mg 3 ie ih i wag sf dy Wile dia a dane ou dag et poh aa & 1 FINITE ELEMENT TECHNIQUES FOR NONLINEAR POSTBUCKLING AND COLLAPSE OF ELASTIC STRUCTURES NORMAN F. KNIGHT, JR. Aerospace Engineering Department, Old Dominion University, Norfolk, VA 23529.0247 Ll. INTRODUCTION ‘The design of structures for automotive, naval, aircraft, and space applica ‘ions exploits the finite clement modeling and analysis techniques to ensure reliability and robustness in a given design, Move and more often, the use ‘of nonlinear structural analyses are required in order to simulate the complex physical response to extreme loading conditions and to utilize the structural ‘design and material to its fullest extent. Meally structural collapse should ‘not occur under normal operating conditions; however, an understanding of the collapse phenomena can lead to increased structural integrity and con- fidence in residual strength predictions. ‘The finite element method is an approximation method for solving dif- {erential equations of mathematical physics. Starting from the differential equations and boundary conditions, 2 weak form of the problem is devel- ‘oped in a variational sense, This variational statement is used to define clemental properties that may be written as matrices and vectors as well as 10 identity primary and secondary variables and all possible boundary con- ditions. Specific equilibrium problems are solved by frst discretizing the spatial domain of the problem, evaluating the elemental matrices and vee- tors, assembling elemental terms to form the global system of algebraic ‘equations, applying boundary conditions, and solving the resulting systemNF KNIGHT ‘algebraic equations. This system of equations is often nontinear. In solid sechanics, the types of nonlinearity include geometric nonlinearity (typi- ally from the strain-displacement relations), material nonlinearity (typi- ally fom the stress-strain relations), and boundary condition nonlinearity ypically from contact or friction). ‘Nonlinear finite element analyses are readily performed using any one of ‘veral commercially available finite element software systems such as ISCINASTRAN, HKS/ABAQUS, or ANSYS. Today's analyst can easily ‘odel the spatial geometry of large complex systems and generate finite ‘ement models which easily exceed one million active degrees of freedom. ‘oupling these two facts with the availability of high-performance comput- 4 systems provides analysts with simulation capabilities that far exceed ¢ capabilites available less than a decade ago. Several aspects of such mputations are described by Hibbitt (1986, 1993). Common to all of the nonlinear solution techniques is the need'to trace at the entire load-displacement response curve. As such, the computational 2st of including such « nonlinear analysis approach within a design opti- ization loop will be high. This feature provides the impetus for the devel- ment of robust and efficient nonlinear analysis methods for designing ructures. Ifthe structure is designed to exploit its postbuckling siffness, fen fective design methodologies need tobe developed ant inconporated to the structural optimization procedure. Within the design optimization “quence, the values of the design parameters may vary, and in composite rictural design, the number of possible design parameters increases dra- saically (¢g., Stroud, 1982). Each change in a design variable typically, ‘quires at least one additional nonlinear analysis if gradient-based optim \Gion methods are used. The use of traditional nonlinear analysis methods ithin such a design optimization sequence is unattractive, ar more pre- sely uneconomical, from a computational standpoint and also from the -signer’s standpoint in that only limited design configurations can be ‘sessed ia a given amount of computational time. ‘The abjectives of this paper are 1 presenta brief overview of the finite ement method, fo review selected finite element analysis techniques as iplied to determining the nonlinear postbuckling and collapse response of astic structures, and to describe selected application problems. First, some the fundamental concepts associated with finite element approximations ee described. Second, the basic equations of nonlinear solid mechanics are ven for the case of small strain, larg elastic deformations including a brief scustion of lamination theory for composite structures. Third, different wiational formulations are presented along with ther finite element mod- 5. Then solution techniques for nonlinear systems (both static and dy- mic) are diseussed. The traditional approach of nonlinear finite element NONLINEAR POSTBUCKLING 3 analysis using a Newton-Raphson approach is reviewed. Reduction methods for nonlineas problems such as the globsl function approach and the reduced basis approach are summarized. Next selected applications are described, and finally some concluding remarks are made. 1.2. FINITE ELEMENT FUNDAMENTALS Finite element fundamentals are reviewed in this section. First, an overview Of the method is given, followed by classification of finite element tech niques, and concluded by a discussion of finite element approximation functions. Only selected aspects ofthe finite element method are presented and described herein. The interest reader is encouraged to look further into the finite element method by studying one of the many available books on the topic (e., Bathe, 1996; Cook, 1995; Cook, Malkus, and Plesha, 1989; risfield, 1992; MacNeal, 1994; Reddy, 1993; Szabo and Babuska, 1991; Zienkiewiez and Taylor, 1989, 1991). 1.2.1. Overview of the Finite Element Method ‘The init elesient mst em Wochnique fee salving’ditierenial equations ‘which eliminates the major dificulty associated with Gnite difference tech- niques, perturbation methods, variational methods, and the method of ‘weighted residuals — the representation of complicated domains and their boundary conditions. An operational definition of the finite element method is one that gives a description of the fundamentals of the method; namely, “the principle of going from the part (0 the whole.” The finite element ‘method was initially conceived as an extension of variational methods 10 handle arbitrary domains. Barly finite element methods were associated with the direct stiffness method from structural analysis, Later it was discovered ‘that the finite element method was an extension of the method of weighted residuals which could handle any differential equation, Fusther discussion of the historical developments of the finite element method is given by Zienkiewicz (1970, 1983). ‘The finite element method is a general discretization technique in which fa given domain is decomposed into a finite number of subdomains or “finite clements” where families of fields (e-., displacements, sresses, tempera- tures), which may have different analytical expressions within each subdomain, are considered. An element may be isolated from the assem. blage of elements and its properties studied independently regardless ofits tultimate location within the domain (Le., within the assemblage of ele- meals),4 NE KNIGHT Classification of Finite Element Techniques Finite element techniques many be classified in many ways: dimensionality, shape, element parameters, fundamental unknowns, formulation, and disei- pline. Dimensionalty refers to two possible aspects of how the differential quation being treated and its associated finite elemeat, It usually refers 10 the spatial domain of the differential equation governing the response such ss one-dimensional problem for a beam or a two-dimensional problem for plane elasticity. It may also refer to the spatial domain in which the element 's able to react such as a one-dimensional beam clement deforming in a plane ot in three-dimensional space (ie., planar beam or spatial bea. Shape refers to the geometry of the element, One-dimensional elements snay be straight or curved and may be planar or spatial elements, Two- Aimensional elements may be triangles or quadrilaterals with either straight or curved edges and with either flat or curved surfaces. Three-dimensional slements may be tetrahedals, wedges, or hexahedrals with flat or curved aces, Element parameters refer to the types of unknowns of the element and ypically include either nodal values, unknown. coefficients, oF averaged ‘alues over the element, Nodal methods express the field variables in terms Df values of the field variable, and possibly their derivatives, at the nodes 2 the element, This fs the most common class of elements employed 1 ‘omimercial finite element codes. Coefficient methods are carry over from he classical Rayleigh-Ritz approximation approach and is sometimes re- ‘erred to as the a-basis approsich. In this method, the field variables are ‘xpressed es a polynomial or asa Fourier series with unknown coefficients The unknowns are coefficients and post-processing is necessary to obtain he solution. This method is most common in p-version Finite element codes vhich feature edge modes and bubble modes in a hierarchical manner (see izabo and Babuska, 1991). Cell methods have element parameters that are lected 10 be averaged (or representative) values of the fundamental us -nowns in a small volume element or over a sueface element, This method 's mainly used in fluid flow problems. In some eases, these methods are combined in order to enhance element performance. One such case is the addition of incompatible modes to the 4-node (uadrilateral membrane element (e.g, Taylor, Beresford and Wilson, 1976). ‘our incompatible modes are added to the shape functions of the 4-node ‘linear quadrilateral membrane clement, and their unknown coefficients are latically condensed out atthe element level. For the general quadrilateral hhape, a modified integration scheme is also employed. This modification, ignificantly improves the element's performance and, inthe limit a¢ the NONLINEAR POSTBUCKLING 5 ‘mesh is refined, the displacement fleld incompatibility dies out and a con- stant strain state is approached, Fundamental unknowns refer to the (ype of element parameters consid- fered for the clement. Single-field methods have only a single field (e.g., displacements or temperature) as unknown. Multt-field methods have mul- tiple field variables (e-.. displacements and stresses) as unknowns. In solid mechanics, these multi-field methods are referred to as mixed finite element ‘methods. In addition, hybrid finite element methods have different approxi- ‘mations within the element and along the element boundary. For example, the stress field may be approximated within the element but interelement continuity of the stresses is not required; only compatibility of the nodal displacements (e.g., Pian, 1964). Another example involves having the displacement approximated in one manner along the boundary of the ele- ‘ment and a different approximation within the eloment. This approach pives rise to “enriched elements” where specific analytic solutions are embedded within the element (ee. Tong, 1984; Krawczuk, 1994; Chen, 1995). Formulation refers to the procedure used to develop elemental properties ‘such as the clement stiffness matrices, element force vectors, and element mass matrices, In the beginning days of finite element analysis, element properties were derived using the direct stiffness method, often based on physical insight and invition, and limited to a simple set of elements. Since {twas recognized thatthe finite element method is essentially a piecewise continuous version of the Rayleigh-Ritz procedure, formulations based on Variational principles in mechanics began to appear. However, no all prob- lems have an associated natural variational principle like the principle of ‘minimum potential energy in solid mechanics. As the finite element method matured, it was recognized that the method of weighted-residuals, such as the Bubnov-Galerkin method, could be used with the differential equations themselves to establish a weak-form statement rather than a solid mechanies energy principle to form a variational statement. This approach placed the finite element method on a strong mathematical foundation for extensions into other disciplines Discipline refers to the type of engineering problem being solved: solid ‘and structural mechanics, fluid mechanics, beat wansfer, biomechanics, electromagnetics, and so forth. For the most par, finite elements have been rmoce widely used in solid and structural mecbanics and dynamics than in ‘other enginecring disciplines. However, wih the exception of computational fluid dynamics, the finite element method is perhaps the more popular approximate solution method for beat transfer problems, electromagnetic problems and coupled field problems. A good resource for background information and formulation aspects of different disciplines is given in the Finite Element Handbook (Kardestuncer and Norrie, 1987).6 NE KNIGHT 1.2.3. Finite Element Approximation Funetions In the finite element method, two basic approximations are made: one for the geometry ofthe problem (eg, shape or domain) and ane fr the response variables or fields of the problom (c.z., displacement, temperature). These approximations are made using interpolation or shape functions over each clement. For example, the field variable approximations in one-dimension can be waitten as o@= YM F a ‘where ¢ (2) isthe field being approximated, are the nodal values of the field, (8) are the shape funetons for the fie, is the natural coordinate of the one-dimensional element, and NNPE is the number of nodes per ‘element. Generally, natural coordinates range from ~1.