Scribd Logo
Upload

Welcome to Scribd!

  • Finite Element Modeling For Stress Analysis - Robert D. Cook Compressed

    Uploaded by

    Toufique Yalsangikar
    131 views330 pages
    FEM by R. D. Cook

    Original Title

    Finite Element Modeling for Stress Analysis - Robert D. Cook Compressed

    Copyright

    © © All Rights Reserved

    Available Formats

    PDF or read online from Scribd

    Did you find this document useful?

    Is this content inappropriate?

    Report this Document
    FEM by R. D. Cook

    Copyright:

    © All Rights Reserved
    Download as PDF or read online from Scribd
    Flag for inappropriate content
    131 views330 pages

    Finite Element Modeling For Stress Analysis - Robert D. Cook Compressed

    Uploaded by

    Toufique Yalsangikar
    FEM by R. D. Cook

    Copyright:

    © All Rights Reserved
    Download as PDF or read online from Scribd
    Flag for inappropriate content
    of 330
    Finite Element Modeling for Stress Analysis Robert D. Cook University of Wisconsin - Madison JOHN WILEY & SONS, INC. New York Chichester + Brisbane + ‘Toronto - Singapore ACQUISITIONS EDFFOR Charity Robey MARKETING MANAGER Sesan Elbe PRODUCTION EDITOR Ken Senor ‘TEXT DESIGNER [Las Golds COVER DESIGNER cual Grote MANUFACTURING MANAGER Sasun Steer ILLUSTRATION Sigmucd Malinowski “Ths Book was set in Times Reman by Gepess] Graphic Servis and printed and bound by Hasina Printing, The coves was pied by Phoenix Color Corporation ecogning te importance of peserving whats heen wenn, itis a policy of John Wiley & Sons, Ine, have books oF enduring value published In dhe United States pints em ae ree pape, ad we exer ot bes feats 0 tat en Copyright © 1995, by Jo Wiley 8 Sons, He mats reserved, Pubilsad sirataneously in Cans, Reproduction os taralaon of any past of this ork beyond dat permed ty Sections GT and 108 af the 1976 United States Copyright AAs wtiout the poms af econ ner is ule, Rey ass foc pension ‘or frtes information shouldbe etressed to ‘he Permasons Departmen oe Wiley @ Sens, Ine brary of Congress Cottoging in Publication Daze: Cock Rober Dis Finite elemont modeling for sess analysis / Robert D. Cook, Fem Inlasesindes IsaN 0.471 107743 [Fie clemest nettod-—Data pressing, 2. Satara analysis (Enginzesngh Dats rocessing. (Tite, TAIT FSCEES 1964 uaa) S417 —te2o CP Printed in the United Ststes of America log 87654 Preface ‘This book is intended for beginning courses in finite elements CFE) that are oriented to watd users of the method. The courses envisioned emphasize the Behavior of FE and i= clude computstions! work in which problems are solved by means of comamercial sof ware and the computed resis are critically examined. The instructor may often sit with students atthe computer to offer advice and to monitor their skil in modeling and assess- ment of results. The courses would use computational problems as vehieles 10 teach proper use of FE, rather than use FE as a way to solve certain problems, The book pre- sents a modest amount of theory, discusses the nature of FE scluioas, offers modeling advice, suggests computational problems, and emphasizes the need for checking the corn. puted results. Problem areas weated are comunon in mechanical engineering and related disciplines. Suggested computational problems include topies often ereatec in a second course in stress analysis, such as spinning disks and elastic foundations. Toe compute- tional problems usually have simple geometry, so that BE may be emphasized rather than detsils of data preparation, Some instructors especially those who leach move advanced seudents, may wish to devise problems of a more “res! world” nature, despite their greater complexity Several commercial FE programs are available for use on microcomputers and work stations, This book is not milored to any particular FE program and therefore does not {In either form, Eqs. 1.2-1 or 1.2.3, the displucernent field a = u(x, y) and v = w(e, ») has six degrees of freedom, abbreviated do. That is, six quantities define the deformed con- ‘Gguration. namely. the six §, in Eqs. 1.2-1 or the three w, and three v, in Eqs. 1.2-3. In ‘Chapter 3 we will explain chat strains are displacement gradients. Therefore a ox « *% sere = Bs ‘ g, e 8, = By O24) au 2 =D hence 1, =Bs+Ps My ae te = Bs +B ‘This three-node element is called a “constant strain triangle” because none of the strains varies over the eletent. This means that the element has a very limited response: ‘could not represent the linear stain field of pare bending, for exampie—bul at lesst there will be no strain concentrations at nodes. Also, from Eqs. £.2.3 we ean conclude that efe- «ment sides will remain straight afer deformation. For example, set = 0 to examine side [in Hig, 12-1: thus a becomes linea in y and depends oniy om do. and a The same will be tise along this side in the adjacent element. Because deformed sides remain straight. elements will not gap apert or overlap when Inod 5 appbed. Similarly, we can show that ¢ along side 1-3 is fincer in.y and depends only on o and vs, whether we ex- amine the clement on the left or the element on the right of side 1-3, Summing wp, itis possible to demonstrate that the tangelar element can cisplay constant strain states and will dsfoom in a way hat is eonspaite with is neighbors, The saute va be detail for otier shapes and types of element. It ean be shown thet these properties allow exact results t0 be approached a a mesh is refined: that is, as more and more elements are used to model a structure Let us also consider briefly a sin-node tiangle, sucit as element L somewinss above E in Fig, 1214, It has three vertex nodes and three midside nodes. In terms of generalized ‘coordinates fits displacement field is + Bax + Bayt Bee + Baxy = Bay* 25) 0 By + Bee Bay + Bao + Bassy + Bay? Deformed shapes of sides can be straight oF perubolic. Some tedious algebra shows that +he deformed shape of » side depends on doz. of nodes attached to Unt side but does nat depend on dor. of nodes not attached wo that side. Accordingly, the element will be com- patible with its nefshbors because adjacent elements share the same uodes and do. along a common side, By applying the differentiation used in Eqs. 1.2-¢, we see thatthe sixstode element contains constant and linear terms in sts srs field. Therefore this ele- ment cop model constant strain states and alsp Tncar strain states thar arise in pure bend ing, Clearly, its a more competent element tha the constant strain triangle. It is also more compliceted, which suggests anoter choice fazed by the user of FE: Is it biter to 6 Inavoduerion use many simple elements or a few complicated elements? We postpone this maticr, as te answer is neither short nor simple. The foregoing discussion ig couched in steess analysis terminology. In plane sires analysis the displacement field is a vector flelé because it has two components, «= u(x, y} and b = v( 33. In othey applications the field may be a scalar field, = 9(s, 3) in two-di> mensional problemas and ¢ = @(, », 2} in three-dimensional problems, where 9 represents temperature in a heat conduction problem, voltage in an electtie flefd problem, and so oa, To restate tie foregoing equations in scalar ficld terns. one smay ciscerd equations that contain v and replace u by gin equations that remain, Equations such es Eqs. 1.21 and 1.2-$ constitute the “basis” of « finite element. What remains is t0 manipulate the basis to generate 2 “stiffness rastcix” that desoribes element behavior, connect