approximate "2ndorder" models can be obtained by adding some 2ndorder terms to the linear modeling aimed at capturing the main nonlinear eects. While based on some heuristics, this leads to quite simple formulations, suitable for FEM implementation, and therefore such models are widely used in FEM practice. However simplied models generally lose objectivity, so, while acceptable in some simple contexts, they can be unreliable in general and, in any case, need to be coupled in actual computations with suitable UpdatedLagrangian or corotational strategies to reduce, or eliminate in some element average, the resulting errors. Geometrically exact models like those proposed in [5, 8, 7] are generally dened in terms of the rigid motion of the cross section (in the beam case) or the transversal ber (in the shell case) and simplied "essential" constitutive laws. The use of an exact kinematics ensures the full objectivity of these models. However, while using the same description format as their linear counterpart, they are generated autonomously by adhoc assumptions and appear little related to 3D Cauchy nonlinear continuum equations and somewhat poorer with respect to the richness and complexity of the 3D behavior. In particular, the section warping, the stress distribution and other relevant details characterizing the linear 3D solution are completely neglected in this approach, and possible generalizations are not so obvious and can lead to such complex expressions as to be of little use in practical applications. Other approaches, like those followed in [40, 15, 16, 41, 42], involve a compatible Galerkin approximation based on the linear solution, by writing the strain energy in terms of GreenLagrange strain tensor g and assuming the displacement eld locally dened by the Cauchy solutions of the related linear problem (e.g. the Saint Vennt or the Mindlin solution). As shown in the cited papers by Petrov and Geradin, this approach can recover, at least in principle, all the details of the linear solution. The use of the g tensor also ensures frameinvariance. Note, however, that this modeling tends to produce a disturbing locking due to the presence of some spurious highorder terms in the energy expression, which have to be aposteriori deleted through simplications or apriori avoided by the use of simplied "engineering" strain measures adapted to the problem at hand and "ad hoc" rotation parametrization as done in [26, 27] (see also [28] and references therein). The problem will be discussed in detail in Section 2.3. We only mention here that the compatible format used in the modeling plays an important role in this locking eect. It also produces an algebraically complex formulation, illsuited for nite element implementations where it tends to produce interpolation and extrapolation locking (see [51, 52] for a detailed discussion of this topic). The corotational approach, based on polar decomposition theorem [1], because it allows decoupling the kinematical coherency from constitutive laws, represents a potentially suitable way for generating coherent nonlinear models, particularly when, as in the case of beams or shells, rotation variables are directly involved in the description. A corotational framing is actually used or, at least implied, in all 9
proposals for nonlinear modelings, even if it is generally associated to the transversal section (or transversal ber) and it is mainly used for providing a-priori geometrical independence from its rigid body motion. So the full potential of the corotational approach in generalizing linear theories in full detail, seems still not to have been completely fullled. The need for an organic methodology, which could give the derivation of the nonlinear model fully exploiting all the information gained by its linear counterpart through a black-box procedure which avoids heuristic assumptions for each particular case, is clear. However, great attention to minor but important details such as the parametrization of rotations, the equation formats and other implementation details which prove to be crucial for FEM implementations, is also necessary in order to obtain models suitable for practical applications.
xb = xa + v ba
We call rotation a second order orthogonal tensor Rba SO(3) (that is, in the Lie group SO(3) := {R : RT R = RRT = I , det(R) = 1}, see [22]), which moves the point from a position xa to a position xb according to the rule
xb = Rba xa
10
Two successive translations, from xa to xb and from xb to xc , obviously combine with a summation rule, while two successive rotations combine with the product rule, i.e. v ca = v cb + v ba , Rca = Rcb Rba We also know that, when applied to a set of material points Pi , both translations and rotations do not change their distances, i.e. the length of vectors xij := xj xi nor the angles between these vectors, nor their relative orientation (righthanded triad are transformed into righthanded triad). That is, a movement obtained by combining constant translations and rotations will be a rigid motion. A rigid motion obtained by a translation v followed by a rotation R can be viewed as a change in the observer which takes a rotation Q = RT followed by a translation c = v . Denoting the positions viewed by the original or by the moved observer with or without a superposed bar, we obtain
x = QT (x c) , x = Q x +c
while material positions are obviously unaected by the change in the observer. Displacements will be dened as the dierence from the current and the reference positions, so denoting with u := x X and u := x X the displacements viewed by the two observers, these will be related by
u = QT (X + u c) X
while the rotations are related by
= QT R R
We will also have
u b u a = QT (X b X a + ub ua ) X b X a
and
b R a = QT ( R b R a ) R
It is worth mentioning that, assuming a reference frame with origin O and base vectors {e1 , e2 , e3 } in R3 , both translations and rotations can be dened in terms of three scalar parameters. A convenient choice is to refer to translations by their (covariant) components vk := v ek (1.5e) and rotations by their axial vector w. Denoting with w3 w2 w1 W := spin (w) := w3 w2 w1
(1.5f)
the skewsymmetric tensor associated to w so that w u = W u, u R3 , and exploiting the internal constraints in SO(3), we obtain the so called exponential map
R := I + [w] , [w] :=
n=1
1 Wn n!
(1.5g)
11
x[X ] := X + u[X ]
The deformation gradient F denes the transformation law for all the innitesimal material vectors dX starting from X in the innitesimal spatial vector dx
dx = F dX
F [X ] := x = I + u
(1.6)
I being the identity tensor and () := ()/ X the gradient with respect to the material position X . From the decomposition theorem (see for example [1]) the secondorder tensor F [X ] can be decomposed into a unique product of a rotation tensor R[X ] and a symmetric, positive denite, stretch tensor U [X ] as: F = RU
(1.7)
= QT R = I ) Recalling eq.(1.5b), the rotation can be reduced to the identity (R simply by a change in the observer making Q = R. So, an objective description requires that the strain measure be independent from R. This requirement is not satised, in general, by the so called linear strain tensor L which is obtained by splitting the deformation gradient u into its symmetric and skew-symmetric parts E :=
and making
1 2
(u + uT ) ,
W :=
1 2
(u uT )
(1.8) (1.9)
L := E
Possible objective strain measures are expressed in the form
(n) :=
1 (U n I) n
which, for n = 1 and n = 2, provides the Biot strain tensor b and the GreenLagrange strain tensor g , respectively:
1 b := RT F I , g := (F T F I ) 2
(1.10)
The latter can be easily evaluated from u through a simple quadratic expression
g = E + uT u
2
(1.11)
12
and so it is most frequently used in applications. However, its identication in terms of linear strain, i.e. g L , requires ||u|| 1 which implies too restrictive hypotheses about the expected solutions to be used in practice. Biot strain b seems to be a better candidate for that purpose. In fact, expressing R in the form (1.5g) we obtain the relation
b = E + (T u + uT T )
2
(1.12)
and therefore we have b L when 0, which is simply obtained by an appropriate change in the local observer making Q R, without needing any further assumption. Anyway, assuming b and are small enough, we could evaluate b through the secondorder approximate expression
b := E + (EW W E W 2 ) b
2
(1.13)
which is simply obtained by making W . Note that denitions (1.10) also provide
2 b (2 b = g 2 b = b E ) 2 2 2 2 so the resulting error in using eq.(1.13) can be evaluated as (2 b E )/2 O [||b || + 2 4 ||b || + || ||]. In the present work, we deal with problems characterized by large displacements but small (smooth) strains, so we can generally assume ||b || 1. The goal in our proposal is to use a suitable change of reference to also render the rotation residual 0 or, at least small enough to allow b to be evaluated through eq. (1.13). 1 1
13
B being the reference body volume. The linear solution is characterized by the complementary energy L [ L ] := 1 2 L C L L , C L being the elastic compliance 4thorder tensor. Taking variation with respect to L , we then have L [ L ] := C L L
So, assuming the same relationship between b and b , i.e. (1.15)
b [ b ] := C b b , C b C L
(1.16)
the identication b L , being L dened by (1.15) and b by eq.(1.13), will imply a local compatibility error
b C b b
1 2
(1.17)
which can be taken as negligible for small . It is worth mentioning that the possibility of assuming, also in the nonlinear modeling, the same constitutive law as used in the linear modeling strictly depends on our choice of referring to Biot tensors. In fact, when referring to Green-Lagrange tensors, the assumption
g [ g ] := C g g , C g C L
will imply
g g [ g ] O(||2 b ||)
modeling in specic structural contexts: the rods theory of Saint Vennt, the Vlasov theory for thin walled structures, the plates theory of Mindlin, to cite only some classical results. All these results have been obtained with the assumption that displacements (rotations) are small enough to allow us to identify the deformed conguration with the undeformed one. So their derivation implies the use of an appropriate frame, that is of an appropriate observer, suitable to lter rigid rotations in order to minimize this dierence. Therefore, the corotational idea is in some way implicit in linear theories. To better explain this concept, we can refer, as an example, to the Saint Vennt rod theory. The general solution we obtain, expressed in terms of the stress eld over the cross section domain and the out of plane warping of the section, is naturally referred to a local Cartesian system {x, y, z }, such that the xaxis be orthogonal to the (average) section plane and y and zaxis aligned to its principal directions, that is to an observer aligned to the section. Actually, apart from the disturbance produced by the section warping, which is somewhat small being related only to the distortion of the longitudinal bers due to torsion and to the shear strains, all the quantities involved in the Saint Vennt solution can be identied as corotational. In more detail, the Saint Vennt stress solution L over the section can be quite naturally identied as a rstorder accurate approximation of the nonlinear solution in terms of Biot stresses b [y, z ]. Analogously, the displacement solution uL [y, z ] can be identied as a rstorder accurate approximation of the corresponding nonlinear solution. The identication b [y, z ] L [y, z ] should require [y, z ] 0 to obtain full geometrical coherence. Actually, this condition cannot be satised exactly, because of the pointwise dierences in rotation due to the section warping, but we can generally assume a small enough to allow Biot strains to be obtained from eq.(1.13), by introducing the expressions for E and W directly provided by the linear theory. The previous discussion can easily be generalized, in particular for the nonlinear modeling of bred continua such as beams or shells whose description is obtained by splitting the original 3D material reference X into a two or one dimensional abscissa z lying on a ber (the cross section, in the case of beams, or the transversal ber, in the case of shells) and a one or twodimensional abscissa s lying on its support (the line axis or the middle surface, respectively), depending on the model considered. The linear modeling denes both stress and displacement elds in terms of generalized strength t[s] and displacement (translation/rotation) d[s] parameters:
(1.18)
The nonlinear model is then generated, by identifying the linear solution as Biot stress eld, taking b [z, s] L [z, t],and by rewriting the second relation, through [s] such that the the use of an appropriate local corotational change from d[s] to d ] into Biot strains could also be recovered by introducing with u L = uL [z, d eq.(1.13). We can introduce these elds into the expression of the potential energy (1.2)(1.4), so completing the modeling. 15
This process will be better detailed in the following subsection. We only note here that, apart from the use of the quadratic formula (1.13), all the nonlinearity of the formulation reduces to the transformation rules relating the kinematical parameters [s], which is governed by purely geometric laws (see section 1.2.1). It d[s] and d is quite easy, through rotation matrix algebra, to treat the rotation Q[s] of the observer as a nite rotation, without introducing any further approximations. In this way we can obtain a nonlinear modeling which is able to both exploit full details of the corresponding 3D linear solution and to satisfy objectivity requirements with respect to the rigid motion of the ber exactly.