<£< 1 which facli- tates the use of common numerical integration procedures in the evaluation ‘of elemental matrices and vectors as well as to facilitate the generation of shape functions. Shape functions are required to satisfy certain requirements within the element as well as at interelement boundaries. Fist, the approximation fonet ueeio desivatives up through those appearing the weak-form statement; that is they must be suffieicntly differentiable. ‘Second, the nodal interpolants need fo produce nodal values when evaluated ta node. Third, the approximation must be able to represent constant values of the primary variables. Finally, the minimum continuity requirement on the approximation functions is defined by the highest derivative ofeach field variable appearing inthe weak-form statement of the problem. Ifthe first derivative with respect to the spatial variables is the highest derivative ‘appearing in the weak-form statement, then only the field variable itself is required to be continuous across interelement boundaries which defines C° continuity (e.g., rods, Timoshenko beams, plane elasticity, three-dimen- sional elasticity, solids of revolution Mindlin plates/shels). Ifthe second derivative with respect to the spatial variables is the highest detvative appearing in the weak-form statement, then the field variable and its first derivative are required to be continuous across interelement boundaries \which defines C! continuity (¢,, Bernouli-Euler beams, Kirchhoff plats! shells. For C® problems the approximation funetions for I-D and for 2-D quad- silaterals are typically based onthe one-dimensional Lagrange interpolation functions determined using the product rule, This approach can be used £9 i inet cal NONLINEAR POSTBUCKLING 7 generate ap" oder, one-dimensional Lagrange interpolation function 2A (6) ‘ing cattle : KAS) fi(ES) 2 “These one-dimensional functions are really caried over to two- and three- dimensional elements and the generation of ther shape functions. For C? problems, the approximation functions for 1-D are based on the Hermite interpolation functions and are generally restricted to the frst order which are cubic spline functions. For the 2-D problems, these approximation functions are developed by examining the continuity requirements along the ‘edges ofthe clement (¢., Melosh, 1963) Ina manner similar to the approximation of the field variable, the geom- ‘try of the element may also vary over the element by using nodal values fof the coordinates and shape functions in tems of the natural coordinates ‘ofthe element. Tht is, one-dimensional curve can be described by x)= >) WUE) R, @ where x (f) isthe coordinate being approximated, ar the nodal values of * the coocdinate, (8) are the shape functions forthe coordinate, and Eis the natural coordinate of the one-dimensional element. I the shape Functions for the field variables N,(2) are the same as the shape function for the geometry of the element ¥N,(é), then an isoparametric formulation is ‘obiained (the field vavables and the geometry ate approximated inthe same mane). Ite fel variable approximations N(2) are of lower order than the geometry approximation §,(6), thon a superparametrie formulation is obtained (geometry approximation is higher order than the fick! variable approximation). Ifthe field variable approximations N() are of higher ‘order than the geometry approximation 1,(6), then a subparametrie fo muvlation is obtained (geometry approximation is lower order than the Field variable approximation) While these tational techniques of generating the element shape func- tions and element characteristic matrices and vectors are commonly taught in a frst-couse in finite element analysis, they are frequently not used in their academe form by the commercial software developers of finite cle- ment codes, Some developers incorporate intemal incompatible modes (see ‘Taylor er al, 1976), others include “drilling” freedoms (see Allman, 1984), sand most have proprietary “uning parameters” that are adjusted to improve« NE KNIGHT the performance and reliability of their elements. In additional, the devel- ‘opment of hierarchical shape functions and hierarchical elements (Le., p- version finite element method, see Szabo and Babuska, 1991) has matared sufficiently and is now appearing in commercial codes c.g, Rasna/Mechaniea and MSC/NASTRAN Version 69) rather than just appearing in research- oriented codes, 1.3, BASIC EQUATIONS OF NONLINEAR SOLID MECHANICS ‘The basic equations of nonlinear solid mechanics for small-sirain, elastic behavior are reviewed in this section. The basic equilibrium equations are described firs, followed by a discussion of the constitutive equations, then the strain-displacement relations, and finally the compatibility relations. The purpose of this section is to establish a common basis from which to proceed for the subsequent sections, 1.3.1. Equilibrium Equations ‘The basic equations for small-sirain, elastic nonlinear deformations are escribed. The first set of equations is the equilibrium equations. These partial differential equations are normally derived by considering a differ- cotial element of the material subjected to a three-dimensional stress state. ‘Summing forces imeach of the three coordinate directions gives Bon Pu “Oe oF A z @ Pu. OF Where oj is the second-order stress tensor, f; are the body forces, p is the mass density, and 41s the displacement field, Summation of moments about 4 point in each direction gives the symmetry of the stress tensor for most engineering materials (see Reissner (1944) for an exception). Most plate and shell formolations assume inextensibilty in the thickness direction, thereby reducing the three-dimensional problem to that of a two-dimensional prob Jem with assumed kinematics in the thickness coordinate direction NONLINEAR POSTBUCKLING ° 1.3.2. Constitutive Equations ‘The constitutive equations relate stresses tothe strains. In general, a poten tial function is postulated such that for an adiabatic process, a stain energy density Function 9 exists from which the stress-strain relations can be developed au = 5 i © ‘where 6; is the second-order strain tensor. For generalized Hookean mate rials, the constitutive relations are defined in terms of the stiffness coefTi- cients Cjy which are dependent only on position within the structure and the corresponding stress-strain relations are y= Co Fu © ‘The fourth-order tensor Cy has 81 terms. Due to the symmetry nature of the stress and strain tensors, only 21 independent terms are needed to ‘describe an anisotropic, homogeneous material. For an orthotropic matetial, ‘only 9 independent terms are required. If there is one plane where all properties are equal in all directions, the materlal 1s termed transversely isotropic giving only 9 independent terms. If there isan infinite number of ~ planes of material property symmetry, then only 2 independent terms occur {usually taken as Young's modulus Z and Poisson's ratio v; the modulus of rigidity G can be computed from these two tenms as G = EA2(I + V))) For a laminated composite structure, a state of plane siress is typically assumed, and some researchers include the transverse shear effects. Each lamina is treated as an orthotropic layer as indicated in Figure 1, Within each layer, the stress-strain relations for an orthotropic material are writen inthe ; a ae Figure 1, Material coordinate system as related to lobal x-y system,0 NE KNIGHT Principal material directions where direction 1, denotes the fiber direction ‘and direction 2 denotes a direetion transverse to the fiber Gx) [21 2 0 0 07 fem on| |Q2 2 0 0 of len af=]0 0 Oe 0 0 {260 o | }0 0 0 e& of |265 as} 10 0 0 0 asl (26, here the Q, terms denote the reduced stiffness terms written in compact form. ‘These reduced stiffness coefficients are related to the engineering material properties by ® Where £; is the longitudinal elastic modulus, E, is the transverse elastic rodulis, vy is the major Poisson ratio, Gp is the inplane shear elastic ‘Modulus. and Gis and Gy ae the transverse shear elastic modol. Then for the given fiber orientation of each layer relative to the «canis, the trans- formed reduced stifinesses areQ, obtained in the form: Di =Qrco'04210,-+20.sin? Beas? 0+ On sin @ Dis = Qs! 0+2(0; +20? oe 6 +0. c00' (Q\1 + On ~40,4)sin® B05" 0+ Q.(sin* 0+ cos! A) NONLINEAR POSTBUCKLING u Ds =( ~ Qs —20ss)sin Boos? 0+ Qn — Os +20,e)sin" Bcos Oss = (hr ~ O12 ~ 204s) 8in? 86088 +(O}2 ~ Orn +2044)sin O c0s' @ Bis = (Qi + On ~20.2 ~204s)sin® Deos? 0+ Og(sin* 0+ cos" 8) Dy = Oupcos? 0+ Osgsin® o Dis = Qu sin? 6+ Qysc0s? @ Dis = Qusin cos 0+ Gass co Hence, the stress-strain relations including the transverse shear terms are dobiained in the x-y coordinate system using the wansformed reduced stiffnesses: Gi) [Or Br De 0 07 [ee leo] fa @ & 0 0! |e Gs Ge Be 9 0 | 126, 10) 0 0 0 Oy Os} |e. 0 0 0 Gs Gs} (2e., For two-dimensional problems wherein one dimension is significantly smaller than the other two (Le., plane stress membrane or plate bending), kinematic assumptions are imposed on the displacement field, First-order assumptions allow the in-plane displacement field components to vary Iinearly thravgh the thickness. For example, 14, y2) = WCE) + 20,08 9) us 65. 762) = VCR y) + 2049) an 0%, 2) = WGK 9D Furthermore these bending rotations (8, and @,) can be related to the deriva tives of the out-of-plane displacement if transverse shear effects are ne~ slecied. In two-dimensional problems, stress resultants are usually defined since the mechanical behavior through the thickness is assumed a priori. For plate and shell formulations, three sets of stress resultants are defined based ‘on the assumed kinematics ofthe displacement field. Then the stress resultants can be defined as2 NE KNIGHT tfcou (Bhtileb= o where Nj, Ny and Nate the in-plane membrane stress resultants, M,, My, land M,, are the bending stress resultants, and Q,, and Q, are the transverse shear stress resultants. Then by substitution of the kinematies given by Equation 11 into the strain-displacement relations (see Section 1.3.3) and ‘grouping the terms by powers ofthe thickness coordinate z gives the point Stains at any point through the thickness of the laminate fu & Xe Sy f=} Gy FE) ey) (2eh) [2% ay lolicre the supcsscsiye zee inuicates that hese ens are comsiunt taro the thickness based on the assumed Kinematics. Integrating these expres- sions through the thickness of the laminate gives the constitutive relations in terms of stress resultants and strain measures (midplane strains and ‘changes in curvatures); Aw By Ba Be 0 0) fed, 4s Ba Ba By 0 0] Leh, As Be Bu By 0 0 | Ih fe Dr Dr Pe 0 O/JaL iy Bx Dy Dy Dy 0 0 ee 0 0 0 0 Sy Sul ly aj lo 0 0 o 0 Ss Ss] (2) ‘where the Ay coefficients represent the extensional or membrane stiffness tecms, the B, coefficients represent the bending-extensional coupling stiff ness terms, the D, coefficients represent the bending stiffness terms, and the 5, coefficients represent the transverse shear stiiness terms based on fiest- NONLINEAR POSTBUCKLING a order, shear-deformation theory (FSD), In classical laminated plate theory (CAPD), the §; terms are neglected, and the transverse shear strains are treated as zero. This change in kinematics affects the Finite element formu- Jation in that the continuity requirement is increased from C® for FSDT to for CLPT since the bending rotations are equated to the slopes of the out- of-plane deflection. Both the CLPT and FSDT formulations convert the ‘multilayer laminate into an equivalent anisotropic single-layer system using classical lamination theory as described by Jones (1975). Other computa- tional models for composite laminates have been proposed and have been reviewed by Reddy (1989), Noor and Burton (1989, 1990), Noor (1992), and Reddy and Robbins (1994). Stffoess coefficients for equivalent single-layer theories are computed using the transformed reduced stffnesses and inte ‘grating through the thickness of the laminate to obtain ; N= Yama ij=h 26 SOMa dy) hint 2.6 as) gus Bama i H= L Wa- 1 should be noted that while transverse sheat flexibility is included in the clement formulation, the resulting transverse shear strains based on the assumed kinematic field are constant through the thickness, and the trans verse shear stresses recovered from the constitutive relations are discontinu- ‘us at layer interfaces in the laminate, Post-processing of the solution by integrating the equilibrium equations through the thickness and imposing transverse shear siress continuity at layer interfaces is needed in order to obtain accurate interlaminar stresses. Procedures for accurate recovery of interlaminar stresses by integrating the equilibrium equations have been escribed by Engblom and Ochea (1985), Govindarajan er al. (1993) and (Canisius and Foschi (1993) among others. Englestad eral. (1992) isthe only ‘paper that compares interlaminar stresses recovered by integrating the equi Hibrium equations with those oblained from the interlaminar constitutive relations for @ postbuckled composite panel4 NF KNIGHT lacement Equations ‘Various forms of the stain-displacement relations are widely used in finite clement cades, For problems associated with postbuckling and collapse of clastic structures considered herein, 2 Lagrangian formulation is used wherein, the deformations are referenced to the original configuration and differen- tiation and integration is performed relative to the original configurat ‘These strain-displacement relations are defined using the Green-Lagrange strains and are given by (168) (16) (162) (6a) (162) ae ® Where €y Gy € are the normal strains and 7. Yur Ye are the shearing strain. For plates and shells, the transverse normal strain is neglected based on the transverse inextensiility assumption. As such, only the nonlinear terms associated with the in-plane strain terms are retained. That is, aay NONLINEAR POSTBUCKLING 1s SSA] are) % am) % are) Furthermore, some researchers only include the nonlinear terms associ- ated with the out-of-plane or transverse deflection w which gives the so- called von Karman strains: du, 1 ay f ean lS «aay by (ab) Ou, Oe, it =a (ase) (18d) (18e) ‘The von Karman nonlinear strains work well for many plate and shell problems provided no plate or shell junctures occur such as for finite element models of built-up structures, When modeling a stiffened panel, if 4 stiffener is modeled using shell elements and attached to