elements (ogether 10 produce the FE model, apply’ loads, impose sup- ‘port conditions, solve for nodal dod. and use the of, 10 compute strains and finally stresses, Some of these procedures are primarily computational and others require that the analyst make decisions, Subsequem chaplers contain a more complete discussion of these mater. Classification of Stress Analysis Problems. Elements summatized above are used for plane problems, in which there is negligible variation of displacement znd stress in the z direction, thet is, in the direction normal to the analysis plane. If displacements and siresses may vary in general wuy with all Uiree coordinates, the object may be called « 3D solid, The special case of solid having axisl symmunetry (ike a bell) is usually called a solid of revolution. Loads may ot may not be axially symmetsi. A flat plate that cazries in-plane loads is 1 plane problem, but ifthe plate is loaded Jaterally so that i bends xt is called a plare bending problem ot simply a plate problem. Floor slabs and highvray slabs are examples of plates. Note that thickness mast be much less than span if tie object is to bbe analyzed us a plate, Ifa plate is curved it becomes a sell. Water tanks and come pressed air tanks are commenly seen shells, Shells can carry both in-plane loads and lat eral loads; thus plane deformation and bending deformation usually appear simaultane- ously in-a shell, Elements have Been devised for all these problems. ‘Thus there are plane elements, generat solid elements, axisymmetric solid elements, plate elements, and shell clements. In uddition, thore are elements for bass and beams ard many specialty elements for elastic foundations, crack tips, pipe bends, and more 1.3 MODELING THE PROBLEM AND CHECKING RESULTS Modeling is the simulation of a physical structure or physical process by means of = sub- stiute analytical or numerical construct. It is not simply preparing a mesh of aodes and aJements, Modeling requires that the physical action of the problem be understood weil enough (0 choose suitable kinds of elemenls, a! enough of them, to represcat the physi= cal action adequately. We want to avoid badly shaped elements and elements (00 large to represent important variations of the field quantity. At the other extreme we want 10 avoid the waste of snalyst time and computer resources associated with overrefinement, that is, using many more clements than necded to adequately represent the field and ies gradients. Later. when the computer has done the catculations, we must check the results to see if they are reasonable, Checking is very important because ié is easy to make mis- takes in describing the problem to the software. The following discussion is a brief sur- vey of these matters. Further discussion appears in subsequent chapters. 1.3 Mordeting the Problem ant Checking Resuls 7 Support conditions are very important but gre often misrepresented. Consider the problem of Fig. {2-1 again. Support along AB is portrayed as rigid. meaning that nodes along AE are not allowed to move at ali. This ts probably unrealistic. No suppor is infi- nitely stiff. t would be better to enlarge the FE model so that thexe aze finite elements be- low AE to cepreseot the elasticity of the foundation. Hosever, perhaps the futent is to an- alyze = G-shaped part that has A£ us an axis of symmetry, and reduce effort by modeling conly the upper half. Supports suited to this situation appear in Fig. 1.3-La, These supports are placed at all nodes along AE. Node A is fixed and other nodes along AE aiv allowed to move in only the x direction, ‘Thus we prevent rigid-body metion in the xy plane snd keep AB a straight line as syraetry requires, The mixture of clement types in Fig. 1.2-1 is unusual, but otherwise is the mesh layout good? We cannot say for sure without knowing more about how elements Behave, However, by anticipating the results we can see that the mesh grading locks reasonable. ‘Suresses near will be low and of little imtcrest, Indeed, theory says that siresses ar B are ‘zero becuse (C48 a point where two free surfaces intersect al an interior angle of Jess than 180°, Accordingly, 5 coarse mesh near & is acceptable: stresses near J may have a large percentage ertor but this does not muller if stresses near B are small. The same is rae near D, so pethaps the mesh near D is more detailed than necessary, At C the stresses are theoretically infinite because of the concentrated load P. In reality, one cannot apply @ load that is toly concentrated at a point, Probably load is a convenient way of repre senting @ load chat is actually distributed over a small span, and stresses near C are not the object of study, se the modeling near C is acceplable, Stresses near E ate probably the stresses of concern, Tere the stresses and stress graiionts ure expected 10 be large, 0 ene ‘model properly displays a finer mesh andor more competent elements in this ara. ‘Che FE method calculates nodal displacements, then (in present software) uses the dis placement information to calculate strains and finally siresses. if displacements are incor rect, siresses will probably be incomect, Accordingly, we should examine the computed displacements first. Without calculation, we anticipate tht the displaced shape of our ex: ample structure will be as shown in Fig. 1.3-2b. If the computed result is substantially ifferent from this we cuspect an ersor in our model The software will permit us to ds play the displaced shape supezposed on the original shape, with displacements scale up so that they ate easily visible. Additiocally, we can animate the displaced shape, so that the mode! appears (0 be vibrating slowly between its deformed and undeformed posi- Hh (tee nt shown) kK» +} el o Co Fig. 13-1, (a) Aliemative support eoncitions at andes along AE of the structure of Fig. 1.21, (6) Dashed lines show the anticipated detonation, greatly exayerated. 8 Introduction tions. Thus we can easily see, for example, if nodes along AB move in only the x direc: tion as was intended. [As for stresses, the software will plot hem either #9 contour ines or bands of different cofors. A stress contour Tine connects points that have the same stress. Users may elect an ‘option in the software that calls for averaging of stresses. This mcans that stresses from individual elements are averaged at odes before plotting, so that sizess coatours have 90 discontinuities between elements, This is pacr practice because it removes information asefol tothe analyst. As we will see, unaveraged stresses ae usually discomtinnous across inierelement boundaries. A contour plot that displays significant interelément discontinu- ities wamns that a finer mesh is needed. This point is mede in Fig. 13-2. In Fig, 1.3-2c, nothing betrays a luck of perfection but small changes of direction where contoue lines cross inteelement boundaries. ‘The separate stress contour plots (one for, one for gy, et.) are examined ia corn, Based on experience, physical intuition, and knowledge of theory (including statics, me- cinanies of materials, and possibly more, itis possible to deseibe the expected stresses qualitedvely. For the problem of Fig, 13-1 we expect the following (his is not an ex- hhaustive lst): + 9, is large compressive stress near E + 4, is tensile near A but smaller in magnitues than , near E + gis compressive bul small in magnitude between A and E + G, and 1, ate very small slong AB because of the free surface condition (computed stresses Will not be exacily zero