b := L , b := E + (EW W E W 2 )
2
(1.19c)
The latter can reduce to b := E , if we can assume W to be suciently small. 3. Recovering constitutive equations The constitutive equations, in terms of ber generalized parameters t[s] and [s], are recovered by entering the corotational tensors into the mixed form of d 16
the strain energy (1.3). When writing the strain work and the complementary energy in the form (1.19d) := W [s] ds , := [s] ds the local strain work W [s] and the local complementary energy [s] will be dened as W [s] := b [t] b [d] dS (1.19e)
[s] :=
1 2
S [s]
S [s]
b [t] C b b [t] dS
(1.19f)
where C b is the compliance tensor that the ICM assumes as dened by the linear theory C b C L . Performing the integration, the previous equations as: can be rewritten in terms of generalized parameters t and d ] , [s] = 1 tT Ht W [s] = tT [d (1.19g) 2 ] being the generalized strain, work-conjugate with t, and H a generalized [d compliance operator. The local constitutive law is then obtained directly from eqs. (1.1) and (1.19g), by dierentiating with respect to t:
] = H t [d
(1.19h)
4. Complete the nonlinear modeling by changing in the observer algebra To complete the nonlinear modeling we only need to express the generalized strain [s], obtained in step 3 as a function of corotational displacement pa[s], in terms of d[s]. As these are related by the change in observer rameters d relationships (1.19a), the frame independence, is assured by denition. It is worth noting that the proposed method provides a blackbox procedure for dening a nonlinear model, without needing any heuristic choices or athand approximations but only the availability of the corresponding linear modeling. The model will satisfy the objectivity requirements with respect to a rigid rotation of the ber exactly, while exploiting full details of the linear solution, including nonlinear eects due to the ber warping which are kept by eq.(1.13) within 2nd-order accuracy. Note also that, being based on the mixed variational condition (1.1) through a separate description for the stress and displacement elds, it provides in general a mixed approximate solution. Apart from simplied assumptions already contained in the linear theory, the error is, however, only produced by the pointwise dierences (1.17) in the strain as evaluated from displacement eld or from the stress eld, which is only related to the rotation residual [s] due to warping. Moreover, its eect is further zeroed on average by the use of integral formulation (1.19e)(1.19h), so it can be considered as negligible. Obviously, accepting less accuracy, a series of variants of the method are possible, such as the use of a linear approximation for b (b := E ), by making 0. Some of these variants will be discussed in the following section. 17
x2
x1 S[s]
y S[s] s Q[s]
L :=
1 y N [s] M [s] , A J
L :=
,y [y ] T [s] A
(1.20a)
18
where, according to the usual notation, A = b h and J = b h3 /12 are the area and inertia of the section, [y ] is the so called stress function expressed in our case by 3 h2 y 4 y 3 , dA = 0 , ,y dA = A (1.20b) := 2 h2 S S and N [s], T [s] and M [s] are the axial, shear and exural strengths dened, as usual, by
N := M :=
S S
L [s, y ] dA , T :=
S
L [s, y ] dA
(1.20c)
y L [s, y ] dA
Furthermore, denoting with u 1 [s, y ] and u 2 [s, y ] the components of the displacement eld in the spatial system {x 1 , x 2 }, these are locally (i.e. in a neighborhood of S ) given by u 1 [s + ds, y ] = u [s + ds] [s + ds] y + L [s] w[y ] (1.21a) u 2 [s + ds, y ] = v [s + ds] where u [s + ds] and v [s + ds] are the average translation of the section in x 1 and x 2 directions u [s + ds] := L [s] ds (1.21b) v [s + ds] := ([s] + L [s]) ds + O(ds2 ) ,
[s + ds] is its average rotation [s + ds] := [s] + L [s] ds + O(ds2 ) , w[y ] is the so called warping function dened by 1 w := /k y , k := ,2 dA = 6/5 A S y
(1.21c)
(1.21d)
k being the so called shear factor, and a rotation parameter which denes the orientation of the local frame chosen in order to maximize accuracy: for = 0 (i.e. [s] = 0) the frame is locally aligned to the section; for = L (i.e. v ,s [s] = 0) is aligned to the axis line. The previous expressions allow the recovery of the displacement gradient eld u[y ] = E [y ] + W [y ] on current section S . We obtain [ ] L L y L ,y /2k E= sym. 0 (1.21e) [ ] 0 L (,y /2k 1) W = skew 0
From the previous equations, we have
(1.21f)
] and W [S ] is so, apart from the alignment parameter , the local kinematics E [S completely dened by the local derivatives u ,s , v ,s , , s of the translations u [s], v [s], and of the rotation [s] associated to the section. It is also convenient to introduce the quantities [s + ds] := [s + ds] [s] (1.21g)
Obviously we have [s] = 0 and ,s = , s . These conditions implicitly dene the alignment of the local corotational observer associated to the section s. In fact assuming [s] be a nite rotation angle associated to ber S , as viewed by a xed observer, and its value viewed by the local observer, this implies a relative alignment angle [s] [s] = [s] between the two observers and denes the rotation matrix [ ] cos sin Q[s] := (1.21h) sin cos relating the two observers.
E and G being the normal and shear elastic modula of the material. Conversely, the Biot strain eld will be obtained by substituting expressions (1.21e) into (1.13). The evaluation of internal work (1.19e) provides W [s] := (11 11 + 212 12 ) dA = N + T + M (1.22b)
S
where the strain parameters , and are dened by: 19 2 1 := L + L + L + 2 48 2 1 := L L ( + L ) 2 := L that is, recalling the relations (1.21f) and (1.21g) 19 2 10 5 := u ,s + v ,s + v ,s 2 48 48 48 1 := v ,s u ,s ( v , s + ) 2 := ,s 20
(1.22c)
By relating expressions (1.22a) and (1.22b) through the Clapeyron equivalence 2c = W , we also obtain
N = EA ,
T = kGA
M = EJ
(1.22d)
which provides = L /2, and corresponds to an average alignment between the section and the axis line. By this choice we obtain
,2 ,s , := ,s := u ,s + v s , := 2 v
12
(1.24)
However, other choices are also possible, e.g. the choice = 0 (align to the section) or = L (align to the line axis). Dierent choices only reect the alignment of the corotational frame, so in the treatment of the average rotation 1 2 L due to warping, which is accounted for exactly through eq.(1.23) by the choice = L /2 and only to a 2ndorder accuracy by the other choices, while the use of corotational algebra (1.23) always assures, by denition, frame independence with respect to a rigid rotation of the section. Dierent choices only produce a small O3 (L , L ) dierence in the nal expression of [d], so they can be considered as equivalent, within our assumption of small strains, and the actual choice could be suggested by practical convenience. A discussion about this point and some further insights into the topic will be given in the following subsection. 21
Further insights
Some further comments and remarks are useful. 1. With regard to slender structures, we can expect the T [s] will be much smaller than N [s] in practical contexts and then assume . So, the quadratic term 2 v ,2 s /12 = L /48 in eq.(1.24) will be really very small and can be considered as irrelevant. If it is obliged we obtain, through eq.(1.23), = (1 + u,s ) cos + v,s sin 1 = 2 (v,s cos (1 + u,s ) sin ) (1.25a) = ,s Note that this approximation corresponds to making ,y /k 0 in the expression (1.21e), that is, because of eq.(1.21d), assuming a linear warping w[y ] y in the evaluation W [s]. 2. The eect of warping, being related to the shear strain [s] can be considered so small as to be neglected in most practical cases. Making w[y ] 0, W [s] will be zeroed by the choice = 0 and consequently b E . We also have [s] = [s], so the ICM kinematics reduces to the well known Antman beam kinematics [7]: = (1 + u,s ) cos + v,s sin 1
(1.25b)
= ,s
3. By neglecting the shear strain everywhere, i.e. assuming 0, we can use the condition = 0 in the second of eqs. (1.25b) for relating to u,s and v,s . With some algebra, eqs. (1.25b) become = (1 + u,s )2 + v,2 s 1 (1.25c) v,ss +v,ss u,s v,s u,ss = (1 + u,s )2 + v,2 s that coincide with the kinematic relationships developed by Nayfeh and Pai in [28]. We can further mention that, when assuming 1, the previous equations can be simplied into + 1 2 = u, + 1 (u,2 +v,2 ) s s s 2 2 (1.25d) (1 + ) = v,ss +v,ss u,s u,ss v,s 4. All previous variants only dier in the treatment of shear and axial strains in the expression of W [s]. When, as generally happens for slender beams, 1 and 1, their dierences become negligible. To give an example, 22
we can refer to a simply supported Euler beam of length L and subjected to a compressive axial force F [] := 2 EJ/L2 . The analytical solution for the postbuckling path coming from the use of (1.25b) can be found in [37] in the form 1 b2 + = b +
2
where is the endsection rotation angle and, introducing stiness ratios s := EJ/GAL2 , a := EJ/EAL2 , := s a , and the buckling load b and b are dened by postbuckling curvature ( ) b = b 1 + 4b b + 2 = 1 , b 4 1 + 2b The same expressions can also be obtained from eq.(1.25a), if := 2 s a , and from eq.(1.25c) and eq.(1.25d), assuming s 0 and s a , respectively (solutions can be recovered using algebraic manipulators, so their details are omitted here). We generally have s 1 and a 1, so all possible choices b 1/4) and can be considered as give essentially the same result (b 1, equivalent for practical purposes. We can also mention that usually a < s , so the simpler expressions (1.25d) could be considered better than (1.25c). 5. Finally, for a better understanding of the role of eq.(1.13) in our proposal, it is convenient, to refer to the limit case where the local and xed frame coincide, i.e. u ,s = u,s , v ,s = v,s . Making L 0 and 1 into (1.22c), we obtain = u,s + 1 v,2 s 2 (1.26) = v,
ss
which correspond to the standard 2nd-order kinematics used in simplied (technical) modelings. Obviously, this equation satises objectivity only to a 2nd order accuracy, and then requires some external tool in FEM management to obtain frame invariance at the element level (see [53]).
y x2 S[s] z s z
Q[s]
S[s]
h
x3
t
spatial system x := {x 1 , x 2 , x 3 }, aligned to the current section S := S [s] as shown in g.1.1, and denoting with xx [s, y, z ], xy [s, y, z ] and xz [s, y, z ] the relevant, normal and tangential, components of the stress eld, these are locally (i.e. on S ) given by: xx = N [s]/A xy = (w[y, z ],y z ) Mt [s]/Jt (1.27a) xz = (w[y, z ],z + y ) Mt [s]/Jt where w[y, z ] is an auxiliary warping which, in the case considered, can be evaluated as y z if t/2 y t/2 w[y, z ] (1.27b) y z if t/2 z t/2
A 2 b t is the area of the section and Jt its torsional inertia, dened by 2 Jt := {(w[y, z ],y z )2 + (w[y, z ],z + y )2 } dA t3 h
S 3
(1.27c)
(1.27d)
Furthermore, denoting with u 1 [s, y, z ], u 2 [s, y, z ], u 3 [s, y, z ] the components of the displacement eld in the corotational frame, these are locally given by: 1 [s + ds, y, z ] = u [s + ds] + L [s] w[y, z ] u u 2 [s + ds, y, z ] = z [s + ds] (1.28a) u 3 [s + ds, y, z ] = y [s + ds] where u and are the average axial displacement and the torsional rotation of the section u [s + ds] := L [s] ds , [s + ds] := L [s] ds (1.28b) 24
The previous expressions allow the recovery of the displacement gradient u[y, z ] = E [y, z ] + W [y, z ] eld on the current section S [s]. We obtain 1 1 L 2 (w,y z ) L 2 (w,z +y ) L 0 0 E= 1 2 (w,y z ) L 1 0 0 2 (w,z y ) L
1 (w,y +z ) L W = 2 1 2 (w,z y ) L
1 2
(w,y +z ) L 0 0
1 2
(w,z y ) L 0 0
(1.28c)
so the local kinematics E [y, z ] + W [y, z ] is completely dened by local derivatives of displacement u [s] and the rotation [s] associated to the section.
1 6
t b3
(1.29c)
the strain generalized parameters and are expressed by (4J1 Jt ) 2 [s] = u ,s + , s 8A 1 [s] = , s u ,s , s
2
(1.29d)
where use is made of eq.(1.28c). By relating eqs. (1.29a) and (1.29b) through Clapeiron's equivalence, we also obtain the elastic laws N = EA , M = GJ (1.29e) so completing the modeling, in the corotational reference frame.
25
[s] = u,s +
(1.30a)
which denes the nonlinear beam kinematics and, together with eqs. (1.29a) and (1.29b) completes the denition of the ICM modeling. Note that terms J1 and Jt appearing in the rst section of (1.29d) take into account the rotations due to the helicoid distortion of longitudinal bers and those due to shear strains, respectively. So we have Jt J1 in general (actually we have 1 1 3 3 Jt = 2 3 bt and J1 = 6 t b ). So, neglecting the small term 2 u,s eq.(1.30a), can be simplied into 1 J1 2 ,s , [s] := ,s (1.31) [s] := u ,s + 2 A which coincides with that derived by Wagner in his study of beams in torsion [14]. 2 It is also worth mentioning that the term 1 2 ,s J1 /A in the expression of implies a nonlinear coupling between axial elongation and torsional curvature which comes from warping and is generally neglected in approaches based on rigid motion kinematics of the section. This coupling, we call Wagner eect, can however play an important role in activating exural/torsional buckling of slender thinwalled beams, as we will show in the sequel, so generally it cannot be ignored.
26
+D=FJAH
the polar and exural inertia of the section. The linear solution is referred to a corotational spatial system x := {x 1 , x 2 , x 3 }, aligned to the current section S := S [s], as shown in g.2.1, Q[s] being the rotation matrix relating the two systems.