the panel skin and the structural response involves global buckling mode shapes, then the16 NR KNIGHT ‘out-of-plane deflections of the skin are related to the inplane deflections of the stiffeners. In such cases, all nonlinear terms, at least for the inplane strains, ae necessary. In addition, solutions predicted using the von Karman nonlinear strains are reported by Kim and Chaudhuri (1995) to overestimate deflections in the advanced postbuckling regime for transversely loaded, laminated curved panels, 1.34. Compatibility Equations “The six strain components can be completely defined at a point provided the displacement fied in terms of u, v, and w is known. However, if the strains are given, the six partial differential equations given by the linear terms of Equation 16 must be solved and cannot be taken as arbitrary functions of x,y, snd z The strains are subject to conditions of compatibility such that a unique, single-valued displacement ficld is obtained. These ‘compatibility equatiis are developed by examining derivatives ofthe normal and shearing strains ia order to obtain common forms of derivatives of the displacement field variables on both sides ofthe equations. A unique single ‘valued displacement is ensured by satisfying the six strain compatibility ‘equations for three-dimensional elasticity: feu “Oya: Oe For two-dimensional plane elasticity, here is only a single compatibility equation; namely, 20) For displacement-based finite element models, these compatibility equa tions are guaranteed (0 be satisfied since the displacement field is assumed rather than either the stress or strain fel. NONLINEAR POSTBUCKLING " VARIATIONAL FORMULATIONS AND THEIR FINITE, ELEMENT MODELS, ‘Variational formulations and their associated finite element models are described in this section. First, an overview of variational principles in solid ‘mechanics is presented as a road map forthe different principles, Then the single-field displacement finite element model is presented, followed by the assumed-stress hybrid model, and finally the mixed finite element model is presented, LA.L. Overview of Variational Principles in Solid Mechanics, ‘The field of solid mechanics is richly blessed with variational statements from mathematical physics. These variational statements range from single- field statements in ferms of displacements only or stresses only to four-field statements involving displacements, strains, tresses, and tractions, These principles ean be readily derived by starting with the basic equations of ‘oalinear solid mechanies, and conversely the basic equations of nonlinear solid mechanics can be derived as subsidiary conditions forthe stationary valves of the variational principles, Traditionally the principle of virtual ‘work has been used in the development of most finite element codes, This principle is « single-ield principle wherein the nodal displacements are the primary unknowns, and strains and stresses are socondary or derived qua tities from the displacements. :By imposing or relaxing various assumptions, alternate variational prio- ciples are easily derived as indicated in the flow diagram presented in Figure 2, This type of flow diagram was first proposed by Washizu (1968) ‘and later modified by Pian and Tong (1969) and Pian (1973). The principle of of virtual work assumes only that a kinematically consistent set of displacements and strains exists and makes no assumption as to the const tutive relations, [fan elastic material is assumed and if the displacement boundary conditions ate satisfied, the principle of minimum total potential ‘energy results. If the assumption of a kinematically consistent set of displacements and strains is relaxed us well as relaxing the assumption on the displacement boundary conditions, then the four-field generalized vari- ational principle is obtained. Next if Cauchy's formula is used to relate twactions to the stress tensor, the Hu-Washizu three-field variational princi- ples obtained. Then by again enforcing the assumption that the displacements ‘and strains form kinematically consistent set and also by defining the strain ‘energy in ferms of the stresses, strains, and complementary strain energy as Ua 04 e-UE eu ‘the functional form of the Hellinger-Reissner two-field variational nrincinle18 NF KNIGHT ES vison arenes |S dn ea Boa ca Price tt) ae ages rai aap ‘sontiyh Ne aes {Sonim acer 12a taaatnaeay (CSuisenstsgoesrsacaad =) mincantereesire iy Ti 1 ea Maino ctaeee ca tae Figure 2. Flow diagram of variational principles from solid mechanics 1.42, Single-Pield Displacement Finite Element Models ‘Single-field displacement finite element models are nearly always devel- ‘oped from cither the principle of vittwal work of the principle of minimum {otal potential energy wherein the unknown field is the displacement field. In such models, the element equations are developed by substivuting the NONLINEAR POSTBUCKLING 19 finite element approximations For each component ofthe displacement field Into the variational statement of the problem and then extremizing the resulting expression with respect to each nodal variable. For the principle ‘of minimum total potential energy, the resulting expression for the total potential energy of a linear elastic system has the form: Sada 2 ‘where dis the vector of element nodal degrees of freedom, kis the element stiffness matrix, and, is the vector of element nodal loads. Extremizing this expression gives the elemental equations as ax 4, f,=0 23) ‘The element stiffness matrix is computed as followed) b= [osterarenro,av Bicbav % here HT relates the deformation gradient vector to the strains, Ts a block- ingonal matrix wherein cach block contesponily «3 cosnponent of the isplacement field and represents the inverse of the Jacobian malrix for the clement, C is the matrix of constitutive relations for the element; and Dy is a matrix of derivatives of the shape functions for the element. Far some simple clements, the integration required in Equation 24 ean be performed exactly; however, for the general case, such integrals are evaluated using Gaussian quadrature or perhaps symbolically. The use of symbolic manipu- lation tools and software are becoming more and more a part of the devel- ‘opment process for new finite element formulations (e.g., Korncoff and Fenves, 1979: Noor and Andecsea, 1979, 1981; Tan, Chang and Zheng, 1991; Rengarajan, Knight and Aminpour, 1995). Numerical integration or quadrature replaces the integrals with weighted sums of the integrand evaluated at selected points. Typically quadrature ‘ules are used based either on an interval of 2er0 to one (Lobatto quadrature) ‘or over the range negative one to positive one (Gauss quadrature), The integration can be shown to be exact ifthe integrand isa polynomial of onder = 2NQP-+ J where NOP means the number of quadrature points rounded ‘o the next highest integer. Infinite element analysis, three terms appear in ‘elation to this numerical integration process. Full integration refers to the number of integration points required to evaluate the integral exactly if the edges of the element are stright, the nodes are evenly spaced, and the20 NE KNIGHT element has uniform thickness. Reduced integration refers to any order of ‘numerical integration lower than that defined by full integration, Itis bounded. fon the lower end by the nuraber of integration points required to exactly Integrate the volume of the element regardless of the element's shape. This is usually one order lower than that required for full integration. Selectively reduced integration refers to using full integration for some terms of the clement stiffness matrix and reduced integration for other terms (e.g, fil integration for the membrane and bending terms and reduced integration for the transverse shear terms). While reduced integration does generally soften the element and seduces the amount of computational work at the clement level, it also generates zero-energy modes within the element that ‘ean corrupt a solution. Such problems are described in many finite element books such as Cook et al. (1989) and Cook (1995). Assembly of the element equations gives a global or assembled system of equations in the form K Dep tine es) where Kis apositive-definite, symmetric, sparse matrix. The solution to this, system of linear algebraic equations is usually obtained by using a direct, method based on Cholesky LU-decomposition, This process involves three steps: the decomposition of K into the provdict LU; the forward-reduetion slep which solves LX'= P; and the back-substitution step which solves UD = X. Evaluations of advanced equation solvers, both direct and iterative, have been reported by many researchers including Agarwal, Storaasit and [Nguyen (1990), Poole (1991), Poole, Knight and Davis (1992), and Bathe, ‘Walezak and Zhang (1993). Recovery of secondary variables such as the element strains and stresses is done after the global displacement solution vector is obtained. Extraction of the nodal displacements associated with each element is based on the connectivity array of the element, and the elemental strains and stresses at selected points within the element are computed from & = HT Dyd,=B a, 26) 0, =C & ee where in most cases and Dy are fonctions of location within the element. Usually finite element developers compute these secondary quantities at the so-called Barlow points (see Barlow, 1976) which correspond 10 Gauss ‘quadrature points one order less than required for full integration of clement stiffness matrix. However, various post-processing codes require either nodal values or a single value per element. To obtain nodal values, NONLINEAR POSTBUCKLING at some global smoothing algorithm can be used such as nodal averaging, topological interpolation, or even a projected least-squares interpolation. If only a single value per element is needed, the value atthe element centroid is typically computed. ‘These single-field displacement models are the most commonly used ‘models in finite element software. Their formulation is straight-forward. Recovery of secondary solution variables such as strains and stresses in volve element level computations and differentiation of element shape func tions. This results in discontinuous strains and stresses across interelement boundaries. Often post-processing techniques are used to obtain smoothed solutions for these secondary variables, but smoothing can mask trie discontinuities inthe model so the analyst must proceed carefully. Smooth- ing element stresses, for example, over a two-dimensional domain with thickness variations is improper. 1.43. Hybrid Finite Elément Models Hybrid finite clement models may be developed from different perspectives: several of which are described by Alusi et al. (1983), One approach is 10 ‘assume one set of displacement approximation along the element boundaries and a different set of displacement approximations within the element. This ‘approach is called the assumed-displacement hybrid method and gives rise {othe so called enriched elements which may feature a closed-form exact solution for some discontinuity (e.g., Tong, 1984: Kaya and Nied, 1993; ‘Krawezuk, 1994), Another approach isto assume similar approximations for ‘more than one field (Le., displacements and stesses) but only impose interetement continuity on one field (ic. displacements. This approach is called the assumed-sttess hybrid method and was first proposed by Pian (1964), Inthe assumed-stress hybrid method, approximations are assumed for the stresses and displacements of the form: o=Pp, uaNd, where P represents a matrix of polynomial expressions forthe stress {Bis a vector of stress parameters for the clement, N is a matrix of shape funetions for the displacement field, and d,s the Vector of nodal displace ‘ment unknowns. ‘Typically the stress field is assumed such that the equilib- ‘ium equations are satisfied within the element. These approximations are then substituted into a multi-ield variational statement stich a the Hellinger- Reissner principlengage [aT as- > 4, A, 28) \where the finite element approximations are made for w and 0, 9 represents ‘2 matrix of differential operators that relate the displacement field to the ‘train components, and C is the constitutive matrix. The last term in the functional given by Equation 28 involves the work done by concentrated point forces and/or moments acting at x; and J ranges from 1 to NCONC. Substituting the finite element approximations given by Equation 27 into this functional and then imposing the stationary condition (i= 0) with respect to the nodal displacement degrees of freedom and the unknown, stress parameters yields the elemental linear stress analysis equations of kad 2) ‘where the linear stiffness matrix is computed from kaP wT 30) na [ pre rav en T=| Planay. (32) where P is @ matrix of polynomial approximation functions for the stress field and N is « matix of shape functions for the displacement field. These clement expressions are derived using the linear strain-displacement rela tions atthe element level and are evaluated using symbolic computations (sce Rengarajan ef al, 1995), Assembly of the element equations gives a slob or assembled system of equations having the same form as the single Field displacement model even though more computational work has been performed at the clement level in evaluating the matrices Hand 7. ‘By using the noalinearstrain-displacement relations and then linearizing the equations, the geometric stiffness matrix can be derived. The element {intemal force vector forthe hybrid stress element is readily computed using the low-order corotational approach and is given by Sin =T Bb oa) In the low-order corotational approach, the deformation process is decom- posed into a rigid-hody-motion component and a strain-producing compo- NONLINEAR POSTBUCKLING 23 nent wherein the linear strain-displacement relations are used atthe element level. The high order corotational approach is similar except thatthe nonlinear strain-displacement