because the solution is approximate) Significant departures trom these expectations warn of trouble with the model or shont- ‘comings in physical understanding of the problem. Discrepancies musi be corrected or Ingicalty explained before the results ean be tusied. ‘The analyst should also obisin analytical or experinnental results for cormparison with FE results. For the problem of Fig. 1.3+1 this task is easy. Cross section A 3s loaded! by ‘ireet force and by bending. The elementary formula for stress in straight beams should provide a fair approximation, the fermla for stess in curved beams should provide a 000 approximation, and tabulated results are available [1.5]. Indeed, FE analysis is prob- ably not needed for this problem. For many problems, approximate solutions cam be b= tained from tabulated formulas in standard! textbooks and handbooks [IS]. Much of this, information is available as software, which makes it far easier to use. However, if this ~ £0 Ne “si L 40 aa 0 20 20 fl o to Fig, 13-2. Three hypathetcel sets of suess contewrs azar a node shared by four elements (2) Without not averaging: imperfect but adequate continuity. (b) Without nodal everasing: Iatiequate continuity. (¢) After nodal averaging: continuity, but difficult to say whether the Faw data were good of bad 14 iscretication and Other Approximations 9 , Apacs | (fanapee anes |! t tee | S mee Larrea] | Pian how | : requis? jee using incigne || i ss amyl. || seadtetr | | : as Festa | | | (ae ariel ] : (oes made revision do ite | Gemputr yee ff Fig. 13-3, Oudine of a FG analysis project phase of verification is done after doing FE analysis, there will be a tendency, perhaps lunconseious, to obtain analytical results that aerce with FE results alzeady obtained. We tond to find what we expect, whether itis there or not. Thereftre some approximete re sults showld be in hand before undertaking FE analysis, Figure 13-3 summiriges the pro codure for FE analysis that is advocated in this book. ‘Organized andl careful work will take less total time tha a hurried approach that pro- duces and propagates carors that must be discovercd later and corrected, Festina lente. DISCRETIZATION AND OTHER APPROXIMATIONS Whatever the analysis method, we do not analyze the actual physical problems rather, we analyze a mathematical model of it, Thus we introduce modeling error. For example, in elementary beam theory we represent a beam by a Line Gls axis) und typically ignore de- Formation’ ussociated with transverse shcar. This is an excellent approximation fur slew er beams bur not for very short beams. Or, for the axial-ioad problem of Fig. 1.4-Ja, we would probably assume thet a state of uniaxial stess prevails throughout the bar, which is proper. f taper is slight but improper if taper is pronounced. Real structures are-not so easily classified, as they are often built of parts that would be idealized mathematically in ifferent ways and have cutouts, stiiTeners, and connectors whose behavior is uncertain. ‘The foregoing considerations must be addressed in order to decide what types of ele ‘ments fo use and how many of them. Ifa beama is deep, transverse shear deformation may became important and shoufd be inciuded in beara elements. Ifa beam is very deep, twor or three-dimensional elements are more appropriate than beam elements. If@ beam has a % ee a ‘a a Tig. 141 (0) A tpeed bar loaded by axil force P. (8) Discretion of the bar into fur uniform avo ode element fel length, 10 Insreduction wide cross section, plate theory may be more appropriate than beam theory (ben, of course, choose plate elements rather thon beam elements). If an axisymmetzic pressure vessel has a thick wall one sbould regard il a8 a sofid of sevelution rather than a shell of revolution and choose axisymmetric solid clements rather then axisyrumetsic shell el- rents, Let us consider the axially losded tapered bar of Fig, 1.4-1a in more detail and de seribe how the FE method implements the mathematical model. We will assume that 4 satisfactory mathematical model is based on a state of uniaxial stress, An analytical solution is then rather easy, bat we pretend not to know i ant ask for a FE solution instead, We discretize the mathematical model by dividing it into two-node elements of constant cross section, 2s shown in Fig. 1.4-1b, Bach element has length L, x conats oaly for a constont uniaxial stress along its Yength, and has an axial deforma ‘ion given by the elementary formula PL/AE. For each clement, A may be taken as constant and equal t6 the cross-sectional area of the tapered bar at an x coordinate comtesponding to the element center. The displacement of load P is equ to the sum of the element deformations. Intaitively, we expect that the exact displacement is ap- proached as more and more elements are. used to span the total length Ly. Hawever, even if a great many efements are used there is an error, known a5 diseretization er- ror, which exists because the physical structure and the mathematical model cach have infinitely many dof. (namely, the displacements of infinitely many points) while the FE model has a finite number of d.of. (the axial displacements of its nodes), How many elements ere enough? Imagine that we carry ous two FE analyses, the sec fond time using a more refined mesh. The second FE model will have tess discretization error than the first, and will also represent the geometry beiter if the physical abject has carved surfaces. Ifthe two analyses yield simitar sohitions, we suspect that results are not inch in error. Or, we might establish 2 sequence of solutions by solving the problem more than twice, using 2 finer mesh each time. By study of how the sequence converges ‘we may be able to state with some confidence that results from the finest mesh are in er sor by less than (say) 5%, ‘Afler the analyst bas introduced modeling error and disczetization error. the computer introduces numerical error by rounding or truncating numbers as # builds matrices and solves equations. Usually numerical error is small, but some modeling practices can realy increase i Finally, it must be ushnitted that the software almost certainly contains errors (5.6). Commercial software packages are large, versatile, and under continual revision, It is precticelty impossible 10 ger everything right, Many exrors either make # software feature inoperable or eause the program to crash, but some can lead %0 erroneous results. It is empting to blame ail strange results on the software, but iis far more offen the case that ‘uc have blundesed in mosieliog or in describing the mode) to the software, Strange results, are obiained so often that (to repeat) it is vital thatthe analyst be able to recognize that re sills are strange. 