27
x1
y
x3
Q[s]
where := 11 is the normal stress and := [12 , 13 ]T collects in a single vector the tangential stresses. Introducing the force resultant N = {N1 , N2 , N3 } and the moment resultant M = {M1 , M2 , M3 } over the section, as usual dened by N1 = 11 dA , M1 = (y13 z12 ) dA S S N2 = 12 dA , N3 = 13 dA (2.1a) S S M2 = z11 dA , M3 = y11 dA
S S
and collecting them into the strength vectors N1 M1 t := M2 , t := N2 M3 N3 the stress solution can be expressed as:
(2.1b)
= D [y, z ]t , = D [y, z ]t
where
(2.1c)
(2.1d)
and vectorial stress functions d2 [y, z ] and d3 [y, z ] need to be evaluated by solving a Laplace/Neumann dierential problem whose numerical solution can be easily 28
obtained through a nite element discretization of the section as, for instance, performed in [84]. Interpolation (2.1c) allows the expression of the complementary energy [s] associated to the current section S [s] in terms of the strength vectors t and t . Following [84], we obtain ( ) 1 1 T 1 [s] := t H t + tT H t (2.2a) 2 E G where
H :=
S
DT D dA , H :=
S
DT D dA
(2.2b)
and E and G are the normal and tangential elastic modula. Matrix H , gives the contribution of normal stresses over the section and, being the corotational local frame aligned to the principal frame of the section, it is simply obtained as a diagonal matrix ] [ 1 1 1 H = diag . A J2 J3 Matrix H gives the contribution of tangential stresses and generally results in a full symmetric matrix which can however be obtained as a byproduct of the FEM solution process which provides the stress functions d2 [y, z ] and d3 [y, z ] (see [84] for details). It is convenient to introduce the vector quantities [ ] N t= (2.2c) M in order to arrange eq.(2.2a) in the more compact form
1 [s] := tT Ht 2
(2.2d)
is related to stress functions d2 [y, z ] and d3 [y, z ], as we will show in the sequel, and Lk , Lk are strain parameters to be related to the stress parameters Nk and 29
Mk through the section constitutive law. Therefore, the displacement gradient u assumes in S the following expression: [ ] L wT u = (2.5a) 0
where
L := L1 y L3 + z L2 [ ] [ ] z L1 w,y := L2 , w := L3 + y L1 w,z
(2.5b) (2.5c)
(2.5d)
By collecting the 6 strain constants L1 , L2 , L3 , L1 , L2 and L3 into the vectors L1 L1 := L2 , d := L2 d (2.5e) L3 L3 we can express all strain quantities in (2.5b) and (2.5c) using the following interpolation dened as a solution of the linear DV problem: L := D [y, z ] d := D [y, z ] d (2.5f) w := D w [y, z ] d
(2.5g)
while D w [y, z ] is related to the stress interpolation (2.1c) through the elastic laws 12 = G12 , 13 = G13 . Using eqs. (2.1c) and (2.2b), we obtain
1 L := + w = D [y, z ] d , D := D H
(2.5h) (2.5i)
and therefore
D w := D [y, z ] D
30
(2.6a)
(2.6b)
and recalling eq.(2.2d), the stationarity of the section strain energy [t] W [t, L ] with respect to stress vector t directly provides the constitutive law
t = H 1
(2.6c)
(2.7b)
and [2 2] matrix is inessential in our treatment, being multiplied by zero stresses when evaluating the strain work W . Using eqs. (2.1c) and (2.7b) and introducing the strain parameter vectors b1 b1 (2.7c) := b2 , := b2 b3 b3 we have
W :=
S
) T + T dA = tT + t
(2.7d)
31
where
T 1 1 J1 2 1 2 L1 + 2 ( A L1 + 2 L2 + L3 ) 8 d H d T 1 1 = L2 L1 L2 + 2J 2 J2r 8J d A3 d 2 L1 2 T 1 1 2 L3 L1 L3 J3r + d A2 d
2J3 1 2 L1 8J3
(2.7e)
+ = (1 L1 ) d
2
(L2 B 3 L3 B 2 ) d
(2.7f)
(2.7g)
and
J2r :=
S
z (y + z ) dA , J3r :=
2 2
y (y 2 + z 2 ) dA
S
(2.7h)
(2.8)
and combining eqs. (2.7c) and (2.2c) we nally obtain the constitutive law in the form L ] t = H 1 [d (2.9)
(2.10b)
(2.10c)
32
where
C 2 :=
A
yDT D dA , C 3 :=
A
z DT D dA
(2.10d)
These expressions can be further simplied if neglecting nonlinear terms related to the shear strain in case of symmetric section. We obtain J1 2 L1 + 1 L 1 L 1 2 A , L2 (2.11) L2 L3 L3 Note that the remaining nonlinear term Wagner axial/torsional coupling.
1 2
u ,s [s] = L , ,s [s] = L
(2.12a)
the rst u ,s being a derivative of a displacement and the second a derivative of a rotation vector. Recalling that [s] = 0, the latter can be related to a derivative of [s] through eqs. (1.5f) and (1.5g): a rotation matrix R 3 ,s 2 ,s ,s [s] := spin( 3 ,s 1 ,s R ,s ) = (2.12b) 2 ,s 1 ,s
[s], dened by reference to the local coroDisplacements u [s] and rotations R tational frame {x 1 , x 2 , x 3 }, are related to the corresponding ones u[s], and R[s], referring to a global xed frame through a change in the observer characterized by a [s]T . As R = I , we have Q = R and therefore, from relative rotation Q[s] = R[s]R eq. (1.19a) we obtain ,s [s] = RT R,s u ,s [s] = R[s]T (u,s +e1 ) e1 , R
(2.12c) Note that, if using a linear evaluation for the Biot strain, that is if referring to eq.(2.6), we recover the AntmanSimo nonlinear beam model kinematics (see [7, 8]). We also recover appropriate constitutive laws directly derived from the Saint Vennt theory without the need for adhoc assumptions. With the use of the complete quadratic evaluation (2.7) for Biot strains, we also recover the full subtle eects due to the section warping. When using the simplied quadratic evaluation (2.10) we obtain a quite simple expression which still however takes into account the nonlinear Wagner coupling due to torsional distortion. A FEM implementation of this model is given in section 4.1 whose results also show the strong inuence of Wagner coupling in cases of exural/torsional buckling. 33
x2 x1
x3
S[s]
s
Q[s] x2 x1
34
where J := h3 /12 is the exural inertia of the plate, [x3 ] is the stress function dened by (see also (1.20b)) h h 2 2 3 h2 x3 4 x3 3 := , dx3 = 0 , ,3 dx3 = h (2.13b) 2 2h h h
2 2
and, making i, j = 1, 2, h h 2 2 Nij [s] := h ij dx3 , Ti [s] := h i3 dx3 2 2 h 2 zij dx3 Mij [s] :=
h 2
(2.13c)
are the plate strengths we collect into the in-plane strength N := {N11 , N12 , N22 }, the shear strength T := {T1 , T2 } and the bending moment M := [M11 , M22 , M12 ].
(2.14b)
are the components of mean displacement and rotation vectors of the ber S [s], and w[x3 ] the warping function dened by 1 w := /k x3 , k := ,2 dA = 6/5 (2.14c) A S 3 Note that assumption (2.14b) provides 1 [s] = 2 [s] = 0, where the alignment of the x 1 and x 2 axis is implicitly set to be orthogonal to the ber. To complete the alignment we need to x their in plane drilling orientation by correspondence to 3 [s] = 0. We assume that it is set such that
u 01 ,2 = u 02 ,1
The expression for the displacement gradient consequently becomes u 01 ,1 + 2 ,1 x3 u 02 ,1 + 2 ,2 x3 u 03 ,1 w,3 02 ,1 1 ,1 x3 u 02 ,2 1 ,2 x3 u 03 ,2 w,3 u := u u 03 ,1 u 03 ,2 0 35
(2.14d)
(2.14e)
(2.15b)
Collecting stress parameters in a single vector t := {N , T , M }, (2.15a) can be written in compact form as 1 (2.15c) [s] = tT Ht 2 The generalized strain parameters can be obtained by comparing the previous equation with the expression of strain work
W [s] =
h/2
x3 =h/2
b b dx3
where Biot stress b is directly identied with the linear solution (2.13a) and the corresponding strain b with the displacement solution (2.14e) by the quadratic approximation (1.13). Performing integration, we obtain
W [s] = tT
=Ht
(2.16)
(2.17a)
u 01 ,1 u 02 ,2 L := u 01 ,2 + u02 ,1 [ ] u , L := 03 1 u 03 ,2 2 ,1 1 ,2 L := 2 ,2 1 ,1
36
(2.17b)
and
(2.17c)
(2.18a)
In order to complete the nonlinear modeling we only need to relate this local description, in the corotational frame, to the corresponding description in a xed global frame through changing the observer algebra (1.19a). The change in the observer [s] = I . We being characterized by a relative rotation Q[s], such that Q[s]T R[s]R obtain Q[s] = R[s] and so, from eqs.(1.19a) and making i = 1, 2,
(2.18b)
R = [i1 , i2 , i3 ]
Moreover, by introducing the plate directors ai dened by
(2.19a)
ai := ei + u0 ,i , i = 1, 2
equations (2.18b) can be rewritten i3 i2 ,i i1 ai ,i = i3 i1 ,i u 0 ,i = i2 ai ei , i1 i2 ,i i3 ai
(2.19b)
(2.19c)
37
and, consequently, obliging the quadratic contribution Q in the expression (2.17), i.e accepting a linear approximation for the strain parameters, we obtain [ ] i1 a1 1 i a 3 1 L i2 a2 1 , L i3 a2 i1 a2 + i2 a1 (2.19d) i3 i1 ,1 i3 i2 ,2 L i3 (i1 ,2 +i2 ,1 ) which coincides with the one derived by Simo in [10].
(2.20c)
(2.20d)
This formulation does not make explicit reference to rotation matrices and could be convenient in FEM discretizations which do not make use of nodal rotations, such as those based on spline interpolations (e.g. see [57]; see also [91] for a general discussion about spline based FEM elements). A FEM implementation of this modeling is discussed in section 4.2.
39
(2.21b)
(2.21c)
Note however that the two expressions dier in the denition of the stress eld which is obtained from b , through the constitutive relation = C 1 b in eq.(2.21b), or directly recovered as b L from the linear solution in eq.(2.21c). These two evaluations do not coincide as already discussed in Section 1.2.3. The stresses are an important part of the linear solution, whose main goal is their accurate recovery. To assume that b is directly recovered from the linear solution can then be generally considered a more reliable evaluation than that obtained from the displacements in a more elaborate way which also includes dierentiations. Actually, compatible formulations tend to generate spurious 3rd and 4thorder terms in the expression (2.21a) which can produce some locking in the resulting nonlinear modeling and generally they need some adhoc treatment to avoid this problem.
obtained by extending the linear solution by perturbation procedures [47]. Even if the use of eq.(1.13) is generally sucient for practical applications, research in this direction could be interesting.
41
+D=FJAH !
u := [u]u p u = 0 , u T
(3.1)
where u := {d, }, [u] is the strain energy, p u the external work and we denote by a prime Frecht's dierentiation. Applying FEM interpolation u := Lu, L being the interpolation operator, eq.(3.1) can be rewritten in the vectorial form
=0 r [u, ] := s[u] p
(3.2)
the response vector s[u] and the unitary load vector p being dened by the energy equivalencies sT u = [u]u , p T u = p u , u (3.3) We obtain a relationship between the conguration vector u and the load multiplier dening a curve (maybe composed of several separate branches) in the {u, } space, usually called equilibrium path. The aim of the analysis will be that of obtaining an accurate evaluation of its natural branch, that is the branch coming from a known initial conguration u0 , usually assumed as the one corresponding to = 0. For reading convenience, we briey summarize both the cited approaches here and refer interested readers to the specic literature, for further details (e.g. see [49, 59] and references therein). 42
g [u, ] (k+1) = 0
(3.4a)
which denes a surface in the {u, } space, so the required solution will correspond to the intersection of this surface with the equilibrium curve (3.2) (variations in the method are obtained by dierent choices for function g [u, ]). Conditions for achieving a proper intersection between (3.2) and (3.4a) are extensively described in Riks' papers, but, for a small step size, they are obviously not so restrictive. In particular the simplest choice of using the linear constraint
nT u (u u1 ) + n ( 1 ) = 0
where
(3.4b) (3.4c)
that is the hyperplane orthogonal to the increment {u1 = u1 u(k) , = 1 (k) } according to a metric dened by
||{u, }||2 :=
1 2
uT M u + 2
2
as originally suggested by Riks, will be eective (see [60] for a discussion about this topic). Solution of the extended system (3.2)(3.4b) can be obtained by a Modied NewtonRaphson scheme:
uj +1 = uj + u , j +1 = j +
(3.5a)
as a suitable approximation for the Hessian of the where by introducing matrix K strain energy [u]: [ ] s[u] K K [uj ] := (3.5b) u uj
43
(3.5c)
which is easily solved in partitioned form allowing us to exploit the symmetry and . banded structure of matrix K The main feature of the arc-length scheme is that it provides a simple way to overcome limit points because the extended system (3.5c) remains not singular even if the Hessian K becomes singular, which was a real diculty before the Riks paper in 1979. Its behavior is, however, strongly aected by the choice for the iteration which is usually assumed as the Hessian evaluated in correspondence to matrix K ( k ) u (i.e. at the beginning of the step) or to u1 (in the rst extrapolation point). The convergence of the iterative process (3.5) has been widely discussed in [51], to which the interested reader is referred. Summarizing the results given there, the convergence speed is essentially related to the relative dierence, along directions orthogonal to the path tangent, between the current Hessian matrix K j := K [uj ] used in the iteration. That is, it depends on how small the step and its estimate K length is, but much more on the nonlinearity in the problem description. Convergence failures arise when K j tends to stien, at least in some allowable directions, during the iteration process. This can occur when using a compatible format, where stress corrections are derived from the displacement corrections using compatibility and elastic laws, as a consequence of a locking (we call extrapolation locking in [51]) due to the interaction of large axial/exural stiness ratios with even small element rotations. The use of a mixed format such as that described in eqs. (1)(4), where stresses are dened separately and determined numerically directly exploiting equilibrium equations, avoids this interaction and noticeably improves the convergence without any need to decrease the step length more than strictly required for an accurate description of the equilibrium path (see [51], for a detailed discussion). As regards the aim of the present paper, it is important to consider that path following analysis only needs the response vector s[u], that is the rst variation of the strain energy, to be evaluated accurately, being directly related to the equilibrium check (3.2). Conversely, the same accuracy is not actually needed for the second variation of the energy which provides the Hessian matrix K , through the energy equivalence uT K [u] u = [u]u2 (3.6) because this matrix is only used as a part of the iteration process and its accuracy only inuences the convergence of the process. A fairly rough approximation is generally sucient for that purpose. As a consequence, if using small steps and an appropriate conguration updating able to reduce, as much as possible, geometrical incoherencies and avoid objectivity errors which could accumulate in the loading process, the accuracy demand in the nonlinear modeling can be noticeably reduced.