relations are used at the element level, For a geometri- cally nonlinear problem, this internal force vector represents the local in- temal force vector which must be projected fo the global system using the ccorotational utilities described by Stanley and Nour-Omid (1990), 1.44, Mixed Finite Element Models Mixed finite element models are defined as finite element models with nodal variables belonging to more than one field (e.g., Noor, 1980; Aduri et al 1983). Mixed finite element models are somewhat related tothe assumed: ‘stress hybrid elements in thatthe formulation of mixed elements also star ‘with a multi-ield variational principle; however, no assumption is made between equilibrium and the nature ofthe stress field approximations. The stress field is assumed in terms of unknown ov, vatiables just as the displacement field is assumed. That is, o=No, d, 64 ‘As such, the primary variables of the mixed finite clement formulation include both displacement and stress nodal values, ad interelement conti- ‘ity of the displacements and stresses is now inherent in the formulation. In this formulation, the stresses are computed directly and not recovered by differentiation of the displacement field. Two sets of element equations are obtained at the element level F048, 4,=0 S29, =P @5) ‘The First set of equations represents the inverse of the constitutive relations (ce. strain-siress relations), while the second set of equations represents ‘equilibrium conditions, The element matrices F, and 5, are defined as f=faretwar oo Sf away en ‘These element equations are then assembled into a global generalized stiff- ness matrix of the formm4 NE KNIGHT =F S][e]_[o ¥ ollotntet on where os the global vector of nodal stress unknowns, D isthe global vector Of nodal displacement unknowns. The soltion ofthis system of equations involves the decomposition of a negative semi-definite global generalized stiffness matrix Gee Johnsen, 1972) “Mixed finite element models involve substantially more nodal unknowns than displacemen-bascl models or even assumed-stress hybrid models, For ‘example, shell element based on first-order shear deformation theory will have five nodal displacement degrees of freedonn (six ifthe driling freedom, is included) and eight rodal stress resultant degrees of freedom, Thus a Anode shell clement will involve a total of 52 degrees of freedom for the ‘xed approach while only 20 degrees of freedom for the displaceiet jbased approach. However, the soltion obtained using the mixed nite einen! mode does impose inferclement contimily ofthe clement sf6iex This requirement may be related and the stresses Computed in terms of the displacement degrees offieedom before assernbly so that the global assem bled equations retain only displacements as iodal unknowns, In this ease, computations] work por element le eimitr to that ofan acexmtod sro hybrid clement. The potential of mixed formulaions has heen described by Noor and Peters (1990) 1S. SOLUTION TECHNIQUES FOR NONLINEAR SYSTEMS. 15.1. Historical Perspective ‘Nonlinear finite element analysis procedures have been under development for more than three decades with temendous strides being taken and yet stil tremendous challenges ae still being faced. Some ofthe fundamental work has been presented by Hartung (1973) and Hinton, Owen and Taylor (1982). Fr these nonlinear problems, the interplay between various aspects of the problem and the analysis (such as structural response, problem formulation, approximations sumerical solves, iteration convergence, interpretation of results, boundary conditions, loading) ean become entangled eausing ana- Iysts grea difficulties in obtaining and verifying robust aecurate solutions. ‘Shell analysis has always challenged analysts with spatial modeling issues, response complexities, and formulation aspects (e., Almoth and Starnes, 1975; Hartung, 1971; Noor, Belytschko and Simo, 1989; Palazotto and Dennis, 1992; Stanley and Felippa, 1986; Stein, Wapner and Wriggers, NONLINEAR POSTBUCKLING 2s 1988; Wempner, 1992), While many issues have been resolved, new issues hhave emerged as the computing technology and our understanding, have evolved. Solution techniques for systems of nonlinear algebraic equations evolved, from the simple methods used to find the zeroes (or roois) of a single nonlinear equation. Early researchers in nonlinear finite element analysis attempted to exploit the solvers available for first-order ordinary differential equations by re-casting the algebraic equations in that form (see Stricklia and Hasler, 1977). These methexls provided ways to exploit certain Features of the problem and were computationally simple and fast; however, like ‘most techniques for initial-value problems, they were also susceptible to solution drift and numerical errors. Because ofthese issues, most researchers tumed to an incremental-iterative procedure based on the Newton-Raphson ‘method. This method has been widely used as a root-finding method of a single nonlinear equation, and various aspects of the method are readily explained from a geometric point-of-view using a single degree-of-freedom system. In addition, this method can be readily extended 10 a system of pponlinear equations and represents the most widely used method for solving nonlinear system of algebraic equations. ‘There arc many variants of this method available that are mostly asso- ciated with tying to reduce the computational effort required to solve a particular problem — i.e, related to the frequency of updating the tangent Sfiffaess matrix and the solution-marching control strategy. Numerous pa- pers and books are available that describe solution procedures, outline ‘computational algorithms, and deseribe structural simulations (c..,Almroth ‘and Brogan, 1977; Bathe, 1996; Bergan, 1980; Crisfield, 1992; Felippa, 1976: Hinton, Owen and Taylor, 1982; Stricklin and Haisler, 197, Solution procedures based on the Newtor-Raphson method require the evaluation of the residual vector and the tangent stiffness matrix, the solution ofa linearized system of equations, and the updating of the solution vector based on the ‘obtained incremental values. Some researchers modify the method so that the tangent stiffness matrix is re-evaluated only periodically based on some criterion such as slow convergence, while others update the tangent stiffness ‘matrix only at the beginning of each load step. When this update is per- formed, the clement tangent stiffness matrices must be evaluated and then assembled, followed by the imposition of boundary conditions and decom- position of this updsted tangent stiffness matrix. This step is perhaps the ‘most time consuming past of the computatios ‘These days, most finite element codes provide a nonlinear analysis ea- ppability for both static and transient dynamic problems. In most cases, the [Newton-Raphson procedure is available and used as the default solution algorithm. In addition, some codes use the full Newton-Raphson method as26 NF KNIGHT siering Behavior career Une Soliton Sortening Behavior Datisction igure 3. ‘Typical pes of nonlinear strictural behavio. their default method in order to provide robustness for novice users of t ‘code. As an alterative, quasi-Newton methods such as the Broyden-Fletcher- Goldfarb-Shanno (BFGS) method as described by Matthies and Strang (1979) and Bathe (1996) are often used. Other quasi-Newton methods are reviewed hy Denpis and Moore (1977). Nonlinear problems are very dif ficult and itis nearly impossible to anticipate all possible solution combi= nations tha users may attempt. Users of nonlinear codes need tobe especially cognizant of the structural problem being solved, options and defaults provided by the code developers, ‘computed results.of. the simulation. In the subsequent sections, various aspects of some of the more comion solution procedures are discussed. year Static Analysis 1.5.2. Solution Techniques for Nor ‘The nonlinear structural response considered herein is typically one of two types as indicated in Figure 3. Ifthe structure tends to inerease in stiffness as the load level increases, then the response is referred to asa stiffening behavior. One example would be a beam or plate with a transverse load. If the structure tends to decrease in stiffness as the load level increases, then the response is referred to 2s a softening behavior. One example would be aan arch or shell with a transverse oad acting towards its center of curvature. Using a linear analysis method tends to give conservative solutions for structures exhibiting a sliffening behavior and unconservative solutions for structures exhibiting a softening behavior. Stiffening behavior is genctaly straightforward to obtain; although in some cases, very small load increments are required. As the load is in~ NONLINEAR POSTBUCKLING. a Extension of Linear Solution Snap-Through Load Deflection Figured. Typical softening behavior exhibiting limit points wih through and sp-ack behavior - ‘creased, the structure stiffens typically due to membrane steiching. For ‘example, trampoline curface will exhibit a atffening behavior as the surface is deflected. As the membrane Strains increase, their nonlinear terms become increasinaly important and add stiffness to the structure. As this nonlinear stiffness increases, large load increments produce only small de- Aleetions until a vertical tangent is reached. In al eases, the tangent stiffness ‘matrix remains positive definite and its determinant increases in valle, Softening behavior leads to overall structural collapse, snap-throxgh and! lo snap-back conditions as part of the nonlinear solution as illustrated in Figure 4. This type of behavior is associated with the loss of positive- finiteness of the tangent stiffness matrix and hence causes numerical difficulties forthe analyst. In most finite element codes, the equation solver assumes thatthe tangent stiffness matrix remains positive definite through the analysis, and when limit points are encountered, the equation solver ‘often fails. For example in Figure 4, as the load is increased from zero (load control), the response softens until a point is reached which corresponds 10 a horizontal tangent (L., zero determinant of tangent stiffness matrix) and if the load is increased further the structure snaps-through. On the other hand, if a key deflection variable is incremented (displacement control), vertical tangent is reached and a snap-back behavior may be present. Softening behavior is also related to another type of structural response Which depends on the applied loading. This response is associated with buckling of structures due fo in-plane loads and is generally related to2 NE KNIGHT Extension of Linear Solution Possible Mode Interaction Load Detlection Figure 5, Typical softening behavior exhibiting isolated critical points reduction in stiffness without the response exhibiting a limit point as shown in Figure 5, Buckling is associated with transitioning trom one deformation ‘mode to another (or adjacent equilibrium configuration) at & éorimon load level. That is a perfect column loaded in compression will respond as a rod ‘until the applied load reaches a critical value (the buckling load) at which time the lowest energy state now becomes a beam bending snode. In this ‘case, there exists « load level with two adjacent equilibrium configurations: ‘one with a rod or membrane behavior and another with a beain or bending ‘behavior. These critical values correspond fo the eigenvalues ofthe structure based on a given stress state. Often times these eigenvalues are isolated or spaced from one another. For example, the eigenvalues of a simply sup- ported column of length £ with an axial load P are given by P, and are widely spaced from each other and hence called isolated bifureation points. Such problems are dificult to solve even though the load-9 mesh, the results shown in Figure 18 also indicate that all models agree until well into the End ispacement ut t0) Figure 18, Bifect of sell oometry using isoparametric elements — 99 mesh NONLINEAR POSTBUCKLING 2° — conversed aotton =O Tota Lgranaion —— Highrondor srotation End eotacement wit10) [gure 19. Comparison of different approaches fr eating lrgeeflection, lage rotation problems — 9x9 mesh and 4LAG element. ppostbuckling regime where large local changes in curvature exist and better representations of both the shell geometry and the shell collapse response are required, Formulations for treating large deflection, large rotation problems influ- cence the prediction of the postbuckling response, Problems with small ot moderate rotations are readily handled by the total Lagrangian approach. As. sotations become large, a corotational approach is often used. The interae- tion between the approach for treating large rotations and the shell clement formulation is evident in Figures 19 to 22 for the postbuckling response of 8 circular cylinder loaded in axial compression. The results obtained using models with 4-node elements use a 9%9 mesh and those for models with 9-node clements use a 9x9 mesh. The posibuckling response shown in Figure 19 obtained using the 4-node Lagrangian clement with uniformly reduced integration (4LAG) indicates litte difference between the total, Lagrangian approach, the low-order corotational approach, and the high- order corotational approach indicating that this collapse problem does not ‘exhibit large rotations. As shown in Figure 19, the solution obtained using the total Lagrangian approach required smaller load steps than either corotational approach. However, the differences between these solutions with the 4LAG element and the converged solution becomes quite notice: able as the solution progresses well into the postbuckling regime ‘The postbuckling response shown in Figure 20 obtained using the 4-node ANS element (4ANS) indiestes that the total Lagrangian approach and the high-order corotational approach give the same response with litleo NE KNIGHT — converged eoton = FT tower corotation —S= 8% 8! toworder cotton S= 9x9, Lagrangian —+— 940, nighordercortaton 20 040 7 80 End apincomont 0) re 2 Companies ies rein ge detect ae ce poe cS ethan SANG ent om Kage a, 199) difference in ether the size of the load stops or the total numnber of load step. Using the low order corotational approach, a slightly stiffer respnce is predicted indicating that the 4-node ANS element is sensitive to the form Of the element strain-displacement relations. “The postbuckling response shown in Figure 2) obtained using the STAGS— 410 element (48TG) indicates thal this element is very sensitive to the End dlapiscement U0) Figure 21. Comparison of diferent approaches for eating lage-deflecton, large frition weak 8 9 mesh nd AST lement rom Kaveh ea. 1995) NONLINEAR POSTBUCKLING ot o 20 a0 eo a0 End depicoront wt) Figure 22. Comparison of diferent approaches for treating arge-deflection, large rotation problens 99 mesh and 8-node elements ‘approach used for treating the large deflection, large rotation problem. The low-order corotational approach predicted the collapse load but failed to predict the postbuckling response. The total Lagrangian approach does predict the postbuckling response but predicts a substantially stiffer re- sponse. The high-order corotational approach predicts the postbuckling response and agrees with the converged solution. This behavior is due in part to the STAGS-416 element formulation and sensitivity to warping “The postbuckling response shown in Figure 22 obtained using the 9-node ANS clement (9ANS) and the 9-node Lagrangian element with uniformly ‘reduced integration (9LAG) indicates thatthe total Lagrangian approach and the high-order corotational approach again give the same response with litle difference in either the load step sizes or total number of steps. Using the low-order corotational approach, numerical difficulties are encountered for both 9-niode models, and a predicted collapse load is not obtained, 16.3. Postbucl 1g Response of Composite Panel ‘A fla, rectangular, unstiffened, 24-ply graphite- 10 miesh of elements over the entire panel ‘The modet has a (otal of 2083 active degrees of freedom, Prior to performing the nonlinear postbuckling analysis, a Tinear stress analysis and a linear bifurcation buckling analysis are performed to aid in understanding the structural response of this class of composite panels. From these analyses, the prebuckling membrane solution and the critical eigenpair are determined Which are used to define the initial load level and load step size for the nonlinear analysis. In addition, an initial geometric imperfection is formed ‘based on the buckling mode shape with a maximum amplitude often percent of the pane! thickness. ‘The postbuckling response determined using the traditional approach is given by the dashed line in Figure 24 whece the filled circles denote con- verged solutions. This structural response is obtained by starting at a load level below the buckling load and inerementing the load until well into the ppostbuckling range. An adaptive solution strategy is used to adjust the load step size during the analysis; however, the arc-length control procedure is not used in the present analysis. The size of the initial load step is cut first64 NE KNIGHT < ad = | Experiment g (Stames and a 2D it Faua Zee wes 1 4 =| 0 1 = 3 Normalized maximum transverse deflection, w/t Figure 24, Postbuekling response comparison for experimentally obtained results and numerically determined results based onthe tational approach in half and then halved wo more times as the lead level approaches the buckling load. Beyond the buckling load, the load step size is increased but never regains the original load step size. A convergence tolerance of 10° ‘on the norm ofthe iterative change in the displacements (Equation 44) is ‘used. These analytical results, as well asthe analytical results of Starnes and Rouse (1981), are in good agreement with the experimentally obtained results NONLINEAR POSTBUCKLING 65 5 Axial Load dn | 35 aa 2 IL | 3 [experiment | z (Starnes and | 2 Louse 1981) 3 | 5 ou o Global Function | 1 ‘Approach (1 vectors) | o 1 ra 0 1 2 3 Normalized maximum transverse deflection, w/h Figure 2S. Posthuckling response comparison for experimentally obtained results ‘and numerically determined results hased onthe global function approach, “The postbuckling response determined using the global function approach is given by the dashed line in Figure 25 where the filled circles denote “corrected” full system solutions. The first four nonlinear solutions are used as the global functions in the transformation matrix, These solutions corre spond to load levels just prior to buckling which begin to exhibit the anticipated postbuckling deflection pattern (i.c., one corresponding to the buckling mode shape and consistent with the moiresfringe pattern of66 NE KNIGHT Figure 23b). The number of global functions is restricted to four vectors throughout the nonlinear analysis, After forming the initial transformation ‘matix using the frst four nonlinear solutions (Le, ist four filled circles), the transformation matrix is updated thereafter on each return to the full, system of equations (ce, filled circles beyond the first four). However, the “predicted” solution from the reduced system of equations only required two Newton-Rapison iterations (on an average) to generate each “comrected” solution, Between the filled circles, several solutions using the reduced ‘system of equations are typically determined. Again, the structural response is determined by incrementing the load; however, only the reduced system. ‘of equations is solved at cach load step. These results are also in good ‘agreement with the experimentally obtained results shown by the solid line ‘on Figure 25. 1.6.4, Posthuckling of Stiffened Composite Panel “The postbuckling behavior of flat stiffened eraphite-epoxy panels loaded in axial compression has been studied by several researchers. Stamnes, Knight, sand Rouse (1985) reposted the results of both an experimental and analytical study for composite panels with F-shaped stiffeners (see Figure 26). The configuration considered herein hax a panel length of 30.0 inches, panel ‘width of 24.0 inches, and a stiffener spacing of 7.0 inches. The panel skin nd stifener were designed using POSTOP (Dickson and Biggers, 1984) s0 that skin buckling wonld occur and posthuekling strength would be exhib- ited by the structure. The loaded edges of the panel were clamped, and the unleaded edges were free. Details of the panel skin, stiffener, and material properties are given by Stames e? al. (1985). ‘As a starting point in this activity, the modeling fidelity needed for stiffened panels was studied wherein three approaches wete considered, The first approach studied was the smeared stiffener model which is by far the simplest. In the smeared theory, the sifiness associated with each stiffener is “averaged over the panel skin such that even an isotropic skin will exhibit orthotropy due to the smeared stiffeners, The second approach studied was the discrete siffener model which involves the same computational effort, 1s the smeared approach since there is no inerease in the number of equa tions to be solved. Here cach stiffener is represented as a discrete beam ‘clement with EA, EI, and GJ stifinesses and possibly offsets to account for stiffener eccentricity. Lines of stiffness are therefore added to the skin model. The third approach studied was the branched plate model which involves a significant increase in modeling and computational effort. Here ‘each stiffener is represented as an independent assemblage of plate elements {hat are atached (oF connected) to the finite element nodes inthe panel skin, NONLINEAR POSTBUCKLING o (b) Photograph of moire-fringe pattern. Figure 26, Flat, rectangular stisened composite pane! (Irom Starnes, Knight and Rouse, 1985)os NB KNIGHT model. The flexibility of the stiffener is accurately modeled inthis approach since cross-sectional deformations ofthe stiffeners are included. Inersectin stiffeners and diagonal stiffeners that donot fall along node lines in the finite clement model pose added challenges. When the buckling response is an overall buckling mode shape for the structure, the three modeling ap- proaches predict essentially the same result. When local buckling is present ‘ise, the stiffeners contribute to the buckling mode shape of the structure), only the branched plate modeling approach is accurate unless the discrete beam model accounts for cross-sectional deformations. For prismatic struc- tures, PASCO is an excellent tool for analysis and design of axially stiffened panels (see Stroud and Anderson, 1980). Comparisons between finite ele ‘ment buckling solutions and PASCO solutions for a variety of loading ‘conditions and stiffener shapes are given by Stroud, Greene and Anderson (1984), Their results are useful for benchmarking new elements and modeling strategies. In addition to spatial modeling jsstes, two approaches for specifying the buckling prestress state can be used: One-approach involves the specifica- tion of the buckling prestress state a priori without the need to compute a linear stress solution, The other approach involves calculating the linear stress solution using a given set of loads and boundary conditions; pertaps using boundary conditions different from those to be used in the buckling analysis. In the frst-approach, combined. inplane load cases are readily solved, However, for the second approach, defining an appropriate load system and boundary condition set is quite complex — especially for the implane shear loading: condition. Using the first approach for defining the buekling prestress state (ie, priori specification) introduces errors which may be detrimental to the accuracy of the buckling results in cases where the smeared or discrete stiffener modeling strategy is used. This error is usualy introduced because the specified stzess state is applied to the panel skin model and not applied to the smeared oF discrete stiffeners. For the branched plate modeling approach, specification of a uniform buckling prestress state in each section of the stiffener panel is then readily accomplished and correlation with results from linked plate analyses can be done. Using the second approach for defining the buckling prestess state (Le., cealeulating a linear stress state) introduces different challenges. Often itis, desirable to simulate the loading conditions that may be experienced in a testing machine (¢g., uniform end shortening). For an inplane normal pre- sess state, loads and boundary conditions that peshaps reflect a testing ‘configuration are usually straightforward to define and specify inthe finite ‘element model regardless of which modeling approach is used. For branched plate models, (wo sets of boundary conditions are generally needed if the [NONLINEAR POSTBUCKLING 69 ‘0000 Applied Asal tha, 40000) 20000) + Tost = Anaysis | — Extonsion of near probucking path | o 006 “0.08 End-shortening, igure 27. End shortening result intent is to compare with another analytical solution which has a certain, inplane stress state. For example, to calculate the buckling load of a panel loaded in uniaxial compression and simply supported on all edges requires ~ 4 set of pre-buckling boundary conditions which leads to a uniform stress ‘tate from the linear solution — uniform end shortening and simple support ‘boundary conditions ead to a combined stress state from the linear solution. However, for the inplane shear loading condition this is not the case “The postbuckling response is characterized by the end-shortening results asa function of the applied compressive load as shown in Figure 27. The ppostbuckling analysis was performed using the finite element model shown, in Figure 288 which is composed of 612 9-node quadrilateral elements, 2,553 nodes, and 12,079 active degrees of freedom. Experimental and analytical results shown in Figure 27 are in excellent agreement. At load values of the applied load, the structure is in a near membrane state, and the internal load distribution is nearly uniform (see Stames et a., 1985), After buckling occurs, large out-of-plane deformations occur, and a redistribation ‘of the load from the panel skin towards the stiffeners takes place. This load redistribution continues until skin-sifTener separation oecurs, and overall pane failure results as reported by Stames etal. (1985). Deformed geometry plots with exaggerated deflections are shown in Figure 28 indicating large ‘out-of-plane deformations occurring in the panel skin segments between ‘stiffeners and litle oF no out-of-plane cleformation of the stiffeners.0 NE KNIGHT (6) Detormed geometry at maximum load Figure 28, Deformed geometries of the Kstifened panel (Grom Knight et al, 1990). NONLINEAR POSTBUCKLING n Free__b. Re25t0 mm 12334 am thks, 6.35 Roo Be 10275 kN? 03 {1501 eiane ie fo. Figure 29, Hingod-frce eylinder — geometry, properties and loading. 