4.5 RESPONSIBILITY OF THE USER FE compater programs have become widely available, exsier 0 use, and can display results with attractive graphics. Even an inept user can produce seme kind of answer. Tt is hard to dishelieve PE results because of the effort needed to get them and the 1.5 Responsibility ofthe User n polish of their presentation. But smooth and colorful stress contours ean be produced by any mode), good or bad. It is possible that most FE analyses are so Flawed that they cannot be tcusted, Even a poor mesh, inappropriate element types. incorrect ods, or improper supports may produce resulis that appear reasonable on casual in spection. A poor model may have defects that are not removed by refinement of the mesh. ‘A responsible user nest understand the physienl nature of the problem and the behay- ior of finite elements well enough to prepare a suitable model and evaluate the quality of {he sesulls. Competence in using FE for stress analysis does not imply cermpetenee in us ing FE for (say) magnet field problems. Responsibiligy for results produced is taken by the engineer who uses the software, not the software vendor, even if results are affected by errots in the software Figure 1.5-] is an example of discrepancies that may appear [1.6]. A pressure pulse is applied to a straight beam with hinge supporis. The loading causes the meterial 10 yield and the beam to vibrate, Analysis seeks 6 track the lateral displacement of the midpoint as « function of time. The results plosted come from tea repuizble analysis codes and were obrained by users regarded as expert. Yer if any of the curves is eomect we cannot tell which one its. Admittedly, the problem is difficult. The results indicate “strong sen- sitivities of both physical and computational nature" [1.6]. This example reminds ws thet any analysis program is based on theory and approximation, and that a user nay push the program beyond its range of validity [1.7], 200 sm —————>4 ~elzomm ie | fryer. EEEEPEERT ARSE: pole 0 t 2 2 = 3 5 7 Times Fig, 15-1. Lateral enigpnint displacement versus sie for « beam loaded by a pressure pulse [1.6)"The material is elasticeperfoctly plbstic. Plots were generated by vacies users and vari- ous codes. 12 Introduction 1.6 ELEMENTARY MATRIX ALGEBRA (ne need not understand matrix algebra in order to use PE software. However, our expla: nations of algorithms and of how elements behave are conveniently stated in mateix for- oat, The following matrix theory is used in this book. ‘A matrix coniains numbers, ané/or symbols that represent numbers, arrayed in rows ‘and columns. A matrix may be denoted by holdfuce ype, A, ot by use of brackets, [A] fy ao [fs 42 Ax a6) far Ae Matrix A has n rows and m columns, where m und m are positive imegers of any magn tude. If m = 1, A is called square, and 1 is its order, Coefficients with like subscripts Guus Haas £163} lie 09 the diagonal of a square matsix, I'm = 1, A is « row mani (also called £ row vector) if n= 1, A Js a coharm mani (zlso calied a column vector), Braces ae offen used f0 indicate a column vector: for exemple, {A} means that A bas only one column, Ifm = n-21, A isthe scalar A =A. Ittwo matrices A and B have the same m and the same n, they may be added or sub- inocted term by term: for example, in C= A+B, Cy = Ay + Bi. A scalar taultiplier of A acts on every term of A; for example, 2A contains AA,,, AA, and $0 on. The Sncegrel {or derivative) of « matrix with respect to a seslar parameter, such as sme. is a matrix Ghat contains the integral (or derivative) of every term “The transpose of A is A but with rows and columns interchanged, Thus Ay Ay Aw As 4n An Ag An Ass 1.6.2) IFAT =A, the mutex is called symmetric. and Ay = Ap), Ays #Agy, and so on. A symmet Fic matrix must be square (mt =». The product of two matrices is ARSE, where = Diab, (6-3) for example, Pix = 4a)B15 + AngBz5 + AzgByst. For multiplication, A and B must be con formoble: that is, if A has nt colurmns then, B must have re rows, Aa example of maltipli- cation is rae 7) a4 9] nes valle 5 isle sa] se Jn peneral, AB ¥ BA. If B is square and symmevic, so is the product P= AYBA, IEA isa ecoluma vector, then ATBA is a scalar. The transpose of a product is the product of the transposes in reverse onder; that is, if P= AB, then P? = BTA’, Analstical Problems 3 A unit matrix, or identity mateix, is denoted by I Ikis a dlagonal matrix of U's: that is, oso [2 bo? (165) ood ‘The inverse of square matrix A is denoted by A“! where A“ is constueted in such a way that ATTA =f is also irs that AA = I In this book we need to know what A“? means but not how to consinuct A“! from A. ‘The inverse of a raduct isthe preluct ofthe inverses in reverse order, that is, P= AB, thon P-! = Br!At A set of simultaneous linen algebraic equations may be symbolized as Kb=R 0.6.6) where K is a square matrix of kaown constants, Ris a column vector of known constants, and D is a column vector of unknowns. Solution for D may be symbolized as D=K'R en In FE work. K is a “stiffness” matrix that is usually large and sparse. Ik wou be wasteful of storage and time to invert it. Thus D = KR usually means “solve for the unknowns,” probably by some efficient form of Gauss elimination or perhaps by an iterative method Solving Eq. 1.6-6 for D is @ major part of FE calculations, bur usually the user need not know how the software goes about it, A square matrix is called singular i ts determinaat is zeo. °K in Eq. 1.66 is sing Jar, there is a0 wnique solution vector D, and staddaidl equation-solving subroutines will fil, As examples, the following mirices ate singuter, Le pi yop 6s oo) tral lee (88) Let K be an m by m matrix and D ann by I column vector. Also let D = 0, wich means that at lees one coeticien Ds nomvera, Then, focal D, ifD*KD> 0, Kis called positive definite (1.683) ifD°KD20, Kis callod positive semidefinite (1.6-90) A positive defiuite matrix is nonsingular. In stress analysis, a stiffness matrix K is posi tive semidefinite (and singutas) if supports of the FE smictare do not prevent all possible ‘igid-body motions. ANALYTICAL PROBLEMS LI (a) Show that Eqs, 1.2-3 follow from Eqs. 12-1 and 1.2-2, {b) Differentiate Kgs. 1.2-3 to obtain expressions for strains in terms of nodal dis- placements, 4 Ineveduesion 4.2 Show that sides 1-2 and 2-3 of the triangular clement in Fig, 1,2-1 remain straight a5 the element is deformed. 1.3 Bquations 1.2-1 may be applied to each of the elements shown, and eoniitions analo- gous to Eqs. 1.2-2 used to express displacements w= u(x, y) and v = u(x, 9) ia terms of nodal dof. u, and v,. Carry out chese operations. (AS a panial check, note that the resulting expressions must yield u = ujand v =u, when x= 2, and y= y,, where fis 1, Dordd tai Problem 1.3 LA Show that side 1-3 of each of the elements in Problem 1.3 remains steaight when the cleaneat is deformed. 1.3 (a) In terms ofthe 6 sn and y. evelumte the strains 6, 6, apd fa associated with the displecement Feld of a sit-node tangle, Eq. 12. 5) Pure moment loading is applied tow cuntlever beams built of these elements, as shown, Exact values of computed stresses are desired, if possible. Why are roller supports piace at A and at B rather than pin supports 8s shown al C? (6) Sketch an alverative arrangement of supports the left end of the beam that would work jst as well. {€) Which ofthe fj in Eq, 1.25 willbe zero for this particular cantilever beam prob lem? Which of the other fi are related to one another, and haw? Consider the stains calculated i part (3) to answer these questions “ * Problem 1.5 1.6. Blaborate on she list of siress predictions ia the latter part of Section 1.3. Gor exam= ple, in Fig. 1.3-1a, where do you expect thal stresses 6, Gy OF Z, should approach zero? Also, where do you expect that Hese stresses may have large magnitudes, and of what algebraic signs? AT Let dimensions ia Fig, |.3-ta ho g= 16 men, #2 18 em, € = 28 ma, and += 6 nun. [Also jet the thickness be 1 and the elastic modulus be , Place roller supports slong AE, as in Fig, 1.3-1. Use mechanics of materials anaiysis to estimate the fallowing in terms of E. f, and load P. fa) Stress 6 at 4 and at E. (Use soralght beam theory, (b) Stress 9, at A and at Z, Use curved beam theory: Computational Problems 18 (©) The 6, of largest magnitude along AZ (G) The