44
uf [] := u0 + u
(3.7a)
where u is the initial path tangent, obtained as solution of the linear vectorial equation K0 u =p , K 0 := K [u0 ] (3.7b) 2. A cluster of buckling loads i , i = 1 m, and associated buckling modes v i are f obtained along u [], dened by the critical condition
K [i ] v i = 0 , K [] := K [u0 + u ]
(3.7c)
We will denote with V := {v = m i } the subspace spanned by the bucki=1 i v ling modes v i and W := {w : wv i , i = 1 m} its orthogonal complement according the orthogonality condition w v i v iw = 0 b u
(3.7d)
where u := Lu , v i := Lv i , w := Lw. We will also denote with b an appropriate reference value for the cluster, e.g. the smallest of i or their mean value, and with a sux "b" quantities evaluated in correspondence to ub := uf [b ]. 3. Making 0 := ( b ) and v 0 := u , for more compact notation, the asymptotic approximation for the required path is dened by the expansion
u[, k ] := ub +
m i=0
i v i +
1 2
m i,j =0
i j wij
(3.7e)
where wij W are quadratic corrections introduced to satisfy the projection of equilibrium equation (3.1) into W and obtained by the linear orthogonal equations wT (K b wij + pij ) = 0 , w W (3.7f) 45
where K b := K [uf [b ]] and vectors pij are dened as a function of modes v i and, i = 0 m by the energy equivalence
wT pij = j v j b w v
4. The following energy terms are computed for i, j = 0 m, k = 1 m:
Aijk = iv j v k b v Bijhk = iv j v hv k b v b (wij whk + wih wjk + wik wjh ) Cik = b w00 wik
1 2 3 b ( b ) 2 v k + 2 v k b u b (b 3)b u 2 6 1 1
(3.7g)
k [] =
where the implicit imperfection factors k are dened by the 4th order expansion of the unbalanced work on the fundamental path (i.e. k [] := (p [u ])v k ). 5. The equilibrium path is obtained by projecting equilibrium equation (3.1) into V . According to eqs, (3.7a)(3.7g), we have
1 2 m i,j =0
i j Aijk +
1
1 6
i j h Bijhk + k []
(3.7h)
b ( b )
2
i,j,h=0 m i=0
i Cik = 0 , k = 1 . . . m
The equation corresponds to an algebraic nonlinear system of m equations in the m + 1 variables 0 , 1 m , with known coecients. Note that, from a computational point of view, the analysis develops into: i) solve linear system (3.7a); this can easily be obtained by standard Cholesky factorization of matrix K 0 ; ii) perform the buckling search (3.7b); this corresponds to a nonlinear eigenvalue problem which can be, however, easily solved, as described in [56], by an iter1 ative scheme only requiring matrix K 0 be available in factorized form, as it already is from the previous step; iii) solve orthogonal equations (3.7f); they correspond to a linear system where the righthand vectors pij are obtained as a function of v i by an elementby element assembling process similar to that used for obtaining s[u]; as K b K 0 within the orthogonal space W , its solution can be conveniently obtained, as described in [56], through a Modied Newtonlike iteration scheme exploiting K 0 as iteration matrix (see [71] for further insights).
46
iv) compute factors Aijk k [] through eq.(3.7g); all of them are scalar quantities which can be easily obtained as integrals of known functions. v) perform a pathfollowing solution of eq.(3.7h); because of the small dimensions of the system (of the order of tenths), this can be obtained very quickly using standard or even specialized variants of the arclength scheme. The actual implementation of the asymptotic approach as a computational tool is therefore quite easy in practice and its total computational burden, which is mainly involved in the factorization of matrix K 0 , remains of the order of that required by a standard linearized stability analysis. It provides the initial postbuckling behavior of the structure, including modal interactions and jumpingafterbifurcation phenomena. Moreover, once the preprocessor phase of the analysis has been performed (steps 1 to 4), the presence of small loading imperfections or geometrical defects can be taken into account in the postprocessing phase (step 5), by adding some, easily computed, additional imperfection terms in the expression of k [], with a negligible computational extracost, so allowing an inexpensive imperfection sensitivity analysis (e.g see [55, 56]). From eq.(3.7h) we can also extract information about the worst imperfection shapes [72, 73] we can use to improve the imperfection sensitivity analysis or for driving more detailed investigations through specialized pathfollowing analysis (see [49, 74] and references therein). For a discussion and further insights into asymptotic analysis the reader is referred to [49] and references therein. Note that this approach can provide a very accurate recovery of the equilibrium path, as it derives from both numerical testings and theoretical investigations [75]. Conversely, it makes great use of information attained from a 4thorder expansion of the strain energy and then requires a 4thorder accuracy be guaranteed in the structural modeling in order to obtain an appropriate evaluation of each term of the expansion. Even small inaccuracies in this evaluation, deriving from geometrical incoherencies in the higherorder terms of the expansion of the [d] law or in its nite element representation, signicantly aect the accuracy or the solution and can make it unreliable. It is also very sensitive to the format used in the extrapolation, to avoid extrapolation locking (see [52] for a discussion about this topic), and the use of a mixed equation format is generally needed to obtain a robust implementation. This high sensitivity of the asymptotic analysis to the accuracy in the nonlinear modeling actually represents an advantage for the purposes of the present paper because it provides a very sensitive context and appropriate benchmarks suitable for an in-depth testing of the accuracy of ICM modelings.
case we obtain an objective (i.e. frame independent) expression for the mixed form of the potential energy { } 1 := t[s]T [d[s]] t[s]T H [s]t[s] ds 2 (3.8) T T p[s] d[s] ds f [s] d[s] ds
where is the ber support domain, t, d and p, f are generalized strengths, displacements and interior and boundary loads associated to the ber S [s], H is a T compliance operator such that 1 2 t Ht is the local contribution to the complementary energy, and s is a material abscissa varying in . The ICM actually provides an appropriate objective denition for t[s], d[s] and H [s], as a generalization of quantities used in the corresponding linear modeling, and an explicit expression for nonlinear displacementtostrain relationship [d[s]]. The latter is obtained by standard change in the observer algebra and so it proves to be frameindependent, by denition. By partitioning the structure in nite elements and assuming in each of them both the internal elds t[s] and d[s] dened as a function of a discrete set of tensions and displacement element parameters te and de
(3.9)
expression (3.8) can be reduced to an algebraic form suitable for use in a numerical solution procedure.
(3.10a)
B t [s] and B d [s] collecting some appropriate interpolation shape functions. We also have a linear relationship between d[s] and [s], so from the displacement interpolation we can derive a linear interpolation law for the strain [s] = B [s]de
(3.10b)
As a consequence, performing integration, virtual works equation (3.1) can be rewritten in FEM format ([ ] [ ] [ ]) H e D e te Ae =0 (3.11a) d p DT e e e
e
where Ae is the standard FEM assemblage operator, summation is extended to all elements in the partition, and the element matrices H e , D e and vector pe are dened 48
by
(3.11b)
Stress and displacement interpolations can be set independently (we only need to satisfy the well known Brezzi-Babuska condition [76, 77]), in this case we call this a mixed interpolation. But more frequently they are related to each other by assuming
(3.12)
This choice, called compatible interpolation, implicitly satises the elastic laws, so allowing eq.(3.11a) to be further simplied into Ae (K e de pe ) = 0 (3.13a)
e
where
K e :=
e
(3.13b)
The accuracy we obtain with this discretization process, in both its mixed and compatible versions, depends on an appropriate choice for the interpolation matrices B t [s] and B d [s] which have to reproduce as well as possible the expected solution, a disturbing locking being produced by inappropriate choices. However there is great experience in this eld, both theoretical and numerical, and a large number of interpolation functions are available in literature, based on known properties of the expected solutions. So we can carefully select them according to the typology of the structure and the specic problem at hand. Obviously, mixed interpolation, allowing more freedom in the interpolation and so a more precise tuning, can provide better performances, in general.
[s] := [s, de ]
(3.14)
because of the nonlinearity of the relationship between d[s] and [s]. This also implies eq.(3.11a) will become nonlinear, i.e. ([ ]) H e te + e [de ] r := Ae =0 (3.15) D e [de ]T te pe
e
49
e [de ] and D e [de ] being dened by [ d ] := B t [s]T [s, de ] ds e e [ ] e [s, de ] T D e [de ] := B t [s] ds de e
(3.16)
But this is an expected minor problem in the analysis, which we can manage quite easily by the numerical strategies described in section 3.1. More subtle diculties derive from the fact that slender elements, such as beams (or shells), generally have a dierent behavior in the axial (or inplane) and in transversal directions, the former being governed by the elongation of the element, the latter by its bending. As a consequence, the expected solution will be characterized by a dierent shape for the axial (inplane) and the transversal components of the displacements, and so this involves dierent requirements for the corresponding interpolation functions. We can easily satisfy these requirements in linear analysis. For instance, by referring to the simple case of a planar beam element subjected to nodal loads, the standard interpolation, linear for the axial displacement, cubic for the transversal ones and quadratic for the rotations, could be appropriate in linear analysis, but the same interpolation could hardly be applied in nonlinear analysis because of local changes in the orientation of the beam axis produced by the beam motion. In fact, due to nite element rotations, the two interpolations become coupled and, as a consequence of the high axial-to-transversal stiness ratio, which is generally of some orders of magnitude, can produce a strong locking in the discretization. Moreover, the derivative of the transversal displacements can be identied as a rotation angle only within the simplifying assumptions of linear theory. In nonlinear analysis we have to be careful in using an appropriate representation for the rotations in order to assure full geometrical coherence in both the internal description of the element and in the interelement continuity conditions. This can prove to be a dicult task when 3D rotations are involved (see [78] and reference therein for more details of this topic). As recalled in subsection 1.2.1, rotations are characterized by a second order orthogonal tensor R SO(3) and combine with a product rule, so this is not easy to manage. To refer to its axial vector := spin(R), nonlinearly related to R through the Rodrigues formula (see eq.[10] of [53]), could be a convenient choice, but this introduces a nonlinearity in the interpolation of translation components of the displacement when they are expressed in terms of rotational parameters and prevents its denition by a simple expression as eq.(3.10a). To avoid these inconveniences the interpolation law should assure frame invariance in the displacement representation and, moreover, interpolations for the translational and rotational components of the displacement should be separately dened.The use of a mixed interpolation could help in this. In fact, we do not have frameindependence problems in the interpolation of the strength t, when it is done independently of the displacements, the components of t being aligned to the material and so frameindependent by denition. This makes it possible to express t[s] as a linear function of the discrete vector te when using a mixed interpolation. 50
By appropriate choices for B t [s] we can also lter, in the virtual works expression t[s]T [s] ds, undesirable aspects of the strain interpolation B [s]. All these possible advantages are lost when using compatible interpolations where the stress is related to the displacement by eq.(3.12). Several techniques can be used in the displacement interpolation to avoid locking, for instance: i) to assume the same shape functions for the 3 components of the translation (and/or the rotation) vector, as in the plate element used in [54, 55]; ii) to assume a separate interpolation for rotations and translations, as in the beam element used in [50]; to assume a stress interpolation which renders matrix D e independent of translation interpolation, as in [51, 52] (refer to these papers for further details about interpolation locking); iii) to lter rigid rotations of the element through an appropriate setting of the element reference frame. All these devices will be implemented in the element formulations proposed in Sections 4.1 and 4.2.