1.6.5. Collapse of a Hinged-free Cylindrical Panel The hinged-free cylindrical panel subjected to a center transverse point load ~ is considered to validte the notlinear aspects ofthe AQA shell clement fased on the low-order corotational formulation. This sratre is shown in Figure 29. The stmctural respoise exhibits a limit pont, a snap-theoush behuvior and then snap-back behavior. ‘The nonlinear response of the pane is shown in Figure 30 for midpoint ‘on the fre edge (Pont b). Two meshes are considered, and the reference Solution is taken o be that obtained using a5 x 5 mesh of 9ANS elements. The resus in Figure 30() indicat the response obtained wing two meshes of the AQH element developed as part of this researc, For both meshes, nearly the same nonlinear behaviors obined. The oly diference appears to be aftr the shell’ free edge has snapped through to an inverted poston ‘he ress in Figure 30() indicate the response obtained using aS 5 mesh of 4ANS elements with both the low-order and high-order cortatonal formulation (ve, NLGEOM = | and NLGEOM = 2, respectively), These results appear to bracket the reference solaton “Te nonlinear response ofthe panel is also shown in Figure 31 for the centerpoint (Point a) To meshes ae considered, andthe reference solution is taken to be that oblained using a 5x5 mesh of 9ANS elements. Then NE KNIGHT Het tan ses 90) Seas Tro agnasan) Kt snes) Figure 30, Hinged Cylinder: Load vs, Free Edge Point Deflection, ‘results in Pigure 31(a) indicate the response obtained using wo meshes of the AQ4 element. For both meshes, nearly the same nonlinear behavior is ‘obtained. Differences appear after the limit point 2s the snap-back response ‘occurs, In this region of the nonlinear response, severe Tocal changes in ‘curvature occur and the 5x $ mesh of 4-node elements withthe low-order ‘corotation approach is not sufficient to reproduce the reference solution Adding more elements (ic, essentially adding more local reference frames) has the effect of refining the solution to obtain the reference solution. As NONLINEAR POSTBUCKLING B Xt win 94) TES soteane Ther ABH ESHA _ eee joecR Na ail 0) ease Figure 31. Hinged Cylinder: Load vs, Center Point Deflection, an altemative, asing the high-order corotational approach should have the same effect, The results in Figure 31(b) indicate the response obtained using 4.5%5 mesh of 4ANS elements with both the low-order and high-order corotational formulation. These cesults are similar to the reference solution ‘up until snap-back occurs, The two solutions exhibit a different character ‘than anticipated from previous results.1” NE. KNIGHT ta vtoty Yon Stine oi prope: =980x 10% eon totesectin* Figure 32. Truncated conical shell — geometry, properties, and loading (from Kaight etal, 1989). 1.66, ‘Transient Response of 2 Truncated Conical Shell “Thiy linear Elastic transient response of a truncated conical shell subjected to-an impulse Toad (initial velocity) was posed as a benchmark problem by Hartung and Ball (1973). The geometry, material properties, and loading is specified in Figure 32, The finite element model shown in Figure 33 has 540, Figure 38, Fite element model of truncated conical shell (from Knight a, 180). NONLINEAR POSTBUCKLING 78 Normal tection, ° ‘0 eo] 0DSS200S(N0 “Time, meroeeoonds Figure 34. Normal deflections at Points A nd on truncated conical shell (from Kaight et a, 1989), 4-node quattilateral shell elements with 589 nodes and 2569 active degrees of- freedom. Two points on the symmetric structural model (Points a and b) are monitored during the transient response. Both points are located at 6.5 inches from the clamped, small diameter edge: one at 6= 0° (Point 3) fone at 6= 180° (Point b). The transient response was computed for an lapsed time of 1,500 microseconds using the Newmark method with a time step of 2 microseconds. The behavior of these two points represented by their normal deflections is indicated in Figure 34 Oblique views of the deformed shape with exaggerated deflections at various points in time are shown in Figure 35. Contour plots of the hoop stress resultant are superimposed on these deformed shapes. These results indicate the added complexity associated with transient structural response predictions. Linear simulations such as the one presented here are readily performed even though the computational task is significantly higher than a static shel! analysis problem whether i is linear oF nonlinear. Nonlinear transient simulations sil pose computational challenges for the finite cle- ‘meat analysts; particularly if plastic collapse, contact, or localized gradients need to be included in the prediction. These results alsa indicate the vole of visualizing the computed solution as an aid to understanding the structural ‘behavior. Inthe present problem, the shell dynamic problem involves the development of both cizcumferential and axial waves along with their in- teraction as shown in Figure 35.16 NB KNIGHT (b) Time ¢ = 2004 sec oe z (a) Time t= 600 see. (6) Time ¢ = 4004 sec. ie (0) Time ¢ = 1000p se 4 (hy) Tine ¢ = 2500p see. (e) Timet Figure 3S. Deformed shapes for truncated conical shell during transient response (Geom Knight eal 1989) NONLINEAR POSTBUCKLING n Uniform end shortening Materia properts: Geometric parameters: 0" pal (68.95 GPa) R= 8.06 i (1582 men) in Som G28.6mm) Figure 36. Circular Cylinder with euouts — geometry, properties, and loading (Grom Knight eral, 1999), 1.6.7. Collapse of a Circular Cylinder with Cutouts ‘The circular cylinder (R/t= 433 and LIR = 1.5) with two rectangular cutouts loaded by uniform end shortening shown in Figure 36 is representative of shell-type strictures with local discontinuities (e.., cutouts for access holes ‘or windows). This problem has also been used as «benchmark problem by Hartung and Ball (1973) for shell analysis computer codes and by Almroth ‘and Brogan (1981) for assessing shel elements, These researchers consid- cred only one-eighth of the shell in their analyses. The results reported herein are compared with their results, and hence only one-eighth ofthe shell is modeled. Symmetry boundary conditions are imposed on three edges and the loaded edge is clamped. Three finite clement models were considered as shown by Knight eta. (1989). Mesh 1 is composed of 101 9.node curved {quadrilateral shell elements, 449 nodes, and 2,012 active degrees-of-free- dom, Mesh 2 has the same nodal layout as Mesh 1 but is composed of 40 4-node flat quadrilateral shell elements. Mesh 3 has the same element layout 1s Mesh 1 but is composed of 101 4-node flat quadrilateral shell elements and has only 124 nodes, ‘A linear bifurcation buckling analysis was performed prior tothe nonlinear clastic collapse analysis. A buckling load of approximately 0.13%) (where Py= 12 xE F) has been reported by previous investigators, and the buck- ling mode shape indicates thatthe vertical edges ofthe cutout buckle local.® NE KNIGHT Figure 37, Deformed geometry fiom Knight eal. 1995) ‘The nonlinear response is characterized by large local radial deflections ‘along the straight edges of the cutouts at a load level of approximately (0.13%, followed by overall shell collapse at a load level of approximately ‘O4OP The deformed geometry with exaggerated deflections is shown in Figure 37 forthe converged solution near the collapse load wherein the one cighth-model geometry and results have been symmetrically reflected to enhance visualization. As the radial deflections near the vertical edges ofthe ‘cutouts develop, the compressive axial stresses are redistributed away from the regions above the cutouts, the stress concentration develops along the straight edges of the cutouts, and the load is carried by the remaining. portions of the shel. ‘A nonlinear analysis ofthe cylinder with cutouts was performed, and the radial deflection, normalized by the cylinder thickness f, for a point at the cylinder midlength at the edge of the cutout (denoted point “a”) is shown in Figures 38 and 39 as a function of the applied load normalized by Py Results for the three approaches to the large deflection, large rotation prob- Jem are shown in Figure 38 for Mesh 1 using the 9ANS element and in Figure 39 for Mesh 2 using the 4ANS element. The solution obtained using ‘Mesh 1 with the 9ANS clement and the high-order corotational approach shown in Figure 38 is taken as the coaverged solution. These results are NONLINEAR POSTBUCKLING p — Lomorder coratation = Total Lagrangian | —+— Highorder coretation \ fe] 4 Radial detection, wait Figure 38. | Compasison of different sppruaches for treating large-deection, large fofation problems — Mesh T using 9ANS elements (fom Knight et a, 1995), — Converged solution —=— Low-order corotaton —O— Total Lagrangian “+ High order corotation oa) o4 Fadia detection, wit Figure 39. Comparison of diferent approaches for treating lrge-efletion large rotation problems — Mesh 2 using 4ANS elemens (fom Knight er al, 1995),0 NE KNIGHT ‘Axial stress at ond b= NONLINEAR POSTBUCKLING 81 12), 0) re Axil stress at mc-plane alt Sito ssf cade ah Sf 0 10 2 9 40 50 60 70 60 90 Circumferential cserdinaise, degraes igure 40, Axial sires at x/L = 0 for various loa steps ftom Knight et a, 1995). ‘consistent with the findings for the previous two shell collapse problems: ‘namely, that nonlinear strain-displacement relations a the element level are needed! for accurate postbuckling analysis and that the 9ANS element per- forms best. Hartung and Ball (1973) reported acollapse load of 0.287 sing a finite difference version of STAGS. Later, Almroth and Brogan (1981) Using the finite element version of STAGS reported a “nearly” converged collapse load of 0.37P 9 using the finite element version of STAGS. The collapse load obtained in this sudy is 0.387, for the converged solution ‘The distribution of the axial stress normalized by 0) (where oy = 0.605 FAR) at the midsurface across the end of the cylinder (x/L = 0) is shown in Figure 40 for various load levels. These results were obtained using Mesh 2 nd the 4ANS element, The distribution ofthe axial stress atthe midsurface ‘across the midplane of the eylinder (x/L= 1/2) is shown in Figure 41 for these same load levels. These distributions indicate that substantial redistribution of the axial stress oocurs as the collapse load is approached. ‘A comparison ofthe postbuckling response of this shell collapse problem for several clement types is given in Figure 42. The slightly higher collapse toad and different postbuckling response obtained using the 9LAG element i attributed to the inherent weaknesses of clements developed using uni formly-redaced integration. ‘The postbuckling response using the 4ANS. clement is obtained for both Mesh 2 (same number of nodes as Mesh 1) and °| 0 2 90 4 60 60 70 0 00 (rcunitereitel coordinate, degrees gure 41. Asis at L172 frou lad (om Ket 1995). i NEEM — Converged solution —— 4 ANS Mesh 3 oa 2 4ANS Mosh 2 —O— 4816 Mesh2 —*— SLAG Mesh? 03, % Radial deflection, Figure 42, Effect of shell clement formulation on posthuckling respanse ond collapse (From Knight er al, 1995)2 NF KNIGHT Mesh 3 (Same number of elements as Mesh 1), These results indicate frst that the 4ANS models approach the converged solution as the mesh is refined, and second that for the same number of nodes the 9ANS model ‘performs better than 4ANS models. The response obtained using the 4STG clement is in excellent agreement with the converged solution up to the collapse load but is unable to proceed beyond the collapse load, ‘These results indicate that the influence of tansverse shear deformations on the nonlinear shell response is minimal. However, the use of higher-order shape functions for the bending field approximations (eg., biquadratte for the SANS element and Hermitian for the 4STG element) dramatically affects the solution as indicated by comparing the response obiained using elements with lower-order shape functions (@.g., bilinear for the 4ANS element) for the bending field approximations. 1.68. Nonlinear Response of the SRB/ETA Ring Interface ‘Mileczident which destroyed the space shuttle Challenger was caused b* the failure of a case joint in the right solid rocket motor (SRM) aft altach- nent segment (Rogers, 1986). The upper end of a lower eylindrieal, motor segment (e.., aft attachment segment) forms a clevis, The lower end of the ‘upper, eylindrical motor segment (e.g. aft center segment) forms a tang witch mates with the lower clevis. Around the circumference of both tang ‘nd clevis ends are 180 holes into which one-inch-liameter connecting pins fne-inserted. Three of the pin holes on the tang end are used as alignment slots to facilitate assembly of the SRM segments, The, seal between two riotor segments is provided by two O-rings in the “inner arm” of the clevis. “The O-rings are compressed by a flat sealing surface on the tang. Several characteristics of the original SRM joint design were identified as potential contributors to the failure (Rogers, 1986). One characteristic ‘was the behavior of the joint under internal pressure load. The motor case expands radially outward due tothe pressure. Because the joint has a higher hoop stiffness than the SRM case wall on either side of the joint, its radial ‘expansion is less than that of the case wall. In addition to the nonuniform stifness in the longitudinal direction due to the presence of the SRM case Joints, a nonuniform stiffness inthe circumferential direction exists in the aft attachment segment due to the extemal tank attachment (ETA) ring. [Nonuniform radial expansion is the primary cause of the relative motion, between the inner clevis arm and the sealing surface on the tang. This relative motion can cause the O-rings fo become unseated and therefore lose their sealing capability. These effects were considered in shell analyses performed by Knight (1987) and Knight et al. (1988). A summary of the finite element models and results are included here to illstrate the com- NONLINEAR POSTBUCKLING. 