vertical displacement component of load P. 1.8 A bar element used for the FE mode! in Fig. 14-1h has twe do.t, namely, the axi- ally directed displacement at cach end. Express the elerment displacement fd in the form of Eq. 1.2-4 and in the form of Eq, 12-3. Let x = 0 at the left end of the ele- sent. 19 Assume thatthe bar in Fig. 14-12 has 2 rectangular eross section, Let A = 34 and Tet thickness + {poapendicular to the paper) be constant. Evaluate the following in teas of P, yA, and (a) The exact displacement of load P. {b) The displacement of P using one, to, and then four wniform elements of equal length. What s the percentage error in exch case? 46) On aser of ates showing x fabscisss) and o,/Pihy) (ordinate), plo te exact ax- ial stress and the axel sitess prediction of the four-element model. By approxi= ately what factor are stress errors redced each time the number of clemens is doubled? COMPUTATIONAL PROBLEMS No specific computational problems are suggested in this chapter, However, students ‘may wish to get acquainted with the FE software chosen for the remainder of the course bby using it €0 solve simple bar and beam problemas for which tabulated solutions are reac sly availabe. CHAPTER 2 Bars and Beams. Linear Static Analysis Stiffness matcioes are developed for bar eemenss and beum elemcn’s. The physical meaning ofthese maces is explained Also explained is how loads ae tested, bat support conditions ate appropriate fora striae bil of bar cr beam element: and how the formulation ylds placements ar steses. Finally, an example uppleation ses how beam efemens may be used practice 2.1 INTRODUCTION Static snalysis omits time a8 an independent variable and is appropriate if deRections ane constant or vary only slowly. A structure forced to vibrate at a frequency less that about ‘one-third of its lowest natural frequency is a case in point. Such “quasistatic” problems may include steady inertia loads, such as those Gue 2 spinning about an axis at constant speed, Linear static analysis excludes plastic action and deflections large enough Wo change the way leads are applied or resisted. Thus cloimonts that fail, lacze yutations, asl gaps that open or close are exclucied After doing an approximate preliminary analysis, ptenaing how to de the computa tional analysis, and pechaps sketching an initial FE model, the analyst tums to softwace. FE analysis requires that the folowing steps be taken: 1. Prepare the FE model, The analyst musi . diveretize the stractuze or continuum by dividing it into finite elements, 1 prescribe how the structure is tended, and 6. preseribe how the structere is supported, 2. Perform the calculations. The software must 2, gongrate the stiffness mairix k of eack element, bb. comneei etements together, that is, assemble the element ke matrices tw obtain te structare or "lobai” matrix K, assemble loads into a globel load vector R, impose support conditions, and . solve the global equations KD = R for the vector D of unknowns. In structural problems D contains displacement components of the nodes, 3. Postprocess the information contained in, D. In stress analysis this meums compute strains and stresses, 18 Bars ond Beams. Hinear Storie Analysis Step Ja requires that the analyst exercise judgment about what types of element to use and how costse or refined the mesh should be in different regions of the mode}. Steps 1 and Je are often more straightforward than step La but iis easy to be inattentive and do ‘them improperiy. ‘The work in stop 1 is greally assisted by the preprocessor postion of the software. Nevertheless, this phase of the analysis will probably take considerable. time. Step 2 is carried out automaticatly by the software, Similarly, step 3 is automatic, al- ‘though the analyst must instruct the program as to which resulis to present and the format oftheir presentation. The displaced shape and various stress contours are usnally plotted. Except for discretization and the plotting of stress contours, the foregoing FE proce- ure is also applied to the numerical analysis of trusses and frames. These structures are inherently diseretized, in the sense that their members are already separate elemests. In ‘our terminology, truss elements are hinged at connection points and resist omy axial force; frame elements are welded together at connection points and resist axial and trans- verse forees and bending moments. All these members can be regarded as special cases ‘of what we will call a 3D beam element, which resists axial force, transverse shear force jn each of two directions, bending about each principal axis of the cross section, and torque about the longitudinal axis of the memes. The response of the member to these Yoads can be formalated exactly, or at least quite accurately, using only the tools of me- chanics of materials. In this chapter we examine beam elements, beginning with plane bar and plane beam elements as special cases, in order to explain the natare of « stiffness me- tix, how loads and supports are treated, and how stresses art extracted Stom displace- ments ‘A crude initial model of a complicated structure is sometimes built of bars and beams because the effort is comparatively small and information useful in subsequent FE model- ing may eppesr. 2.2 STIFFNESS MATRIX FORMULATION: BAR ELEMENT Direct Method, Consider 2 unifonn prismatic elastic er of length L, Fig. 2.2-), with elastic modulus £ and cross-sectional area A. A node is loesied at each end. For now we allow only axially directed displacements. We éisplace fist one node and then the other and, in each case, calculate forces that must be spplied to nedes in order to maintain the O while = 0, = Og = O, and so on. Thos there are four casas. In each case, show that nodal loads in Fig 25-26 are work-equivalent. 2.12 Disiibuted lateral force q and the cantilever beam ere both uniform (see sketch) Compate the ip defection and zoot bending rnoment esing work-eqaivalent nodal loading (Fig. 25-28). Then repeat the calculation, ibis time with lumped loading (Ge, omit the nodal moment portion ofthe vork-equivalent loading). Compate per~ centage errors in exch case. Suggestion: deflections of the FE mode! can be ob- {sine by use of standerd bear formas. (=) Usc a singte element (©) Use two elements, exch of tenth Laf2 2.43 Te bar shown is confined benseen rig walls. Cross-sectonal aren A vasies Hin- carly from Ay 10 1.64y, The bac is nitally stess-ive, then is oniformly heated st ‘umount AT. Compute bresses in @ BH model that contsins three elements, cach of Tength £3 and having the respective enss-sectional aes 1.14. 134q. and LSA Oo aes x (abscissa) anc a AT (ordinalc), plot the exact stress fielé and the FE stress field. : ‘ % 2640 (ErEreeay} { ee Proiem 212 Problem: 2.13 COMPUTATIONAL PROBLEMS, In the following problems compute peak values of displacement and stress er bending moment. Exploit symmetry if possible. When mesh refinement is used, estimate the max imum percentage error of FE results in the finest rnesh. Where materiel properties are needed lut not stated, use those of steel ‘A PE analysis should be preceded by an alternative analysis, probably based on statics and mechanics of materisis, and oversimplified if necessary. If these results end FE re sults have substantial disagreement we are wamed of trouble somewhere. 