e := g e [de ] d
(3.17a)
to translate the element displacement vector de , referring to a global frame dened e in the current element frame. once and for all in the analysis, to its representation d They also use the same displacement interpolation in locally aligned components:
[s] := B d [s] d e d
(3.17b)
However, in the (updated) Lagrangian description, the element frame is assumed as xed (while frequently updated), so eq. (3.17a) states a linear relationship between e , i.e. de and d g e [de ] := Qe [dR (3.17c) e ]de where dR e denes the reference conguration and matrix Qe [] expresses the dierence in alignment between the global and the local frame. Conversely, in the corotational description, the element frame is assumed to move with the element and then the e . vectorial function g e [de ] := Qe [de ] de states a nonlinear relation between de and d Note that, while in the former case all nonlinearities of the problem are contained e ] and D e [d e ], in the latter in the internal element description through functions e [d case, the internal description is decoupled from the external nonlinearities due to 51
the nite rigid rotation of the element, which is taken into account by g e [de ]. As a consequence, in the corotational approach the need for an accurate modeling of the e , element could be somewhat relaxed. In fact, internal displacements, ruled by d are only related to the element distortion and so, by mesh renement, they can be made small enough to allow the use of the same interpolation laws derived from the linear analysis, without introducing noticeable interpolation locking. The smallness ] through e also allows an evaluation of the strain/displacement relationships [d of d its Taylor expansion without noticeable loss in accuracy. A 2ndorder expansion or, even more, a 1storder one can be sucient with ne meshes. In this case, the element description will coincide, in the corotational framing, with that already used in linear or the so called 2ndorder analysis, so allowing a direct reuse of standard software libraries widely available for these contexts. This is a great advantage, as pointed out by Rankin in the pioneering paper of 1988 [79], which is the reason for the diusion of the corotational approach in literature (see [87][90] to only cite recent proposals), but it is paid for the need for some complex algebra to obtain derivatives of the nonlinear function g e [de ] which are involved in the variations of the strain energy which are used by the analysis. These can be obtained in standard recursive form and so are easily managed in FEM analysis, as shown in [53], but in any case their derivation adds an extracost to the analysis. This extracost and the need for an appropriate mesh rening is the weak point of corotational approach. The Lagrangian approach does not have these disadvantages but, conversely, requires a full accurate modeling of the element: that is it needs an appropriate choice for the structural model to be used in the discretization and a careful setting of interpolation laws. Geometrical coherence plays an important role in pathfollowing and, even more, in asymptotic analysis and we have to be very careful, especially in the presence of coupled buckling, because even a minor loss in coherence can result in the mislaying of secondary bifurcations and modal jumping phenomena within the path recovery (e.g. see [72]-[74]). The need for an appropriate modeling could be a serious drawback, in general, but the possibility of obtaining geometrically exact models through the ICM procedure described in [?] is a strong advantage of the approach. So a Lagrangian description will be used in the element proposed in Sections 4.1 and 4.2.
To clarify this point, we can refer to the Hessian dened by eq.(3.6). By assuming an interpolation based on eqs. (3.10a) and (3.14), it will be directly obtained as ([ ]) He De [de ] Km [u] := Ae (3.18a) DT e [de ] Ge [de , te ]
e
[ Dk ] (DT e te ) = tk Ge [de , te ] := de de
k
(3.18b)
Dk and tk being the k th columns of matrix De and tk the k th component of vector te and summation being extended to all components of te . That is, denoting with index G the assembled global matrices, we obtain a mixed format [ ] HG DG [dG ] Km [u] := (3.18c) DT G [dG ] GG [dG , tG ]
which, however, can be reduced to a compatible format by static condensation:
1 Kc [u] := DT G [dG ]HG DG [dG ] + GG [dG , tG ]
Obviously, this rewriting does not change the element interpolation or its behavior, but only the format of its description. Conversely, by introducing the compatibility assumption (3.12), the Hessian is obtained in compatible format Kc [u] := Ae (K0e [de ] + Ge [de ])
e
K0e [de ] :=
(3.18d)
e and H e , the latter can be rewritten However, by an appropriate setting of matrices D in the form T 1 K0e [de ] = D e He De
and so, by introducing the further equation
1 te = H e De [de ]de
it can be split into the same mixed format as eq.(3.18a). Also in this case, the rewriting does not change the element nature, but only the format of its description. The mixed or compatible formats, while completely equivalent in principle, behave very dierently when implemented either in path following or asymptotic solution strategies. This is an important, even if frequently misunderstood, point in 53
practical computations which has been widely discussed in [51, 52]. By referring readers to these papers and to the general discussion given in [49] for more details, we only recall here that both numerical strategies described in Section 2 need function K [u] to be appropriately smooth in its controlling variables u. In pathfollowing := K [u0 ] when u moves in analysis, its smoothness will imply having K [u] K the neighborhood of u0 of interest, so allowing a fast convergence of the Newton iterative process. Analogously, matrix K [u] being the Hessian of the strain energy, its smoothness implies, in asymptotic analysis, that the higherorder energy term neglected in the 4thorder expansion (3.7h) be really irrelevant, so allowing an accurate recovery of the equilibrium path. We know that the smoothness of a nonlinear function strictly depends on the choice of the set of its control variables, that is on the format of its description, and can change noticeably when referring to another, even corresponding, set. As a consequence, the mixed and compatible format, even if referring to the same problem, can be characterized by a dierent smoothness and so they behave dierently in practice, when used within a numerical solution process. Actually, the compatible format is particularly sensitive to what we call extrapolation locking in [51, 52] which can produce a loss of convergence and then premature arrest of the incremental process, when used in pathfollowing analysis, or unacceptable errors in the path recovery, when used in asymptotic analysis. These inconveniences are easily avoided by changing to a mixed format.
54
+D=FJAH "
minimal set of parameters, is singularity free and gives a onetoone correspondence in the range 0 < 2 (see [78]). Taking := [s], it is also useful to introduce the operator 1 cos sin 2 T [] := I + W [] + W [] (4.2a) 2 3 which satises the following equivalence
( ) R ()T R () ,s = spin T T [] ,s
(4.2b)
N and M being the vectors collecting axial/shear strengths and torsional/bending couples and [ ] [ ] Q [L ] L L := , Q := (4.3b) Q [L ] L
where L and L are given by
L := RT (v ,s +e1 ) e1 , L := T T ,s
(4.3c)
as a function of translations v [s] and rotations [s] associated to a material abscissa s varying from 0 to e , and Q and Q collect the linear and quadratic terms in the expression of dened by eqs.(39). The modeling is completed by the compliance operator H dened in section 2.1. Note that a large part of the components of Q could refer to numerically negligible eects and so it can be omitted in practical computations (see section 2.1.2). Also note that, when using an appropriate reference frame xe suitable to render 1 within the element and making W := spin[], we can substitute the nite expressions (4.1) and (4.2a) with their Taylor expansion 1 1 1 R[] = I + W + W 2 + W 3 + W 4 + 2 6 24 (4.4) 1 1 1 2 3 T [] = I + W + W + W +
2 6 24
relating the alignment of the section S [s] to the element frame, and with Re the rotation matrix relating the element frame to the global frame xG which refers to the overall beam assemblage. A mixed separate interpolation is assumed for stresses and displacements. Stress is dened by a linear interpolation for M [s] := {M1 [s], M2 [s], M3 [s]} while N [s] = N e := {N1 , N2 , N3 } is assumed as constant over the element. Displacement is dened by a quadratic interpolation for the rotation vector [s] = {1 [s], 2 [s], 3 [s]}, and by a linked interpolation for the translation vector v [s] = {v1 [s], v2 [s], v3 [s]}. As a consequence, the element behavior is controlled by 9 stress parameters {N e , M i , M j } and by 15 displacement parameters {v i , v j , i , j , m }, where indexes i and j relate to quantities evaluated in the end-sections of the beam (s = 0 and s = e ), and 1 e [s] ds (4.5) m := e 0
57
Bn :=
, Bm :=
I I
)I I
(4.8b)
The interpolation for translations will be assumed as linked to that of the rotations, according to the following rule
:= v ,s [ ] spin ([ ]) e1 = const.
(4.8c)
Substituting stress interpolation (4.7) into eq.(3.11b), we obtain an explicit expression for the compliance element matrix He : 12 H nn 6 e H nm 1 (4.9b) He := 32 6 e H T e H mm nm 12 e 2 e H mm 58
(4.10a)
eL = {Lr , Ls , Le } , eQ = {Qr , Qs , Qe }
(4.10b)
(4.10c)
(4.10d)
with W m := spin (m ), W s := spin (s ), W e := spin (e ), and wm := W m e1 , ws := W s e1 , we := W e e1 , and constant , dened by eq.(4.8c), is obtained by integrating over the element: = r wm (4.10e) In a similar fashion it is possible to obtain Q . Its complex expression is not reported here, it being more simple and eective as it is derived directly using algebraic manipulators.
Assuming that the reference conguration be straight (r = s = e = 0), we obtain [ ]T [ ][ ] e e t He De t 2 e u = (4.11b) T e De Ge d de where He is given directly by eq.(4.9b), and matrices De and Ge are dened by I 0 0 spin (e1 ) 0 (4.11c) De := 0 I 0 0 0 I 0
0 0 0 Gr 1 1 0 2 Gs 1 3 Gm 2 Gs Ge := 1 1 1 0 2 Gs 5 Gm 2 Ge Gr 1 1 Gm 2 Gs 2 Ge
where
(4.11d)
Gr := spin (mr ) , Gs := spin (ms ) , Ge := spin (me ) Gm := (I e1 eT 1 ) mr1 + spin ({0, mr3 , mr2 })
2 1
(4.11e)
for planar beams, could mean a noticeable improvement in the element accuracy and allow rougher meshes. An investigation in this direction, which is outwith the scope of the present paper, could be of interest.
61
w,11 + (w,11 u,1 u,11 w,1 v,11 w,2 +w,11 v,2 ) + (u,2 v,11 w,1 +u,1 w,11 v,2 v,2 u,11 w,1 u,1 v,11 w,2 u,2 w,11 v,1 +v,1 u,11 w,2 ) w,22 + (w,22 u,1 u,22 w,1 v,22 w,2 +w,22 v,2 ) + (u,2 v,22 w,1 +u,1 w,22 v,2 v,2 u,22 w,1 = u,1 v,22 w,2 u,2 w,22 v,1 +v,1 u,22 w,2 ) w,12 + (w,12 u,1 u,12 w,1 v,12 w,2 +w,12 v,2 ) + (u,2 v,12 w,1 +u,1 w,12 v,2 v,2 u,12 w,1 u,1 v,12 w,2 u,2 w,12 v,1 +v,1 u,12 w,2 )
(4.12b)
u[s], v [s] and w[s] being the components of the average displacement (translation) d[s] of the transversal ber, according to a xed spatial frame xe := {x1 , x2 , x3 } aligned to the element. Expressions (4.12) are frame invariant, so the actual choice for this frame is irrelevant for their accuracy, even if it is convenient to refer to the local average alignment of the plate and make x3 orthogonal to its average plane, for practical reasons. It is worth noting that the Kirchho assumption of zero shear deformation allows the rotation parameters to be taken away from the kinematical description, so the nonlinear model is expressed in terms of the displacement elds alone, avoiding the use of rotation algebra. Also note, that the membranal strain coincides with the inplane partition of the GreenLagrange strain tensor, while the curvature takes a complex expression with respect to the linear approximation assumed in the classical Karman-Marguerre theory. The dierence lies in the addition of some quadratic and cubic extraterms which take into account pointwise dierences in the orientations between the local normal to the middle surface and the x3 axis. The modeling is completed by the compliance operator [ ] Cn H := (4.12c) Cm Cn and Cm being dened by eqs (51b). It is worth mentioning that both the two approximate models, previously implemented in KASP, were based on the same modeling while a linear approximation was used for the curvatures. Membranal strains were dened by the complete quadratic expression (4.12a) in that we called Complete Lagrangian (LC ) model and by a simplied expression, which only maintains quadratic terms in w, in the Simplied Lagrangian (LS ) model. The LS version is characterized by a better exural behavior, so it was the standard option in KASP.
this interpolation, which the reader can nd in [57, 60] and only recall here that its main advantage is that of assuring interelement continuity for both displacement components and their derivatives, with a minimal use of displacement parameters (approximately 3 kinematical DOFs per element) and without needing rotational degrees of freedom. As already done in KASP, a [2 2] Gauss scheme has been used to integrate all the energy terms required by the analysis. Letting sg be the position of the Gauss point g within the element e and denoting, from now on, with the index g quantities evaluated in point g , the strain energy can be evaluated by Gauss integrations. The element contribution to strain energy will be obtained by e := (Wg g ) (4.13a)
g
where
Wg := tT g g Ag , g :=
1 T tg Hg tg A g 2
(4.13b)
Hg being dened by eq.(4.12c), Ag := 1 4 1 2 the area pertaining to point g , the Gauss strain g [de ] := {g , g } obtained by eqs. (4.12a) and (4.12b) making s = sg , and the Gauss point stress tg := {N g , M g } dened by Ng11 Mg11 N g = Ng22 , M g = Mg22 (4.13c) Ng12 Mg12
where Ng11 , Ng22 , Mg12 are directly assumed as element parameters controlling the stress interpolation within a mixed format. This simple integration scheme is generally sucient for our purpose. Note that, even if more sophisticated interpolations could also be used, this choice provides an exact correspondence with the results obtained in previously published papers [55, 57, 60] by allowing attention to be focussed on the dierence due to the use of a more accurate structural plate model without introducing any disturbing discretization improvement. Energy variations needed by the analysis can be derived from eq.(4.13) by standard algebra and assemblage procedures. Explicit expressions, similar to those already given in [57], are simply obtained through the use of algebraic manipulators, so details are not included here.