83 pee Factory is igure 43. Definition of SRB/ETA cing interface, (fom Knight 1987) plexities of nonlinear shell response predictions and capabilities of nonlinear shell analysis tools, ‘The solid rocket booster (SRB) structural subsystem provides the meces- sary structural supputt for dae shuttle vehicle ch the Intiich pad, transters thrust loads to the Orbiter and external tank (ET), and provides the housing, structural support and brackets nieded forthe recovery system, the electrical components, the separation moitors, and the thrust Yector control system. ‘This subsystem assembly consists of the nose cone assembly, the forward skirt ineloding the forward SRB/ET attach fitting, the alt SRB/ETA ring and struts, the aft skirt including the heat shield, the systems tunnel, and other structure for mounting other SRB subsystem's components. Each SRB is approximately 1-4 fect long. Nominal dimensions for the SRM case radius and thickness are 72.6 inches and 0.479 inches, respectively. These corre- spond to a radius-to-thickness ratio of approximately 150. The SRB includes four SRM segments that are assembled at the Kennedy Space Center, and these joints are referred to as the “field joints,” ‘The definition of the SRB/ETA ring interface region considered here and shown in Figure 43 ineludes both of the BTA rings (ring webs are approxi- ‘mately 12 inches apart), a portion of the SRM aft attachment segment including the factory joint at station 1577 (approximately sixty inches of shell stricture), and a portion of the at center segment including the field joint at tation 1491 (approximately 64 inches of shell siacture), The total length of the SRB/ETA ring interface region considered is 136 inches (coughly one shell radius on either side of the ETA rings). The center ofthea NF KNIGHT [SRB/EETA ring interface region is located at station 1511, approximately 20 inches below the aft attachment segment field joint. The BTA ring assembly is comprised of two tapered, partial rings, H-fitings to attach the ET struts, cover plates, and various other intercostals and brackets. The ETA ring assembly extends only 270-degrees circumferential sround the SRM seg- ‘ment. The ETA rings ae bolted every two degrees around the circumference {othe stub rings which ate integral parts of the SRM aft attachment segment. ‘Three struts attach the uft end ofthe SRB and the ET. These three attachment struts are designated the lower strut (P9), the diagonal strut (P10), and the ‘upper strut (P8) “Two-dimensional shell finite element analyses were performed by Knight (4987) and Knight, Gillian and Nemeth (1988). The analyses were per- formed using the STAGS shell finite element code (Rankin, Stehlin and Brogan, 1986), and the finite element models were developed in order 10 study the overall deflection and stress resultant pattems for the SRB/ETA ring interface. Both the field and factory tang/clevis joints are modeled by ‘using equivalent stiffness joins instead of detailed models of the joints. As such, the effect of local joints on the shell response is included; however, local joint behavior (ic, O-ring sealing) can not be obtained from these ‘lobal models, Global shell behavior of the SRB/ETA ring interface can be ‘obtained using an equivalent stiffness join, and an evaluation ofthe effects ‘such as shell collapse and ovalization can be perforiued ccnoss SECTION demas (6° oe ees) Figure 44, Baseline finite clement model of SRB/ETA ring interface (from Knight, 1987) NONLINEAR POSTBUCKLING 85 element mode wed by Knight (1987 forthe SRB) ETA‘ing nrc it shown in igre adh 45 node shel lements nifonly spaced around the sel cicumfeence and 2 clomens alongs ten Synciy bownday canon ae ipod tte forward en of the ode! with he exeption that the longi deen is unestane Simpl-supgrtbewadry condone ae imped a the tend of be model sith ihe exception thatthe rail ection is anvestaine, Ths set of Boundary condons equies the aft end of he shel oemain circular The fre tment mode ad approximatly 12,00 active degrees of feedom, andthe average ser-bandwidh ofthe global stiffs ats I 10, ‘The lner an nolnearslatons foc he bassline fine element med ofthe SRB/ETA rng interice were obtained whon jected oe ner press of 1000 ps. Deford geometnes wth exaggerated deflections Corresponding othe iar and roninearslstons re shown n igre 48 Defooned aeons of he enti inode te shown inte tp a ofc figure an ho ofthe SRM sib ring lone ae shown in he Tower hall. Both nats snlnerslations exit an abr change in deiecons near he ent of he ETA ing sy the part ings or shaped ings). The taal detecton pater rom he nonocar as neats a sfenig of th el epee do ot incon of peomee nouparaes The high fea! bending causes Inge tango shearing und nore forces o evel tw the SRM si ge andthe BTA ng Figure 45. Deformed geometries of the baseline finite element model duc to internal pressure only (from Knight, 1987)6 NF KNIGHT wi # & g Figure 46, Axial distibution of nonlinear radial deflections (rom Knight, 1987) ‘The axial distibution of the nonlinear radial deflections forthe baseline model subjected to 1000 psi internal pressure only is shown in Figure 46 for several circumferential locations. The radial deflections are normalized by the niominal shell thickness and are shown as a function of the SRB station number, Station numbers corresponding to tang-clevis joims (eld and factory) and the upper and lower ETA ring webs are also noted in Figure 46, The first circumferential location (Point A) corresponds to mid- way between the ends of the ETA ring, The radial deflection pattem is denoted as a solid curve, This pattern exhibits a marked change in radial deflections near the field joint at station 1492 and is such that a tang-clevis joint would tend to open. The second location corresponds to Point B which is approximately 180 degrees opposite to Point A. The radial deflection pattem a¢ this location is denoted by a dashed line. The pattern near the field {int is again similar to the pattems at the other locations. However, near the ETA ring webs, the radial deflection pattern is different, and the ening influence of the ETA ring on the shell response can be seen. The third location (Point C) corresponds to an end of the ETA ring, The radial deflection at this location is denoted by a fine with filed symbols. This pattern is similar to that of Point A except for the increased amplituces of the radial deflections. For comparison, two additional curves are shown on Figure 46, One curve represents the membrane solution, including the NONLINEAR POSTBUCKLING 87 RADIAL CIRCUMFERENTIAL LOCATION « DEGREES igure 47. Nonlin (Grom Knight, 1987) lial deflections of SRM stu ring for pe-ftft losd eases biaxial effect, for a uniform thickness (0,479 inches) cylindrical shell with an internal pressure of 1000 psi. The other curve represents the nonlinear solution forthe same finite element model used to generate the other results except the partial ETA rings are removed from) the model. Noninear analyses of the SRB/ETA ring interface model sing the recon- sinucted flight loads for ies 5.3 seconds and 7.2 seconds after space shuttle main engine (SSME) ignition were reported by Knight (1987). Time seconds corresponds to the time at which maximum bending occurs and has been referred to as “max twang.” AC this time, the SRM is unpressutized since it has not yet been ignited, Time t= 7.2 seconds corresponds to the time at which the SRM reaches full pressure (approximately 1000 psi) and lift-off occurs. These nonlinear analyses included the asymmetric loads resulting from the aft ET strut loads and the internal pressure, ‘The radial deflection pattems as a function of the circumferential coordinate 6 are shown in Figure 47 for these two conditions. The ET strut loads at maximum bending result in an asymmetric radial deflection pattern: however, the amplitudes of these deflections are small compared to either the nominal shell thickness or the radial deflections caused by the internal pressure loading only. Deformed eometries for these two pre-liftoff load cases are88 NE KNIGHT ‘Figure 48. Deformed geometries ofbateline model fr pre-lifloff load cases (om Keight, 1987). shown in Figure 48, At SRM pressurization, the overall shell response is dominated by the effects of the internal pressure, and the effect of the ET sirut leads appears to be secondary, 1.7. CONCLUSIONS: Various aspects of finite element analysis for predicting postbuckling and collapse of clastic structures have been deseribed in this chapter. Finite ‘element fundamentals have been reviewed to provide a common basis for readers, The basic equations for nonlinear solid mechanics assuming small strain response have been given and different approaches to the formulation identified. Next, the vasiational formulations commonly used to develop finite elements were presented and discussed. The finite element approxi- ‘mations and assumptions in single-field as well as multi-field models were described and computational aspeets mentioned, Details of nonlinear solu- tion algorithms for postbuckling and collapse problems were examined in ‘detail and implementation details provided. The various numerical solutions presented depict different aspects of performing finite clement simulations ‘af complex response problems. NONLINEAR POSTBUCKLING © While much progress in these areas has been made in the past three decades, much research remains to be done. Problems still semain with clement performance and reliability even though production codes are widely used in analysis and design, As the complexity of the finite element models increase, the likelihood of detecting a problem associated with an inherent clement deficiency becomes inereasingly remote, Some problems continue ‘o overwhelm even the most robust solution algorithms as a result of nom- ‘convergence, mode jumping, and mode interaction, While some analytical \work has been done, it may be several more years or even a decade before these “new” methods begin to appear in commercial software systems, In losing, the finite element analysis tools available today are extremely powerful and able to solve very complex problems. The emergence of advanced high-performance graphies workstations has enabled the develop- ‘meat of very large detailed finite element models that “look like" the Structural component, and analysts need 10 step back to ensure that the physics of the simulation are also accurately represented (@g., see Smith, 1986). As the speed of the computers increase, larger and larger finite clement models will be used. The analyst has the responsibility 10 thor- ‘oughly interrogate the computed solution — just because it is three-dimen- sional with results compated on a supercomputer and displayed in color does ‘not mean that they are correct. Hopefully this chapter has provided some sulditional iusighn for solving nonlinear problems, References 1, Agarwal TK. 0.0, tras nt D7. Nesyen, 190," Paral Vitor Agito for Rapid Sirota Analy on High Perfomance Compuiers." ALAA Paper Nv 0-119, ‘Allman, D3, 1984, "A Compabe Triangular Element lcaing Vester Raton for Plane Blsticty Anais" Conger and States, 1K 1-2) 1-8 3. Almrot, 8.0. an 341. Stmes, Je, 1975, "The Comput a Shell Stability Anyi” ownl ofthe Engineering Mechonice Divion, ASCE, HOLESIS), 73-888 4 “Aim B10. and EA. Bicenn, 1977, “Automate Choice of racedre in Compe {aed Siratural Atay” Computers and Sractres, 1, 35-31 5. Almroih, 8.0, P Stem ond PA. Brogan, (978, "Automate Choise of Global Shape Function in Sacral Avani” AA Jounal 16), 525-528 6. Alot, 8.0. P Stein and EA. Brogan, 1981, "Useof Global Euacton for Improve ‘nen ia ficiency of Nonlinear Analysis.” ALAA Paper 81-0575 7. Amr B.O. and FA. Brogan, 1981, “Conputurna Eceny of Shell lemeets In FIR. Hughes, A idk and Atay eins, Noninear Fine Element Analyte of Ploes and Ses pp. H-168. AMD Vol. 4% ASME, {8% Aminpour Mu. IAD, “Assousment of SPAR Elements and Formation of Sons Basic 2-Dand 3 Flementfr Use with Tested Genenc Element Pros In NP. Sy, ttn, Proceedings of NASA Workshop on Compatorional Stra! Mechanice (53-682, NASA CP-IODID, Pat 2. 9. Anon, 1962. Coleced Papers on Inability of Shell Srucures, NASA'TN D-1510, 10 Aur, SN. RH, Galgher and LC. Zeshewic, editors, 1983, dnd and Mixed Finte Blane Methods, Jon Wiley and Son, Inc, Cicer” 2 B 4 a, 2 2%. m. ». M 2. xm » NF KNIGHT ‘Bartow 1, 1976, “Optimal Suess Locations in Fine Bement Models.” Iernetional “eure for Recricd Methods In Engineering. 12}, 245-251. Alo se dvsson of ire HC), co BBthe, KJ, 199, Fine Bement Procedures. Prestice Hal Bae, 1, J, Walerak and H. Zhang, 1993,"Some Recent Advance or Patel Fite Element Aniyis™ Computers and Sirurune, 484), 51121, Bator, JL, and ©, Dat 197, "larementa Displacement Alors or Nonlinear Problems” Inerationl Jural for Nanerial Methods m Engineering, 15, 1262 ely. 174, Transient Analysis. WD. Pie, K, Saczals and. Schaef, fits, Sucre Mechenes Canputer Programs, pp. 285-276. The University of Ninna Presa, Carotecil Belyeehco, T, 1075, “Nosiner Analyse Desrigton and Numerical Stability” In Wiallerand Ratna Pile, editors Shock and Ubration Computer Programs — Reviews and Summaries SVM 10 pp. 337-362 Beljctko.T and TAA Hoghes tors, 1982, Cmpuraonal Methods for Transient Anata Elaeier Science Publisher. Borzan PG. 1980, "Solon Algotins foc Nonlinear Stra Problems.” Come fr and Stractures, 12, 497-508 Bergan, PO and RWW. Clough, 1972, “Canerzeace Cte fre ‘AA sourna. 108), 107-1108, Sens CAV. ad PN. Fra 1974, Composite Material Mectnc: Struct Mecham Jes AIAA Journal, 128) 173-1186. ‘Bushell D. 1985, “State Collapse: Suey of Methods and Modes of Bebavion Finite Ele i Anat and Design 3 VS=205, nel, 1987 PANDA? — Proprar for Minturn Weight Design of Silene, ‘Compose Lecally Buch Panels” Computers and Simerures, 2(4), 4B-105 (Cin HDG. and RO. Foss, 985," Aspects the Calevation of Transverse Shear Seses in Minin Plates” Coumanictions tv Neral Methods in Engineer ve Proce: ng, 9 651-085 E {Cth 105, "Sens Analyse Laminates wth Holes by Special Fine Element Compete Scars, 31, 9-106 it ad fr 198,opaon tod Scere Egg” Enepeon fu schnce Aba ta) 13 eer 0s Pc Beno Sol fr Sas an ab Whey Sn. Gane DS: Mah th Pa 19 ims nd Apt Fine ‘lon Als Tie onl Wey & Sn, EER 18h, Sree! yc mato Cone Matas ohn ‘iy nt Sm, ne New fone CdR a, Ars nemeneiv itn Psd ht Hales Sesto Compute ond Sct, 13, 3-62 EXANEA, 00 "hnArcLeg cb nding Line Senses a Aes Sect tuna! for Hered Meads in Ege, 1209-129 Perera sR Pte Bont nap tf Sa and Smee — Stume snes a Wy snd Sven Cnehene. Bere ec Siu 117 “QueNewns Mate, Matvaon ad Theo’ Sar ony 1, iets ges, 184, STOR Poul Ope Steer Orin Pans Thay and Cai” NASA CR 259 Perc td Wn st es tera cm cig of Tn ints Unreal Conpssne AE ur of es Meche rem, ”. 2 2 6 6 4 48 ”. NONLINEAR POSTBUCKLING ot [Enghiom, 33. and 0.0, Ochoa, 1985, “Mhough-he-Thickxess Stress Predictions for Laminate Pates of Advanced Compsite Mater” Iereationa! Jounal for Nun al Methods in Engineering 21, 1739-1716. TEngebd, SP, JN. Rody and NE. Keight Je, 1992, “Postbckling Response and Fue Predation of Grghite Epoxy Pes Lonel in Compression ALAA Toure 30). 