2.14 (a) Does the software you use include Gransverse shesr deformation in its beam cle- nents? Disregard the software documentation; instead, devise and run test cases 4 find out directly. (b) Similarly, use sest cases to discover if nodal moments due (o distributed load, are applied to beam elements. (These moments may appear in R and/or in stress comppiation)) Compunarienal Probiems 29 2s 216 247 18 29 ‘Modify the problem of Fig, 2.7-1 by changing the following dimensions 1 the ¥al- tues nicated: (2) ¢=d=9, others asin Fig. 27-1. () a= d= 10 om, others as in Big. 2.71. (6) b= c=D, others as in Fig. 2.71; also add a central vertical member of cross seotion by “The veaon shown is uniformly tapered in width. Let Poisson's sudo be zare. Create FE models by using clements of eonsiant cross sestioa, in che manner of Fig. 14-1, but with beam: elements, Use one, then two, then Foor, and soon, elements of equal ‘eng Choose convenient numbers for length L and force P. The problem may be repeated using « tip moment rather thun a tp free, or using a distributed load “Members of the plane structure shown may be bars pinned together a joints to ere- ‘ae u tuss or beams rigidly connected iogetier et joints to create a frame. For the frame mode one may assume that rotations at de wall are either permitted or pro- hibited. Investigate how much difference dhere fs between the cuss model and the frame moset, Assume that all members have @ square cross section. b units on a side. Let H/= 120 mm, £.= 160 un, aad P= 1.0 N, Consider the cases 6 = 3 oun, 19 mom, and b= 30 mm. Teo ew | fo a he Problem 2.16 Problem 217 ‘The bown structure shown has unit thickness normal to the paper. Depth fof the ‘cross scotion vaties linearly in the axial direction. Confine displacements to the pplane of the paper, ie yee! rote 248 Problem 2.18 ean be repeated with one end hinged. Consider the problem of a beam on an elastic foundation. The foondation can be modeled by bar elements that connect nodes of the beam 10 a fixed support and act as linear springs of stiffness k = AE/L (see skereh), This is not she best foundation amiodel, bat it is instructive to see how displacements und bending moments in the bbeam converge toward exact results us the mesh is refined. Analytical solutions for 0 Bars and Beams. Linear Static Analysis vVacious kinds of loading and suppost on s uniform beam are available (1.5, 2.1]. Beams of finite length or with step changes in cross section may also be considered. Typical bor lane Problem 2.20 2.24 The structure shown has unit thickness normal to the paper. Depth i of the cross section varies inesly in the ciroumferendal direction. Coafine displacements tthe plane of the paper. Some possible choices of geometry are as follows: 0%, R= SOG ram, y= hy=30 mm. 49°, R= 500mm, Hy : 90, R=S00muo, hy =h,= 30mm, R=S00mm. y= 30am, y= 10mm [I 1N (unt 08a) Problem 2.21 2.22 Problem 2.21 can be repeated with one or both ends hinged, Also, uniform or aacnunsform heating cxm he upplied. Yet another set of problems is generated By pre- soribing nonzero values of translational andor notaiond] df, one end, 2.23 Use the geometry shown for Problem 2.21, bu orient load P so that i acts normal to the plane of the paper, an et represent the diameter of a cireular cross seciton. Thus the structure becomes a belcony beam, which hos both bending and twisting deformations. The specifi con‘igurations in Problem 2.21 can be analyzed. Or, pie scribed nonzera d.o.f. may now inchude twist 2.24 Idealize a bicycle wheel as & planar structure having 36 radial spokes. Properties are 25 follows (2.3), Spokes: diameter = 2.1 mm, E= 210 GPa, lengt = 309.4 mm from the center of the wheel to the cemtoidal axis of the rim eross section. Rim: 1384 mn, £ = 70 GPa, v = 0.33, centraidal J of A = 1469 mm*, Ye = (76 mm* or stress celcularion), Assume that initial tension ja dhe spokes is sufficient to maintain tension in every spoke when load is anplied, Consider the following load- ings (@} A vertical force of 490 N applied by the road. () A force of 100 N applied tangentially by caliper brakes at the top of the whee! Neglect the mass ofthe wheel CHAPTER 3 Plane Problems Necessary preliminacies from solid mechanics theory are reviewed. Next, plane elements of several types are discussed. with paticolar attention to element displacement fields and what thay portend for element behavior, Trestment of loads and calculation of stresses are discussed, An example application clases the chapter. 3.1 INTRODUCTION ‘Stress-Strain~Temperature Relations. By definition, a plane body is flat and of con- stant thickness. Let Cartesian coordinates 2¥ lie in the plane of the body. As explained in elementary mechanics of materials, the plane sitess-sirain relation (or constitutive rela. Lon) of a faneariy edastic anu isorropic material is & VE wB 0) fa, few gp = |-WE YE 6 yr + 1 Ee GID ln ee Ke | ie} where £ is the clastic modulus, is Poisson's ratio, and G is the shear modulus, @ = 9.8 BCL + ¥). The last column vector in Eq, 2.1-1 contains inital strains (described be- Tow). Abbreviated, Eq. 3.1-L 38 vaitten € = “Yar + e,, If this equation is solved for the sitess veetor «7 we have os ketoy B12) ty 0 v1 0 | forplane stress B13) 9 6 dH, Equations 3.1-1 and 3.1-3 pertaia to a plane stress condition, in which =, initial suzsins , caused by temperanure change AT are €q = Go = @AT and Yu where @ is the coefficient of thermat expansion. The thickness is free co increase or de- crease in response to stresses in the xy plane. In 2 plane strain condition thickness change at 2 Plane Problems is prevented, Equation 3.1-2 still states the stress-strain relation, but Fis o-) oy a (i) | forplane strain 34-4) 0 6 G-2vy2 oxi) and inital strains due to temperature change AT are &,o= &9= (1 +s} AF end ‘As an exemple of plane stain, if «flat plate is bemt $0 as to become « cylinder with 2 is axis, mss Sections in planes normal 10 the z axis axe in a state of plane strain (except ‘ery near ends of the eylindey), Stresses are independent of zn plane stress and in plane stain conditions. As the thickness of a plane body increases, from much less to greater tim in-plane dimensions of te bedy, there is a transit of behavior from plane stress ‘war plane stain. If » = 05 the material is incompressible. ify approaches 0.5 and plane stain conch tions prevail, Bq 3.1-4 shows thar strains ¢, and €, are associated with very large stresses , and 6, This citcumstanse may cause trouble ib FE analyses because of numerical tl conditioning. Equation 3.1-2 need not be restricted to isotropy. In the most general case of anisotropy, B is a full matrix, aad for « plane problem it contains six independent elastic constants, The theory and computational processes of FH ate not made more complicated by anisouopy. However, there is often practica difficuley in obtaining numerical values of elastic conslanis. Also, itis harder to judge che validity of results because response 0 Toads is not as easily visoalized and approximate calculations become more dificult Stresses in plane stress problems may be called membrane siresses. They are constant trough the z-direction thickness. In contrast, the bending stresses that appear in plates snd shells vary from tension to compression through the thickness and by definition are sbseut if the problem is plane. One should bear in rind that afl physical structures ase three-dimensionsl, so that regarding 2 problem as plane (or as a bar, beam, plate. or shell) inmpties thal at least a small amount of idealization has already taken place Strain-Dispfacement Relations. FE theory makes extensive use of siraiit—displacement relavions. They aro used to obtain a strain field from a