63
+D=FJAH #
Numerical validation
5.1 Numerical validation
The numerical results are mainly related to the asymptotic analysis, this context being more sensitive to the correctness of both the structural model and its nite element implementation. The results of pathfollowing analysis are based on the same modeling and discretization, while using the corotational updating strategies described in [57] and [53], so they can be considered very accurate and are used, essentially as a reference to check the accuracy and robustness of the asymptotic results. In all the numerical tests comparisons are performed with results obtained by other authors and/or with analytical solutions, when available. The proposed 3D beam nite element, implemented into FEM code named COBE [83] has been tested for compact and general thinwalled cross sections. In all cases the evaluation of the compliance operator H e has been obtained through the solution of the Saint Venant equations using the FEM approach described in [84] to which the reader is referred for further details (see also [85]). The proposed Kirchho plate element has been implemented in the KASP code [57]. In this way, it was easy to compare the results provided by the proposed plate model (denoted as ICM) with that obtained, for the same nite element discretization, using the approximate plate models complete Lagrangian (LC) and simplied Lagrangian (LS) already implemented in the code. Thinwalled beams can be modeled as Saint Vennt rods but also as assemblages of thin plates. This allowed further comparisons between results provided by the two dierent modelings and so made it possible to obtain an independent check on their robustness. The comparison also allows us to emphasize the eects of Saint Vennt simplifying hypotheses on boundary conditions and cross-section distortions, in some relevant cases. The section is organized as follows: i) rst we present some simple tests, with a known analytical solution, showing the improvement in accuracy provided by ICM modeling with respect to the (LC) and (LS) ones; ii) we then present some tests to show the role played by postbuckling stress redistribution in reducing the accuracy demand in the plate modeling; iii) furthermore we present some tests to compare 64
the results obtained for thinwalled structures modeled as beams or as 3D plate assemblages, to show the general accuracy of the beam model recovered by the ICM, but also its possible shortcomings in the presence of relevant boundary disturbances, local buckling and inplane section distortions; iv) nally we report some tests referring to multimodal coupled buckling phenomena to show the real eectiveness of the proposed approach in general contexts and complex structural problems.
Despite their simplicity, when analyzed with an asymptotic approach, both problems are taxing with regard to the accuracy of the structural model and its FEM discretization. In fact, in both cases the buckling is not followed by a signicant stress redistribution, and in order to obtain the correct postbuckling results the modeling has to assure full objectivity and be free, in its FEM implementation, from both interpolation and extrapolation locking phenomena (see [50] and [52] for a dis65
cussion of these topics). The two tests have been analyzed using either the beam or the plate modeling and, in the case of plate modeling, referring to either an in plane or an out-of-plane buckling. The results are compared with known analytical solutions and with the ones obtained using the approximate LC and LS plate models also implemented in KASP. An independent analysis has also been made using the commercial code ABAQUS [94] to investigate the convergence behavior of the proposed elements. In all cases the ICM-beam pathfollowing results are assumed as reference values.
Euler beam
The test refers to the Euler beam shown in g.5.1. The resulting paths are drawn in g.5.3 and show a good agreement with that provided by pathfollowing analysis. b , obtained by The values for the buckling load b and the postcritical curvature identifying the expansion parameter as the endsection rotation of the beam, are reported in tabs. 5.2 and 5.1, for dierent discretization meshes. Note that both the ICM models recover the analytical solution [50] b = 2 and b = 0.25 for suciently ne grids exactly. The LC and LS models provide a / correct answer for the buckling load, but have a dierent postbuckling behavior in the inplane or outofplane analysis: LC agrees perfectly with the exact solution b = 2, eight times in the inplane case, whereas LS provides the wrong result / greater; conversely, LS behaves better in the outofplane case, by providing the ap b /b = 0, while LC gives a completely erroneous unstable postbuckling proximation curvature /b = 0.75. So the LS model could be viewed as a better compromise than LC, considering the importance of exural buckling in the slender plates (see also [54]). Note also, from the results reported in tab.5.1, the good convergence behavior, in recovering the buckling load, of both the proposed beam and plate elements in comparison with the ones used in ABAQUS (3node beam element, S4-R and S8-R plate elements). The improvement in convergence is apparent considering the larger number of variables implied by the ABAQUS meshes: generally more than double in the case of the beam elements while about 2 or 6 times larger in the case of S4-R or S8-R plate elements.
out plane buckling ICM S4-R S8-R 9.901 9.933 9.868 9.877 9.885 9.868 9.872 9.873 9.868 0.290 0.250 0.250
in plane buckling ICM S4-R S8-R 9.918 33.380 9.868 9.870 14.964 9.867 9.870 10.326 9.867 0.332 0.252 0.250
66
11
11
10
10
lL2 / EJ
9 Path-following Asymptotic ICM beam Asymptotic ICM plate Asymptotic LC Asymptotic LS 0 0.05 0.1 0.15 0.2
9 Path-following Asymptotic ICM beam Asymptotic ICM plate Asymptotic LC Asymptotic LS 0 0.05 0.1 0.15 0.2
wa / L
wa / L
Roorda frame
The test refers to the simple frame shown in g.5.2, rst investigated by Roorda. An analytical solution for this problem can be found in [86], by obtaining b = 13.89, b /b = 0.3805 and b /b = 0.7576 by reference to the choice := b for the ex pansion parameter. These values are recovered exactly by both the ICM models for suciently ne meshes (plate models are analyzed in outofplane conditions). The b are also recovered by the approxbuckling load b and the postbuckling slope imate models LC and LS, which, however, provide an unsatisfactory result for the b ; we have b /b = 0.2844 for LS and an even worse estipostbuckling curvature mate b /b = 1.4352 for LC. Note, from g.5.4, that paths recovered by asymptotic analysis show a good agreement with the reference path-following solution, when using both ICM models (they practically coincide up to a transversal displacement va of 10% of the beam length), while the ones obtained by LC and LS models rapidly deviate with the increasing displacement. A convergence test is also reported in tab.5.3, including a comparison with ABAQUS 67
13
12
lL2/ EJ
11
Path-following Asymptotic beam Asymptotic ICM Asymptotic LC Asymptotic LS
10
wa / L
results. Note the good convergence behavior of the ICM model, which conrms the one obtained for the Euler beam. lcr
b b
b b
el.n 16 32 64 16 32 64 16 32 64
ICM 13.954 13.903 13.890 0.3815 0.3807 0.3805 0.9070 0.7594 0.7570
beam 13.886
0.3805
0.7574
chosen to represent cases with no or very small postbuckling stress redistribution, which can be considered as quite unusual for slender plates. If present, the eect of stress redistributions dominates the postbuckling behavior, so reducing the relative size of the error. Stress redistribution can be considered as a typical phenomenon in the post buckling behavior of slender plates, especially in the case of relevant local buckling eects. We can show several examples for this, but also counterexamples, as shown by the following tests.
E=2.1 10
=0.25 s=0.1
a
Figure 5.5: Shear plate: geometry and loads
Note, from g.5.7, the excellent agreement between the asymptotic and path following paths for moderate transversal displacements (of the order of the plate depth). As the displacement increases, the asymptotic solution tends to deviate by overestimating the load. This deviation is an expected consequence of the asymptotic approach, because of the omission of higher order terms in the expansion. The error, which is higher than in the previous test cases, can be related to the importance, in 69
lcr
convergence for b at the n.elem. ICM a/b=1 9x9 173.2 27x27 172.2 81x81 171.8
a/b=1.5 9x18 27x54 81x162 132.08 130.52 130.28
mesh rening S4-R S8-R 186.7 170.6 172.5 170.2 170.6 170.2
139.0 130.69 129.63 129.53 129.46 129.46
this case, of the stress redistribution process which is largely inuenced by corrective modes wij , taken into account in eq.(3.7h) only to within a 3rdorder accuracy (see also the discussion in [54]).
70
240
Asymptotic: ICM Path-following
180
l 120
60
15
12
l
6 Path-following Asymptotic ICM Asymptotic LC Asymptotic LS
0 -1 9 19 29 39
va
Figure 5.8: Cantilever beam subjected to shear force: geometry and equilibrium path.
to reproduce the assigned resultants. The boundary constraints are described either by assuming zero displacements and rotations in all points of the clamped section (Full Clamped), or by assuming zero displacements and rotations in the web and a stress distribution in the anges corresponding to the one obtained by the Saint Vennt solution, (S.V. Clamped). The latter assumption can be considered more adequate for reproducing Saint Vennt clamping condition in the shear test. Relatively ne element meshes are utilized for both the beam and plate analysis (16 beam elements and (5 + 10 + 5) 50 plate elements, respectively) to avoid interference from discretization errors. In all cases, pathfollowing results were practically coincident with the one provided by asymptotic analysis, at least in the range considered, so their plot was generally omitted. In the compression test, a exural buckling (see g.5.11) occurs at b = 164.17, coincident with the analytical solution in [86], for both beam and plate analysis. We also obtain an excellent agreement between the beam and the two plate solutions, which are practically coincident, in the recovery of the equilibrium path, as shown in g.5.10, by reference to the transversal components of edge displacement (in the plate case, the plot refers to the average displacement of the edge section). 72
180
3 1
120
l
1.0 7
60
0 0 10 20 30 40
Figure 5.10:
When considering the results of the shear test, some dierences can be seen in g. 5.13 and 5.12, between the results from the ICM plate and beam models, even if the elastic compliance matrix of the beam section is evaluated by the procedure described [84] and so the beam modeling correctly accounts for all Saint Vennt couplings between shear and torsion. Moreover, in the case of plate analysis, the results are very sensitive to the boundary constraints details. Note that we have a quite large dierence ,6%, between the beam and the plate results in both /u3 and / curves, when considering Full Clamped constraints. This dierence reduces noticeably when considering SD Clamped constraints, even if a residual error still remains in the / plot, which could be referred to the overestimate of the torsional exibility in the Saint Vennt beam solution caused by the omission of the increment in torsional stiness due to Vlasov's nonuniform torsion.
73
Figure 5.11:
120
90
90
3 1
60
60
2.1 105 0.3
1.0 7
30
30
3
0 -0.4
Figure 5.12:
Figure 5.13:
30
3
85.73
20
q
12.7 2.0 105 0.3 0.001
3
142.88
l
10
85.73
0 -600
-400
-200
Figure 5.14:
Figure 5.15:
In the compressive test, a exural buckling occurs at b = 24.371 for both beam and plate modeling. Plate analysis is practically uninuenced by the two dierent boundary conditions. An excellent agreement can also be seen in the equilibrium 75
12 10 8
85.73
10 8
6 4 2 0 0 0.05 0.1
10
6 4 2 0
0.15 0.2
q
142.88
103
85.73
50
100
150
200
Figure 5.16:
Conversely, in the shear test, we can see from the path comparisons shown in g.5.16, a large dierence between the plate and beam modeling. This is apparent in the Full Clamped case, which has an error of about 18%, but is also present in the SV Clamped case, with an error of 4%, perhaps related to a residual dierence in boundary constraints between the plate and beam modelings or/and to the omission of the Vlasov increment of torsional stiness in the Saint Vennt beam modeling.
Figure 5.17:
of Full Clamped constraints, can be assumed as reference. Beam modeling, uses 8 beam elements and the analysis is repeated with (Q-ICM ) or without (L-ICM ) the quadratic term Q , as an expression of the ICM strain (4.3). In all cases a exural/torsional buckling is recovered, as shown in g.5.19.
20 L=300 L-ICM Q-ICM Plate-ICM
16
12
4
8
0.5
2
4
0.5
0 0 10 20 30 40 50
0 0 10 20 30 40 50
Figure 5.18:
b /2 b 103
A notable dierence in path recovery is obtained from the three FEM models, as shown in g.5.18 and tab.5.5. Neither of the two beam solutions recovers the reference plate result exactly. The largest discrepancy occurs for the L-ICM solution as a consequence of the loss of the Wagner coupling. A residual error is however present also in the Q-ICM solution, which could be related to in-plane distortion of the section, neglected in Saint Vennt solution. The relative error rapidly reduces as the beam length increases and this conrms these interpretations.