2106-2112 Flop CA, 1976, “Pocedwes er Computer Analysis of Lage Nonlinear Stet Systeme". In A. Wexler, dior. Lange Figneering System, pp. 101, Petganion res, Oxtrd Fiji, F. BC. Pere and KK. Choons, 1992, “Selstion of Con Parameter in Displacement Intemenaion.” Computers an Sractres, 420), 161-174, (Gane, Hand A uk, 1983," item Triangular Piste Boning Fine Element for Crash Simulation” Conan aad Stcmrs I, 371-579. \Gvinarjan. RAN risa Morty K. Vaaykona and PV Ragham, 193, “Fite heme Estimation of ElancInerlarige Susie ia Laminates." Conponies En reerng, 19). 431-466 Hartong, RE, e6tor. 197, Computer Oriented Analisi hell Strctues, AFFDLTR- Hartung, REF, 1973, Mamerical Satuion of Nopionar Stach! Prats, AMD-VbL ASME. Harang. REE and RE. Bal, 1973. A Conpariaoe of Several Comper Solan 0 Due Senctural Shel Problens. APPDL TR +15, boi, HID. 1986, "Pacical Aspects of Fst Element Computations in Sold Me- ‘haies” Applied Machanie Review, 3911), 1678-168 Ubi, HD 1995, "NonneatSaibd Mechanics from the Commercial Softare View int” Fine Element ie Avabis an Dessn 51, 3-10. Hier. HM. TI Hoghas and RL Teslor 1077. “ingroved Numerical Deion fer Time Itegraon Alport Satur Dysamicn= ntretonal Jounal for Enrihguke Bnginering and Stractral Dynamics, 53, 283-292, Hinton. EDR Owes aC. Tae, eies, 1982, fect Acces bn Nomina Computation! Mechonies,Fineilge Pr, Swans olf NU. 1966, "The Perle Behavior of Tin Cicaar Cyndie Sl in Axil Compson” laser of Technhogs, 1), 1-28, Hurtot, 8. and BP. Sten, 191, Computaiona I Mechonice Texted (COMED) Generic Conttaive Processor Manual. Loctheed Cotact Report F384, Jetnses, T1.. 1972, "Note on Symmetric Decempontion of Some Special Syrametic Matices" Computer Methods Apped Mechanis and Enpiacering 1, 301-306. Jones, RM, 1975, Mechanics of Compotte Materials. Scinta Book Company ‘Kardestuncer Hand D.H. Nomi, eos, 187. nite Element Hands McCaw Hil Book Company Kays, AC ard HF Nie, 1993, “iteface Fracture Analyse of Bonded Ceraic Layers Using Eich Fs TK Kolin, eos Ceramic Coatings, pp. 47-71 ASME Rone No. HODKS3, MD. Vol Kim, D. and R. Chan, 1995, “Full and vow Karwan Geometically Noninea ‘Analyses of Laminated Cylindneal Panels” AIAA Journal, 3411), 2173-2181 nh A Je, 1987, Nowlinear Shel Anaesth SRBUETA Ring Interface. NASA aansied ight. NF, Je, RE. Gillin and MP Nemeth, 1988, Preiminry 20 Stel Anata ofthe Space Shutle Solid Rocket Boose. NASA TML1ODS1S Kipht NF Jr, 1994, "actors Inflancing Nonlinear Static Response Prediction and “Test Anals Corclaton for Compose Parca” Compote Stacey, 2, 1-25, Knight NF, Je, SL_ MeCleary, SC Macy nd M.A. Aminpour 1985, Large Scale Staetaral Anais: The Sractal Anat, The CSM Testo! and The NAS Sytem NASA TM-100683,2 . n, n 5 1. n. ms. NF KNIGHT sight, NF, Je, .: Lots and RE ian, 1980, “Compattiona rca Mecha see Metbads Revco Using an volving Framework” AIAA Pape No. 50-1185, Fight NB, dex SC Macy and S1L- MoCleay, 1995, “Assssment of Stucual ‘Analysis Technology fr State Coliseo Este Cyl Shell” Fine Bleu lv Aaland Design, 18, 403-43. Kom apd. Fees, 1979, “Symbolic Generton of Fiske Element Sites Mattes” Computers and Siruxees 0. 119-128 KKrameeat, M1994 Rectangular Shell Fite Eemene with an Open Crack” Fae Elements in Ais and Design, 13), 238-253, Lee, Sango and. Belysco, 1995, “H-Adpve Methods fer Nenlincar Dynamic ‘Aye of Shel Stites.” Shock ad Vortion, 23). 198-208, MacNeal, R. 194, Pinte Elemons: Tir Design and Performance, Maret Dekkes, ‘Mathiss, Hand G. Strang, 1979, “The Solution of Nonioar ite Elemeat Ba tho internal Journal for Naerial Methods in Enginecring, 14, 1613-1826, ‘Meta, Rl 1963, "Basis or Derivation of Mares fr he Dect Shes Method.” ‘AIA Sournal. 17} 1631-1687, Also oseion Of paper in ATAA Journal, (2). 05 196 26), 1261, 1964 6), 1215-1216, 1905, ‘Noor A. 68" Metta of Amisie—Sats a Projections InN-Perone fini WD. Tiley edo, Srctral Mecho SifwareSerier~ Nolame I, p. 263 505, The Univeruty of Vigna Press Charbel Noor, Ak, 1992 "Machanis of Anisotropic Plates an Shalls — A New Look at an (O1 Subjct™ Computers and Sirctnes, 443), 99-514 Noor, and €M. Andersen, 199, "Compatrized Symbolic Maniatton in Stae- thal Mechanics — Proper and Poti" Comper: and Secures! 1, 95-118 Noor AK andCM. Andersen, 98, Computize Symbolic Manipal in Nonlinear Noor AK /CM. Aaderen and) Pte, 1979,"Glnbal Local Approsth es Nooinest ‘Shell Anyi" In Precings ofthe Seventh ASCE Canfrence on Erectonie Com- puetion, pp 64-657. Wang Univers St, Lovie, Misoa” = Noor ARM. Teer an’C-M Ales, 140, "Twn Stage Rayleigh-Ris Technique for Nonna Ansel Sarte" In Shaw, W. Pilkey, B. PAY: TOWson, A. Liki A. Chandsosr and G. Maras, ees, movie Monreal Anos for the Engineering Scencen, pp. 745-TS5. The Unies of Vezina Press, Chrisie Nove AK. and JM. Peters, 1980, "Reduced Basis Tene for Nonlinear Analysis Sct” AAA Journal 13), 485-462 Roce AK. and JM, Peter, 1990, “Penal of Mixed Foansations for Advance ‘Artis Systema” Conpters and Sercnres, 35), 369-380. Foor AK. and IM. Peters, 1981, "Difeation and Pos ling Analysis of Lami ated Compania Pics va Reduced Hass Techrige"In Hughes, Gating and Spies, ‘airs New Concepts Fite Element Anases pp. 07-131, ASME, AMD-YO. 4. Noor AKC, Belyehko and J. Simo irs, 1989, Anica and Computational Models of Sei: ASME. CED-Nol 3 [Nowe AK and WS. Buroa, 1989, “Astssinent of Shear Defonasion Theos for Manitayered Composite Piste.” Applied Mechanis Review, 420), -T2 ‘Nor, and WS Burton, 199, “Assesment of Compuaonal Medes for Malayer ‘Aniowopie Plates” Conpaders and Sacre i, 283-208, Peaaot, AN. std ST Denis, 1992, alin car Analysts of Shell Sacturs. ALAS Eeatonal Series, AIAA, Washington,DC. Fk, KC. and GM. Stay, 1966, "A Carved C Shell Element Based on Assume Natur Conrdine Sain” ASME Jour! of Applied Mecharis, 108, 278-290. Phan THAN, 1968, "Dervtion of Element Stes Matices by Assumed Sires Functions" IAA Journal, 2(), 1335-1886 Alo ension of pager in ATAA Jor SM) 186-187, 1965, 8 6, wr 95, on oe NONLINEAR POSTBUCKLING 93 Pian, THLE, 1975, “inte Bmore Methods by Variations Princip with Rane Conny Requirement" InC.A. Brebia and H. Toten eins, Variational Met ‘dein Eegnering, No.1 p. S/-S724. Southamon Univesity Pex Pian, EHH. and P Tong, 1989, "Bais f Fite Element Method fe Solid Comino” Insematonal Journal for Nunerical Methods in Engineering, 1 328, Poole, EL, 1851, “Comparing Direct an lsttve Equation Savers na Large Struc ‘url Anat Software System™ Comparing Systems i Engineering, 22, 397-08, Poole, EL. NF Knight rand DD. Davis Je, 1092. “igh Perormance Equation Solvets andiceImputon Fate Element Analysis" Itermational Journal for Name ol fehods i Enginecring, 343), 885-888 Ramm, B 1981, "Stags for Tescing the Nonlinear Response Neat Lit Points, In Woden, er, Notinear Fine Element Anais i Sacral Mechanics, PE 63-89, Springer Vee, Bern Ramm, 1982, "The Riks- Wenner Approach — A Extension of the Displacesnent Contd Method i» Nonlinear Analysis" In Winton, Oven and Taylor, edits, Recent ‘Advances in Neninear Campuaonal chanics pp 68-86 Pinendge Pes, Swansea Ranka, CC. PSeblin and FA. Brogan, 1986 Exoncements to the STAGS Computer Cote: NASA C000, Rankin. Cand A Droea, 1986, “An Element-Inependent Cortona Poche for the Treatment of Large Rotations” ASME owrnal of Prestre Verse! Tete 108, 165-174 ania, CC. and B, Nour Omid, 1988, “The Use of Projects 0 Imprive Fine Element Performance Conputrs ond Sructres, MMV), 257-267 Ranhis, CC anal FA. Broga, 1991, The Computational tecture Sachanie Test SiractralElentProcestor BSS: STACS Shell Element. NASA CR-88 Redd. ZN, 1989, “On Refed Compuutional Models of Conpaste Laminates” International Jal for Nunereal Methds in Enncrin. 27. 361-382 Reda 1, 199%, An hovoducan 10°the Fite Elonen Heth. Second Eton, Mia Book Company ne [Redy, Nand DH. Robbins, J, 1994, “Theos and Computational Models Yor [Composite Laminates” Applied Mechanics Reviews, AH), Pat 1, 147-169. Reise. B14, "Noton the Teeeem ofthe Syne fhe Stress Teor"Youra (of Maheoaies and Physct, 28, 192-194 engarsjin, GNF Knight ead M.A. Amiepour, 1985, “Companion of Symbolic and Numer oteeaton Methods foram Astmed Suess Hyd Sell Element CCommunicaions in Numero! Methods fe Engineering, 1, 307-316 Rengirajet.G. MA. Ansiopour al SF Katt Tr, 195, “proved Assumed Stes Hybrid Shei Eiemen’ with Dring Degree of Freedom fr Linear Suess, Bucking a Free Vibration Analyses" Interational Journal or Munrical Methods a Engine, 134 1917-1993 Ris, 1970, On the Mamet Sotuton of Shaping Problems the hear of Elastic Sibi. Stanord Univesiy, Repo SUDAAR No. 401 (ASO valle a AFOSR 70. 225878, Rts, E, 1987, “Propress in Collapse Analysis” ASME Jounal af Presive Vessel Techrlogy, 109, 33-1. its, F192, "On Fermolations of Pah flloing Tcigus fr Stata Stability ‘Alyse "InP Ladevere and OC. Zenbewice etre, New Advances in Comput tonal Siactaral Mechonis, pp, 65-79. Euevier Science Pushers, Amst, Ris, B, CC. Rankin and FA. Brogan, 1996, "On the Solution of Mode Tumping homens in Tin Walled Shell toca.” Computer Methods Aplid Mechanics land Engineering 0 app. Rogers, W. 1986, Report ofthe Presidential Commission on the Space Seale Chal Tener Accident, Wastrgion, DC.us us us, Ws, 120 NE KNIGHT Smith G4, 1986, “The Dangers of CAD. Mechanteal Engineering, Febaary, pp. 58 6 Stanley, GAM. and S. Now-Omid, 1990, The Computational Setural Mechavis Tate Genre Srctral Element Proceso Monae. NASA CR-181728. Staley, GM KC. Park and Hughes 1 1986, "Contam Based Resaan Shell lent" Ia TJ. Hughes and E- Hunton, estore, Fine Element Methods for Plate (and Shell Struct. Vole I: Element Tecnology, pp. 1-45. Pine Pres ater ‘aston, Swanen, UX. Sanky, GM. and CA. Felipe, 1986, “Compuaional Procedures for Pssbckling of ‘Cormpost Shells” In PG, Bergan... Bale snd W. Wupdetich editors, Proceedings [FS Euope Symi on Fite Bement Methods fr Noincar Problems, p35 $85, Universi of Tondheim, Nore. Stanley, GM, 191, The Canputtiona Sractural Mechnies Tested Sirctura Ele iment Processor ESL: Basie SRI and ANS Elements, NASA CR-4357 Stames: 1411, Knight Jc and M. Roos, 1985, Poxtuching Behavior of Selected Flat Rectang rapt Epoxy ancl Lodi Compression AIA A Jeurel, 258), 1236-1246, Shanes, HL Je and M. Rowse 1981,“Poshockting at Faire Characteristics of ‘Sclected Far Stfened Grape Epox) Plies Loaded im Compression.” ATAA Piper No, 81-0543. — Stein, apd. Brogan 1984 “Analysis of Strcral Cols by the Reduced Basis “Tsien Sobel and Thoms eis, Collapse Analysis of Siucures, pp 2-3, ASME PUP-Vol. 84, Sean B,¥e Wagneand. Whigs P18, “Conoeps of Modeling nd Discretion ‘or Elatc Shells for Nolineae Fite Element Anais.” In J. Wasa, eo, The “Mothemacir of Bite Elements and Appian VI MAFELAP 1987, pp 205-29. ‘cadene res Limited, London Stewart CB compile), 1980; Te Computational Siucural Mechanic Tested Us fs" Mtenal. NASA TNGIOD64S (pats). Stewart CL. (compile), 191, Fhe Computions Strcrra Mecha Testbed Pro telurt Maral, NASA FMIHOD6E6, Stastny A an WE, Haier 197% “Frsatsont snd Solution Procedures for [Nonlinear Stns Analy ™ Computersand Sacer, 7, 15-136 St Wand MS, Anderson, 1980, “PASCO: Structural Panel Anal and Sing Cole — Copabiiies and Anata Foundations." NASA TM. 3018 ‘Svoud, 1982, “Optimization of Compose Stacie" In ashi ana Herve, of Composite Materials — Recent Advnees bp. 207-321, (Alo sealable as NASA TMEBIS4). ‘Seoud WWM Greene and M.S. Anderson, 198, cing Load of Stifened Panels Subjected 19 Combined Longitana! Compression and Sher: Results Obtcined with PASCO, FAL and STAOS Compcer Programs. NASA TP-215, Srabo, Band. Babs, 191, Finite oman! Anais. John Wiley are Sons, Ic New York ‘Tan HiQs TX. Chang and D. Zheng, 199, “Om Sjntaie Manipulation and Coe Generation fs HybatTvce-Dimenstna Sli Bement Eninecring with Compt oA. {Tjlo, Ribs PI, Rersfnd sid EL, Wilson, 1976, “A Now-Conformiew Blame for Suess Atalsis” ntcratine Jornal for unorcal Methade in Engineering, 10), 176 1211-128. “ews, A. and THR, Maples, 1985, “A Three-Node Mindlin Pate Bement with Improved ‘Transverse Shear Canpuer Merhds In Applied Mechanics an Ensineer~ ing 8071-101 ms, 14 wos, bs vos, 130 ne NONLINEAR POSTBUCKLING 95 “Tharson, GA. 1980, “Newwon's Method Link Between Comtauous and Disc Solon of Nonlinear Frolons"Ia McComb and Noo eta Research Nonlinear Sirctral and Solid Mechanis, 9p. 2-28. NASA CP 2N7 “Thorson, GA, FA. Bropan and P.Siehiin, 1985, “Posbackling Analyst Using General Pupove Code" AVAA Journal, 246), 1013-1020. “Tileron, 1, 1975, "Selecting Solon Pocedaes For Nonlinear Stal Dyan Ex!" Shock aad Wvavion Digest, 78) 2-13 “Tong 1984, “A Hybrid ante Bement Meth for Damage Tolerance Analy: Computers and Soaces, YU), 288-20. ‘Wathsh, Ky 1968, Vertnona! Methade im Eloucy and Plestciy, Pergamon Yess, Wastes, K, 1972, “Saxe Considerations of Basic Tory forthe Finite Elemnt Meth In. Odes, RW. Clough and V. Yano, eis, Adionces i Compa fatonl Methods in Sirctra Mechanie and Design, 39-83. Unieraty of Alabina tn Homvile Pres Weingarten, VE-P Seide an 1. Peeson, 1968, Bucting of Thin-Wlled Chew Chiindere NASA SP-8D07, weve, Wenner GA, 1971, "Disrete Approximations Related to Nonlinear Theories of Sold ntertanal Jounal of Ste ad Stace, 7, 1581-1389 ‘Wena, 1092, “Shell: Mechanics ni Approximations — Linear and Noninse ‘Arpt In BY, Cheng and F Zab, eitors, Computational Mechanics in Sacra ‘Ensncring: Racont Development ad Futire Tends, pp 14-127 ever Sconce Pabishers, Ld Wilkos, EL. MLY. Yuan and JM. Dickens 1982, “Dynamic Analysis by Direct Suparpesiton of Ritz Vectors" Intemational Jownat for Earthquake Brgincering nd Shactural Dynes 1016) 813-821 “Lenkieviez, O.C., 1070, "The Finite Element Method rum Into to Gener.” ‘Applied mechancs Renews, BP. 249-250 “enkiewie,O.C, 1971 “lastenentl Displacement in Noieat Analysis tioned doural for Numerical Wess in Engineering. 3, 87-392. ‘oeniewier 0.6, 185, "The Generalized Fane Eemene Metiod Ste of he At ‘nd Pare Dison" ASME Jounal of Applied Mechanis, 80 1210-1247. “ekiewie, CLC and RE Tyler 1989, The Fue Element ft: Volume J Basie ‘Formulations and Linear Profan. Furs Eden. McGraw-Hil| Hook Compe Lond, Zeenkiewicr, O.C and RI. Tayo, 191, The Finite Ble Method Volume 2: Solid fend Fi Mechanis, Dynes ond Nonlinear, Fourth Edition, MsGra-Hil ok {Company London. cea.