displacement fickd. Recall that no mal strain is defined as change in length divided by original length and that shear sta defined as the amount of change in a right angle. Thus im Fig. 3.1-1 we have &, = AufAx, £, © AVIAY, and jy = Away + AV/Ax, However, in general the x-direction displacement L . - ; a o A ae a w wo Fig. 31-1. A rectangle of incremental size, subjected to (2) x-direction strain. ¢b) »- ‘Sreston stain. and (c) shear strain. A Inroduction 43 und the y-direction displacement v are both functions of the coordinates, w = (x, ») and v = v(x, 3) in a plane problems. Therefore we must use partial derivatives. Doing so, and passing tothe limit, we write ® : cae (3.1-3) a & dae 0 , Pe th or e=du G16) re) lala aft ‘These strain definitions are suitable if the material has sma strains and small rosations. Otherwise the strain definitions must be more extensive 3.1) Displacements in plase finite element are interpolated from nodal displecements 1, and o, as follows: si} f}- fm geod fel or vend GED le oy 0 om mE where the 1, are separate shape (or interpolation) polynomials und N is ealled the shape Junchon matrx, Agcording to Eg, 3.1-7, 1 depends only on the, © depeass uly om the ‘uj ane w and v use the same interpolation polynomials. This is a common arrangement ‘bot it is not mandatory. An insiavee of Pq. 3.1-7, for « particular triangular element, sp- ‘pears in Eq, 1.2-3. From Eqs, 3.1-6 and 3.1-7 we obtain e=aNd or e=Bd where B= aN G8) Matix B is called the strain-displacement matris, A General Formula for k. Several texts on mechanies of materials derive an expression for Up, the strain energy per unit volume of an elastic material, To terms of strains and in matrix format, this expression is Uy = é”Re’2. Upon integrating over element volume ¥ and substituting from Eq, 3.1-8, we obtain clement straia energy U as pfetBeay =ye"feTeBava = dd’ka G19 (One can interpret Bq. 3.1.9 as follows, Let any element dot, say the ith da. be in- creased from zero to the value d, This is accomplished by spplying to the Co.f. x force that inereases from zevo to F,. The work done is Fjdj/2, just as it would be if stretching @ linear spring an amount d,, This werk is stoced as strain energy U. Equation 3.1-9 says that work Fid/2 is equal 9 strain energy in the element when the element displacement 44 Plane Problems Geld is that produced by d, and the clement shape functions. For example. if'd, = wy. we see from Eq, 5.1-7 thet the element displacement field is w(x, ») = Nun and vt, 3)= 9. In Bq. 3.1-9, the expression k= [pres av 1-10), js identified 4s an expression for the element stifvess matrix. Equstion 3.3-10 is not ro- stcted to plane problems; itis applicable to all displacement-based finite elements. The dizect method of generating ik. applied to bars and beams in Chapter 2, is not general be- cause there are nu formulas that relate nova forces to nodal displacements for elements of arbitrary shape. We see from By. 3.1-10 that for a given E, the nsture of k depends emtinely on B, ‘which in tm is derived from N by prescribed differentiations. In other wones. she behav for ofan element és govemed by its shape fimczions. In subsequent sections we will ex- amine the displacement and strain fields of several elements and use the field informa tion to predict how the element will behave and what its defects will be, There are practi- cal reasons for this study, In FE modeling, one secks # good match between behavior that the actual structure is expected to display and behavior that clement are able to display, and one chooses element «ypes, shapes, and sizes accordingly. Also, en anslyst who un- detstands the limitations of element bebavior will not have unrealistically high expecta sions of the capabilities and accuracy of the FE method. Londs. Mechanical Joads include surface trictions, body forces, and concentrated forces and moments, Surface action is a disiributed load upplied co a Boundary of the structure, that is 10 a boundary Tine in two-dimensional problems and t a boundary surface in uee-dimensional problems. Pressure loading is called 2 taction even though it pusies rather than pulls on the boundary. Also, truction may act either normal or tangent 10 8 boundary (Fig. 3.1-2). Body forces act throughout the volume of a structure rather than only on its surface. Body forces are usually caused by acceleration and occasionally by & magnetic Geld ‘Typical accelerations are the centripetal acceleration in rotating mackinery and the accel. eraiion of gravity, which produces self-weight loading. In one-dimensional elemients ere is no distinewon bevseen body and surface Toads because mathematically the ele- ments a fine, Similarly, the sel-weight of 2 horizontal plate can be replaced by pressure acting normal to the plate midsusface, which represents the plate mathematically Tig. 34-2. Plane body’ with nonzero action along ACD, ato taetion slong AB, and fixed suppor slong D4. Axes neand + are, respectively, normal and tangent 10 che boundary, 3.1 Imsroduction 8 Concentrated forees and moments come from prescribed leads and from the reactions of point supports. & concenitated load is & convenient substitute for a foad of high inten~ sity distributed over a small area, Thermal loads are teaied within the sofiware in the avanner described in Section 26. Boundary Conditions. Boundary conditions inckide presctibed displacements and pre- scribed sucface tractions. Usvally both appear in a given problem because part of the boundary is supported and another part is loaded. In Fig. 3.1-2, for example, displece- rents ate prescribed along DA and stresses are prescribed along ABCD. In a EE model, nodes along AB would receive no loads, nodes along BCD would be loaded by forces from the norm! and tangential tractions, and all nas alony ABCD would be free to dis- place, Nodes along DA would have their d.0.. set to zero and woukd nat be loaded by prescribed forces, All boundary and internal nodes of the mode! except nodes along DA aight be loaded by forces associated with sel-weight loading and would be free 0 dis- place, In structural mechanics, the teem “support condition” is used as a synonym for & displacement boundary condition. Nature of the FE Approximation, We smust preface our brief discussion by writing cocuiibciom equations and defining “compatibility.” Stresses are in gonerel Functions of the coordinates, so thet each stress has a rate of change with respect to 2 and y. In « plane problem the rates of change satisfy the equilibrium equations [6.1] Bry, a0, 5 Foy Gry ae where, and F, ure body forces per unit volume, If Eqs. 3.1-11 are satistied throughout a plane body, every differentia! element and the body itself ae in static equilibria. As tor Gefocmations, they are called compatible if displacciment boundary conditions ate tet and the material does not crack apart or overlap itself Elasticity theory shows that if displacement and stress fields simeltsacously satisfy equilibrium equations, corpatiility, and bouadary conditions on stress, then the solu- tion obtained is exact. How is an exact solotion approached by a FE approximation? Let clements be based on polynomial displacement fields, ss in this book and indeed a5 for most elements in common use. ‘Then the compatibility requirement is satisfied exactly within elements. Equilidriam equations and boundary conditions on sitess are not satisfied: thet is, Bos, 3.1-11] are not satisfied at most points within the FE model, and stress boundary conditions are not satisfied at most points on the boundary (@8. in Fig, 3.1-2 the computed stresses 7, and r,, will not be precisely zero all along the tnloaded boundary 4B), Stress boundary conditions and equilibriua equations are sat- isfied in an average sense: it can be shown that integrals of the left-hand sides of Eas. S.L-L vanish over eacit element, As a mesh #8 repeatedly refined, pointwise satisfae- tion of stress boundary conditions and equilibrium equations is approuched more and more closel The foregoing discussion also applies to theee-dimensional elastic problems, for which Eqs. 3.1-11 must be expanded to include all six stresses, The nature of a FE soludon is further discussed in Section 4.8. 