Figure 5.19:
even minor deviations from the Saint Vennt solution can have a nonnegligible effect on both the buckling and postbuckling recovery. So, there is a cost for that simplication. The relative error could be explained by reference to three possible eects: i) the in-plane distortion of the section, which is allowed by the plate modeling and ignored by the beam one; ii) the Vlasov increment in torsional stiness due to non uniform torsion, which is obliged by Saint Vennt solution; iii) the unsatisfactory correspondence in boundary constraints between the two modelings which requires a dierent average alignment of the clamped section. The latter seems to play a major role in some of the examined test cases. It is worth mentioning that a more accurate recovery could be obtained by rened onedimensional modelings which can be derived through ICM procedure from more sophisticated linear beam theories, such as that of Vlasov [92] or even better the ones obtained by the so called GBT approach (see [93]) which allow a more realistic modeling of boundary conditions. An investigation of this would be of interest, but outwith the scope of the present paper. Note also that, in any case, when analyzing beam assemblages, great care has to be paid to a detailed modeling of both boundary constraints and interelement connections.
detects three distortional modes, occurring at b = 1387, b = 1979 and b = 2063, respectively (see g.5.22), and reproduces a complex coupledbuckling phenomenon characterized by a strongly instable behavior. Note, from gs. 5.23 and 5.24, that the interaction between the rst two modes changes the postbuckling making it unstable, and a secondary bifurcation, which redirects the path to a strongly unstable branch, is produced because of the activation of localized distortions due to the interaction with the other two modes.
=1289.8 =1387.1
=1979.5 =2062.8
The contribution of 4th mode is actually very small and to include further buckling modes in the analysis does not change the overall behavior. This is also conrmed 79
1200
1200
c
900
900
b a
l 600
Asymptotic ICM: 1-2 modes Asymptotic ICM: 3-4 modes Path-Following
l 600
Asymptotic ICM beam
300
300
0 -7 -6 -5 -4 -3 -2 -1 0
0 -60
-50
-40
-30
-20
-10
va=-vb
wa=wb
by an independent pathfollowing analysis, whose results are also shown in g.5.23. Note the very good agreement between pathfollowing and asymptotic analysis. The two paths are practically coincident and only tend to deviate, after the secondary bifurcation, in correspondence to a quite large distortion of the middle section (see g.5.24-b). Note also that, in plate analysis, the same unstable behavior determined by the twomode analysis is also recovered by considering the rst (purely exural) mode alone. This result might be surprising but it is due to the fact that a section distortion, similar to that coming from the second buckling mode is also contained in the corrective mode w11 (see g.5.25). So the onemode analysis can actually reproduce the exural/distortional evolution of the buckling process. Previous results clearly show that the beam modeling, at least if based on Saint Vennt theory, is completely inadequate in problems presenting coupled buckling with relevant inplane section distortions. So it has to be used with certain care in the analysis of thinwalled folded plates. But they also show, from the results of plate analysis, the accuracy and reliability of the ICM approach in these quite dicult contexts.
80
81
+D=FJAH $
Conclusion
In the current state of the art in nonlinear structural analysis there is a lack of suitable nonlinear model, for use in nonlinear FEM analysis, in comparison with the richness and sophistication of existing linear theories. The declared purpose of this paper was that of reducing this gap by developing a general tool able to exploit information from existing linear models for bered continua, such as beams or plate, to setup the corresponding nonlinear models. We proposed a general procedure, called Implicit Corotational Method (ICM), which consist of two logical steps. In the rst, we exploit the stress and displacement elds provided, as a base solution, by the linear theory to derive appropriate expressions for the nonlinear elds as viewed by a moving local corotational observer; in the second we transfer this description to a xed global frame directly exploiting the change in observer algebra, so completing the nonlinear modeling and assuring frame invariance, by denition. The former step only needs a linear solution be available, provided by some existing linear theory. The recovery of the nonlinear elds is straightforward through a standard procedure: the stress eld is directly taken from the linear eld and the strain eld is obtained from the linear displacements by the quadratic formula (1.13). The latter introduces in the modeling the geometrical nonlinearities due to the nite motion of the ber, simply by exploiting the objectivity requirements in a constructive way. Only the simple standard algebra dened by eqs.(1.19a) is involved in this step and its implementation is straightforward and does not require any adhoc adaptation to t the particular problem at hand. The ICM method was implemented in two special but still technically relevant, contexts: that of 3D beams, according to Saint Vennt general rod theory, and plates, according to MindlinReissner and Kirchho plate theories, a homogeneous isotropic material being assumed in both cases. We discussed these cases in detail and have shown that the resulting models can actually recover all the richness of the underlying linear solution. We also indicated the eect of dierent simplication choices and have shown that, by appropriate simplifying assumptions, we can recover existing nonlinear models (e.g. the beam and plate models of Antman [7] and Simo
82
[8]).The method could however easily be generalized to anisotropic materials and also applied in dierent contexts, such as for instance Vlasov thinwalled beam theory [92], GBT theories [93] or laminated plate theory [28]. In all cases the advantage is that it provides fully objective nonlinear models by a blackbox procedure which only needs the corresponding linear model to be already available. The use of ICM as a numerical tool, able to provide relevant results in technical practice, requires the implementation of the recovered models within a nite element approach. While the implementation procedure could be as similarly straightforward as that used in linear analysis, some diculties can arise due to the presence of nite 3D rotations and the need to assure frame invariance, also in the interpolation of the mechanical quantities internal to the element. Dierent aspects of FEM implementation have been discussed, in relation to path-following or asymptotic analysis approaches, and some suitable strategies have been suggested to avoid locking problems related to the loss of frame invariance in the element denition or to an inappropriate format in their representation. Two elements have been described in detail. The elements are derived from nonlinear models previously dened, through a separate (mixed) interpolation of the displacement and stress elds and they are described in a mixed format which uses both stress and displacement control variables. The rst element derives from the nonlinear 3D beam model based on Saint Vennt rod theory described in section 2.1. It uses quite simple interpolation laws, directly derived from the linear case and can be considered the simplest choice in the nonlinear beam context. The second is derived from the thin plate model based on Kirchho linear theory described in section 2.2. It uses a spline interpolation for the translations which allows a rotation free kinematical description. This interpolation, which has already been used in [57], implies C 1 continuity for the displacement eld and it is highly suitable for the discrete modeling of rectangular plates and their 3D assemblages. The elements have been inserted into two computer codes COBE [83] and KASP [57], aimed at the FEM analysis of 3D beam or plate assemblages, using either pathfollowing or asymptotic solution strategies. Extensive numerical testings have been performed with these codes and some of the results have been reported and discussed in section 6. In all cases the proposed elements show strong robustness and high accuracy when compared with existing reference analytical solutions. The good agreement between results obtained by an independent modeling of thinwalled structures, such as Saint Vennt beams or Kirchho plates assemblages, gives a further mutual validation of both modelings, and dierences, if present, are clearly related to some inadequacies in Saint Vennt linear theory. The inadequacy of Saint Vennt beam modeling in presence of relevant inplane section distortions, but also the overall eectiveness of the proposed approach and its ability to recover even complex buckling behavior is apparent from the results reported in section 5.1.4. It is worth mentioning that both proposed elements should be viewed, as rst simple implementation examples, only aimed at showing the eectiveness of the ICM as a computational tool to be used in FEM analysis for solving practical problems. More rened elements with an even better behavior can be derived for both beams 83
and plates, starting from the nonlinear models presented in the rst part of the report. Further elements can also be obtained by more rened nonlinear continua which can be derived through the ICM approach as a nonlinear generalization of more sophisticated linear theories, such as for instance the Vlasov or GBT beam theories or the laminated shell theory.
Acknowledgements
The present paper has been developed within the national joint research project "Performance-based modeling and analysis of nonlinear structures", supported by the Italian Ministry of University Scientic and Technology Research (MIUR). We would like to thank all the participants in the project for their comments and suggestions. Special thanks are addressed to Giuseppe Zagari, for his help in setting up of the computer codes and performing the numerical tests.
84
Bibliography
[1] L. Malvern, 'Introduction to the Mechanics of a Continuous Medium', Prentice Hall, New-York, (1969). [2] J. Bonet, R.D. Wood, 'Nonlinear Continuum Mechanics for Finite Element Analysis', Cambridge University Press, New York, (1997). [3] A. Ibrahimbegovi, R. L. Taylor, 'On the role of frame-invariance in structural mechanics models at nite rotations', Computer Methods in Applied Mechanics and Engineering, 191, pp. 5159-5176, (2002). [4] E. Cosserat, F. Cosserat, 'Thorie des corps dformables', Hermann et ls, Paris, (1909). [5] E. Reissner, 'On one-dimensional nite strain beam theory: the plane problem', J. Appl. Math. Phys, 23, pp. 795-804, (1972). [6] E. Reissner, 'On nite deformations of space curved beams', J. Appl. Math. Phys.', 32, pp. 734-744, (1981). [7] S.S. Antman, 'Nonlinear Problems of Elasticity', Springer- Verlag, New-York, (1995). [8] J.C. Simo ,'A three dimensional nite-strain rod model. Part II: Computational aspect', Computer Methods in Applied Mechanics and Engineering, 58, pp. 79-116, (1986). [9] D.D. Fox, J.C. Simo,'A drill rotation formulation for geometrically exact shells', Computer Methods in Applied Mechanics and Engineering, 98, 3, pp. 329-343, (1992). [10] J.C. Simo, D.D. Fox, M.S. Rifai, 'On a stress resultant geometrically exact shell-model, part III'. Computer Methods in Applied Mechanics and Engineering, 79, 1, pp. 21-70, (1990). [11] P. Wriggers, F. Gruttmann,'Thin shells with nite rotations formulated in Biot stresses, theory and nite element formulation', Int. Journal for Numerical Methods in Engineering, 36, pp. 2049-2071, (1993). 85
[12] DJ Allman, 'A quadrilateral nite element including vertex rotations for plane elasticity', Int. Journal for Numerical Methods in Engineering, 26, pp. 717730, (1988). [13] Barr de Saint Vennt A., 'Mmoire sur la torsion des prismes, avec des considrations sur la exion', Mmoires des savant trangers, 14, p. 233 (1855). [14] H. Wagner, 'Torsion and buckling of open section', NACA TM 807. (1936). [15] E. Petrov, M. Geradin, 'Finite element theory for curved and twisted beams based on exact solutions for three-dimensional solids Part I: Beam concept and geometrically exact nonlinear formulation', Computer Methods in Applied Mechanics and Engineering, 165, pp. 43-92, (1998). [16] E. Petrov, M. Geradin, 'Finite element theory for curved and twisted beams based on exact solutions for three-dimensional solids Part II: Anisotropic and advanced beam models', Computer Methods in Applied Mechanics and Engineering, 165, pp. 93-127, (1998). [17] G.A. Wemper, 'Finite elements, nite rotations and small strains of exible shells', Int. Journal of Solids & Structures, 5, pp.117153. (1969). [18] T. Belytschko, L. Glaum, 'Application of high order corotational stretch theory to nonlinear nite elements analysis', Computer & Structures, 10, pp. 175-182, (1979). [19] C.C. Rankin, 'An element indipendent corotational procedure for the treatment of large rotations', J. of Pressure Vessel Technology, 108, pp. 165-174, (1986). [20] B. Nour-Omid, C.C. Rankin, 'Finite rotation analysis and consistent linearization using projectors', Computer Methods in Applied Mechanics and Engineering, 93, pp. 353-384, (1991). [21] J.C. Simo, 'A geometrically exact rod model incorporating shear and torsion warping', Int. Journal of Solids & Structures,27 (3), 371-393, (1991). [22] A. Ibrahimbegovi, F. Frey, I. Kozar,'Computational aspects of vector-like parametrization of three dimensional nite rotations', Int. Journal for Numerical Methods in Engineering, 38, pp. 36533673, (1995). [23] G. Garcea, A. Madeo, G. Zagari, R. Casciaro, 'Asymptotic postbuckling FEM analysis using a corotational formulation', Int. Journal of Solids & Structures 46 (2), 523532, (2009). [24] T. Hughes, J.A Cottrell, Y. Bazilevs, ' Isogeometric analysis: CAD, nite elements, NURBS, exact geometry and mesh renement ', Computer Methods in Applied Mechanics and Engineering, 39-41, 4135-4195, (2005).
86
[25] R. Goncalves, P.B. Dinis, D. Camotim, 'GBT formulation to analyse the rstorder and buckling behaviour of thin-walled members with arbitrary crosssections ', Thin-Walled Structures, 47 (5), 583-600, (2009). [26] Pai P.F., Nayfeh A.H., 'A new method for the modeling of geometric nonlinearities in structures', Computer & Structures, 53 (4), 877895, 1994. [27] Pai P.F., Palazotto A.N., Greer J.M.,'Polar decomposition and appropriate strains and stresses for nonlinear structural analyses', Computer & Structures, 66 (6), 823-840, 1998 Times Cited: 9 [28] Ali H. Nayfeh, P.F. Pai, 'Linear and Nonlinear Structural Mechanics', Jon Wiley, (2004). [29] G. Garcea, G.A.Truno, R. Casciaro, 'Mixed formulation and locking in path following nonlinear analysis', Computer Methods in Applied Mechanics and Engineering, 165 1-4, pp. 247-272, (1998). [30] G. Garcea, G. Salerno, R. Casciaro, 'Extrapolation locking and its sanitization in Koiter asymptotic analysis', Computer Methods in Applied Mechanics and Engineering, 180 1-2, pp. 137-167. (1999). [31] E. Riks, 'An incremental approach to the solution of snapping and buckling problems', Int. Journal of Solids & Structures, 15, pp. 529-551 (1979). [32] W.T. Koiter, 'On the stability of elastic equilibrium'. Thesis, Delft, 1945. English transl. NASA TT-F10, 883 (1967) and AFFDL\TR70-25 (1970). [33] B. Budiansky, `Theory of buckling and postbuckling of elastic structures', Advances in Applied Mechanics , 14, Academic Press, New York, 1974. [34] R. Casciaro, 'Computational Asymptotic PostBuckling Analysis of Slender Elastic Structures', CISM Courses and Lectures NO. 470, SpringerWien, NewYork, (2005). [35] A.D. Lanzo, G. Garcea, R. Casciaro, 'Koiter postbuckling analysis of elastic plates', Int. Journal for Numerical Methods in Engineering, 38, pp. 2325-2345, (1995). [36] A.D. Lanzo, G. Garcea, 'Koiter analysis of thin-walled structures by a nite element approach', Int. Journal for Numerical Methods in Engineering, 39, pp. 3007-3031, (1996). [37] G. Salerno, A.D. Lanzo, 'A nonlinear beam nite element for the postbuckling analysis of plane frame by Koiter's perturbation approach', Computer Methods in Applied Mechanics and Engineering, 146, 325349, (2001). [38] G. Garcea, 'Mixed formulation in Koiter analysis of thinwalled beam', Computer Methods in Applied Mechanics and Engineering, 190, 33693399, (2001). 87
[39] R. Casciaro, 'Nonlinear analysis: main problems and solution methodologies', in R. Calzona ed. "La ricerca non ha ne. Il Ponte sullo Stretto di Messina", DEI pbl., 2008 [40] M.Y. Kim, S. Kim, N. Kim, 'Spatial stability of shear deformable curved beams with non-symmetric thin-walled sections. I: Stability formulation and closedform solutions', Computer & Structures, 83, pp. 2525-2541, (2005). [41] Y.L. Pi, M.A. Bradford, B. Uy, 'Nonlinear analysis of members curved in space with warping and Wagner eects', Int. Journal of Solids & Structures, 42, pp. 3147-3169, (2005). [42] H.H. Chen, W.Y. Lin, K.M. Hsiao,'Corotational nite element formulation for thinwalled beams with generic open section', Computer Methods in Applied Mechanics and Engineering, 195, pp. 23342370, (2006). [43] Argyris J. H., Scharpf D. W., 'Some general considerations on the natural mode technique', Journal of Royal Aeronautical Society, 73, pp. 218226. (1969) [44] A.S. Petrolo, R. Casciaro, '3D beam element based on Saint Vennt's rod theory', Computer & Structures, 82, pp. 2471-2481, (2004). [45] G. Fichera, 'Remarks on Saint Vennt principle', I.N. Vekua 70th anniversary volume, Moscow, (1977). [46] Timoshenko, 'Strengt of Materials, Part I and II', D. Van Nostrand Company, 3rd Ed. (1955/1956). [47] Ali H. Nayfeh, 'Introduction to Perturbation Techniques', Wiley-Interscience, NY, (1981). [48] E. Riks, 'An incremental approach to the solution of snapping and buckling problems', Int. Journal of Solids & Structures, 15, 529-551 (1979). [49] R. Casciaro, 'Computational Asymptotic PostBuckling Analysis of Slender Elastic Structures', CISM Courses and Lectures NO. 470, SpringerWien, NewYork, (2005). [50] Salerno G.,Lanzo A.D., 'A nonlinear beam nite element for the postbuckling analysis of plane frame by Koiter's perturbation approach', Computer Methods in Applied Mechanics and Engineering, 146, 325349, (2001). [51] G. Garcea, G.A.Truno, R. Casciaro, 'Mixed formulation and locking in path following nonlinear analysis', Computer Methods in Applied Mechanics and Engineering, 165 (1-4), 247-272, (1998). [52] G. Garcea, G. Salerno, R. Casciaro, 'Extrapolation locking and its sanitization in Koiter asymptotic analysis', Computer Methods in Applied Mechanics and Engineering, 180 (1-2), 137167. (1999). 88
[53] G. Garcea, A. Madeo, G. Zagari, R. Casciaro, 'Asymptotic postbuckling FEM analysis using a corotational formulation', Internationl Journal of Solids and Structures, Int. Journal of Solids & Structures 46 (2), 523532, (2009). [54] A.D. Lanzo, G. Garcea, R. Casciaro, 'Koiter postbuckling analysis of elastic plates', Int. Journal for Numerical Methods in Engineering, 38, 23252345, (1995). [55] A.D. Lanzo, G. Garcea, 'Koiter analysis of thin-walled structures by a nite element approach', Int. Journal for Numerical Methods in Engineering, 39, 30073031, (1996). [56] R. Casciaro, G. Garcea, G. Attanasio, F. Giordano,'Perturbation approach to elastic post-buckling analysis', Computer & Structures 66, 585595, (1998). [57] G. Garcea, 'Mixed formulation in Koiter analysis of thinwalled beam', Computer Methods in Applied Mechanics and Engineering, 190, 33693399, (2001). [58] Riks E., Progress in Collapse Analysis. J. Pressure Vessels Tech., 109: 3341, 1987 [59] Riks E., Buckling, in Encyclopedia of Comutational mechanics, vol. 2, 139 164, edited by E.Stein, R.de Borst and T.R.J.Hughes, John Wiley & Sons Ltd., 2005. [60] G. Garcea, G.A. Truno, R. Casciaro, 'Pathfollowing analysis of thinwalled structures and comparison with asymptotic postcritical solutions', Int. Journal for Numerical Methods in Engineering, 55, 73100, (2002). [61] F. G. Flores, L. A. Godoy, 'Elastic postbuckling analysis via nite element and perturbation techniques. Part 1: Formulation', Int. Journal for Numerical Methods in Engineering 33, pp. 17751794, (1992). [62] R. Casciaro, G. Salerno and A. D. Lanzo, `Finite element asymptotic analysis of slender elastic structures: a simple approach', Int. J. Numer. Methods. Eng., 35, 13971426 (1992). [63] C. Pacoste, A. Eriksson, 'Element behavior in post-critical plane frame analysis', Computer Methods in Applied Mechanics and Engineering 125 (1-4), pp. 319343, (1995). [64] B. Wu, Z. Wang, 'A perturbation method for the determination of the buckling strength of imperfection-sensitive structures', Computer Methods in Applied Mechanics and Engineering 145 (3-4), pp. 203215, (1997). [65] P.N. Poulsen, L. Damkilde, 'Direct determination of asymptotic structural postbuckling behaviour by the nite element method', Int. Journal for Numerical Methods in Engineering, 42 (4), pp. 685702, (1998). 89
[66] E.H. Boutyour, H. Zahrouni, M. Potier-Ferry, M. Boudi, 'Asymptoticnumerical method for buckling analysis of shell structures with large rotations', Journal of Computational and Applied Mathematics 168 (1-2), pp. 7785, (2004). [67] N. Silvestre, D. Camotim,' Asymptotic-Numerical Method to Analyze the Postbuckling Behavior, imperfection-Sensitivity, and Mode Interaction in Frames', Journal of Enginering Mechanics 131 (6), pp. 617632, (2005). [68] B.W. Schafer, L. Graham-Brady, 'Sthocastic post-buckling of frames using Koiter methods', Int. Journal of Structural Stability and Dynamics 6 (3), pp. 333358, (2006). [69] H. Chen, L. N. Virgin, 'Finite element analysis of post-buckling dynamics in plates'. Part I: An asymptotic approach', Int. Journal of Solids & Structures 43, pp. 39834007, (2006). [70] T.Rahman, E.L.Jansen, 'Finite element based coupled mode initial postbuckling analysis of a composite cylindrical shell', Thin-Walled Structures, 48, pp. 2532, (2010). [71] Casciaro R., 'A fast iterative solver for nonlinear eigenvalue problem'. LABMEC, University of Calabria (2004), report 35 (can be downloaded from /www.labmec.unical.it/pubblicazioni/collana.php). [72] G. Salerno, R. Casciaro, 'Mode jumping and attrative paths in multimode elastic buckling', Int. Journal for Numerical Methods in Engineering, 40, 833861, (1997). [73] G. Salerno, G. Uva, 'Ho's theorem in global-local mode interaction of pinjointed bar structures', International Journal of Non-Linear Mechanics, 41 (3), 359376, (2006). [74] Casciaro R., Mancusi G. 2006. Imperfection sensitivity due to coupled local instability: a non-convex QP solution algorithm. Int. Journal for Numerical Methods in Engineering, 67: 815840 [75] F. Brezzi, M. Cornalba and A. Di Carlo, `How to get around a simple quadratic fold', Numer. Math ., 48, 417427, 1986. [76] Babuska I., 'Error Bounds for Finite Element Method', Numerische Mathematik, 16, 322333, 1971. [77] Brezzi F., 'On the Existence, Uniqueness and Approximation of Saddle-Point Problems Arising form Lagrange Multipliers', R.A.I.R.O., 8, R2, 129151, 1974.
90
[78] M. Ritto Corra, D. Camotin, 'On the dierentiation of the Rodrigues formula and its signicance for vector like parametrization of Reissner-Simo beam theory', Int. Journal for Numerical Methods in Engineering, 55, 1005-1032, (2002). [79] C.C. Rankin, B. Nour-Omid, 'The use of projectors to improve nite element performance', Computer & Structures, 30, 257-267, (1988). [80] O. Rodrigues , 'Des lois gometriques qui rgissent les dplacements d'un systme solide dans l'espace', Journal de Mathmatiques Pures et Appliques, 5, 380440, (1840). [81] J.H. Argyris, H. Balmer, S.St Doltsinis, P.C. Dunne, M. Haase, M.Kleiber, G.A. Malejannakis, H.P. Mlejnek, M. Muller, D.W. Scharpf, 'Finite element method - The natural approach ', Computer Methods in Applied Mechanics and Engineering, 17-18, 1106, (1979). [82] M. Aristodemo, 'A high-continuity nite element model for twodimensional elastic problem', Computer & Structures, 21, 987993, (1985). [83] A. Madeo, 'The Implicit Corotational Method: General theory and FEM implementation', Phd Thesis, (2008), LABMEC, University of Calabria, can be downloaded from //www.labmec.unical.it/dottorato/tesi/Madeo.pdf. [84] A.S. Petrolo, R. Casciaro, '3D beam element based on Saint Vennt's rod theory', Computer & Structures, 82, 24712481, (2004). [85] W. Lacarbonara, A. Paolone, 'Solution strategies to Saint-Vennt problem', Journal of Computational and Applied Mathematics, 206, 473497, (2007). [86] Bazant Z.P., Cedolin L., 'Stability of Structurers', Dover Publications, New York, (1991). [87] Battini J. M., Pacoste C., 'Co-rotational beam elements with warping eects in instability problems, Computer Methods in Applied Mechanics and Engineering 191, 1755-1789, 2002 [88] R. Alsafadie, J.-M. Battini, H. Somja, M. Hjiaj, 'Local formulation for elastoplastic corotational thin-walled beams based on higher-order curvature terms', Finite Elements in Analysis and Design, 47 (2), 119-128, 2011. [89] R. Alsafadie, M. Hjiaj, J.-M. Battini, 'Corotational mixed nite element formulation for thin-walled beams with generic cross-section', Computer Methods in Applied Mechanics and Engineering, 199 (49-52), 3197-3212, 2010. [90] H.H. Chen, W.Y. Lin and K.M. Hsiao, 'Co-rotational nite element formulation for thin-walled beams with generic open section', Computer Methods in Applied Mechanics and Engineering, 195, 2334-2370, (2006). 91
[91] T. Hughes, J.A Cottrell, Y. Bazilevs, ' Isogeometric analysis: CAD, nite elements, NURBS, exact geometry and mesh renement ', Computer Methods in Applied Mechanics and Engineering, 39-41, 4135-4195, (2005). [92] V.Z. Vlasov, 'Thin walled elastic bars', Fizmatgiz, Mosca, (1959). [93] R. Goncalves, P.B. Dinis, D. Camotim, 'GBT formulation to analyse the rstorder and buckling behaviour of thin-walled members with arbitrary crosssections ', Thin-Walled Structures, 47 (5), 583-600, (2009). [94] Hibbitt, Karlsson, and Sorenson, 'ABAQUS Version 6.7. Pawtucket', (2007).
92