46 Plena Problems 3.2 CONSTANT STRAIN TRIANGLE (CST) ‘Tne CST elersent, Fig. 3.2-1, is perhaps the earliest and simplest finite element, In terms. of generalized coordinates fits displacement field is w= B+ Bx + By G24) B+ Bax + Boy and, froma Egy, 3.1-5 and 3:2-1 the resulting strain field is eth 6B te Bt Bs 3.22) ‘We see that strains do not vary within the element; hence the name “constant strait trans gle” (CST for short). The element may also be called a “linear tangle” because is dis- placement fic is linear in x end y. Blement sides remain straight as the element deforms, (Element sides have the appearance of bar elements discussed in Chapter 2, but # plane, FE is not an assemblage of bars; a plune FE is the region bounced by its sites.) We omit the algebra neoded to recast Eqs. 3.2-1 in the shape-function form of Bq. 3.4-7. ‘The algebra is tedious [2.2] snd does not help us understand how the clement behaves. ‘The strain field obtained from the shape functions, in the Form e = Bd, is S] | Pps 2 yn 9 ye 0] et=t}O xy 0 4 0 aul} G23) 26 . sete x and yay ada coowines = 1.2 Deg ry—ayand yng CF 2, 3). and 2A $s twice the grea of the trisngle, 24 = 23,99) —-%5,¥2y. Node numbers are ‘tbitrary except that the sequence 123 must go counterclockwise around the element if 4 4s to be positive. Again we see that strains do not vary within the elemeat, Applying Eq, ‘S.1+10 we obtain ths element stiffness matrix: k= BYEBA B24) Fig. 32-1. A eonsten: strain wianele, lis six nodal xu dob-are shown. 3.3 Linear Shain Triangle (EST) a i te Fig, 32:2, (a) Suess 9, along the x axis in a beam modeled by CSTs and loded in pure bending. (b} Defonpation of the lower-lefi CST inthe rade where # is the element thickness, assumed constant, and FE comes from Eq, 3.1-3 if plane stress condicions prevail, Integration in Eq. 31-10 is uiviel because B and E contain only anstants The CST gives good results in « tegion of the FE model where there is litte strain gra- Siont, Otherwise the CST does not work well. This is evident if we ask the CST to model pare bending (Fig. 3.2-2). The x axis should be stress-free because it is the neutral axis of the beam, Instead, the FE mode! displays 6, 26 2 square wave pattem, Alse. the FE madel predicts deflections and ¢, stresses that are ymly sbout ene-quurter of the correct values. ‘The inability of the CST to represent an ¢, that varies linearly with y is partly to blame for this poor result, But the CST also develops a spurious shear stress when beat. This is soen in Fig. 3.2-2b. It is proper that 1, and 1, appear as shown, bat v, creates a shear stress thar shew not be present. an expression for the shear strose eun ba obscined by sie of gs, 34-2, 31-3, and 3.2.3. Despite defects of the CST, correct results are approached as a mesh of CST elements is repeatadly refined, 3.3 LINEAR STRAIN TRIANGLE (LST) ‘The LST element is shown in Fig, 3.31 Iehes midside nodes jn addition to vertex nodes. The dog. are ig and at cach node &,2= 1,2, ..., 6, fora tot of 12 do. In enms of generalized coondinates 8 ts displacement ficld is Boot Bry + Bax’ + Boxy + Bex? 33-1) Br Bot By + Bro Busy Bax? . and, fom Eqs. 31-3 and 33-1 the resuhing sain fed is = f+ 2B Bay 8 = At Buxt 2ay G32) toy (Bs + Bs) + (Bs + 28,0) + 286+ By dy 4B Plane Problems a Fig. 33-1, (9) A tincar sosin wiange and its six nodal do (6) Diplacsment mode associates wih nod sof. v 46) Displacement mode associated with nal das. e.(Forvinuaizaton onl imagine dor dsplacomenteocurs norma ote plane ofthe element) (> ano rerite tory [22] by pemssion of Join Wie & Sons, ne) ‘The stcuin field can vary linearly with + and y within the element; hence the name “linear strain viangle” (LST for short ‘The clement may also be called a “quadratic triangle” be~ cause its displacement field is quadzatic in x and y. Element sides deform into quadratic curves when a single dof. is activated, as shown in Figs. 3.3-3b and 3.3-1e. The LST has ail the capabitites of the CST, few as they are, and more. For exarnple, Eq. 3.3-2 shows that stn €, can vary linearly with. Ifthe pate bending problem of Fig. 3.2-2ais solved using LST clements, exact results for deflection and stress are cbtuined. Additional au- merical examples appear in Section 3.11 ‘The clement stiffness matrix is most easily generated using “area coordinates.” The procedure does not help in understanding how the element bebaves. Details may be found elsewhere (2.2). We note only that the product BER is quadratie in the coondinates and that integration required in Eq. 3.1-10 can be done either in closed fonn by speciat for~ imulas oF numerically (Section 4.5). Numerical integration is necessery if elerment sides ase not straight but ourved, chat is, iniiatly curved. wher ail nodal 0, are zero 34 BILINEAR QUADRILATERAL (Q4) ‘The Q4 clement is # guaditaveral that has four nodes. Its nodal de.f. are shown in Fig. 34-1, Lu temus of geucralized cooidiastes 2 its displacement field is B+ Bt Bay + Boy 36+ Bix + Bry + Bay ‘The name “bilinear” arises because the form of the expressions for sand v is the product of two Hinewr polynomials, that is, (cy + ex¥){cs + egy}, Where the ¢, are constants. There are four parameters in each displacement expansion: four , or w and four fi for v in Bq, 34-1, or four shape fonctions N, in By. 3.4-3 below. From Bgs, 3.1-5 and 3.4-1, the ele» ment strain field is B41 = Bet Bor = B+ Bex G42) Fo = Bet B+ Bae + 34 Bilinear Quacritareral (Q4) 49 Fig. 34-1, A bil sodal do. Important aspeets of clement behavior can be deduced from Eqs. 3.4-f and 3.42. ‘The strain field shows that £, is independent of «, which means that the Q4 element cannot cxactly model a cantilever beam under transverse tip force (Fig. 3.4-2a), where axial strain varies linearly with x, Moreover, the Qd element cannot exacily modal a state Of pure bending, despite its ability to repzesent an €, that varies linearly with 9. Consider Fig. 3.4-2b, which shows » block of misterial loaded jn pure bending. We know from beam theory that shear strain %, is absent, that plane sections remain plane, and that top and botiom edges become arcs of practically the same radius of ccurvatute, as shown by dashed lines in Fig, 3.4-25. A Q4 element loaged in pure bend- ing is shown in Fig, 3.4-2c. Sides rolate, as shown by dashed ines, but top and botwom cdlges remain streight. ‘This result is dictated by Eqs. 3.4-1: along edges y displacement v is Bncarin x. indeed aif sides of a Q4 element deform as ‘Therefore right angles in the element are not preserved under psc moment loading and in consequence shesr strain appears everywhere in the element except along the y axis. The saan result can also be seem from Kgs. 3.4.2: the displacement mode of Fig. 3.4-2c requires that B, be nonzero so that é, will vary lineacly with y, but fi also ap- pears ih the expression for 7,,; therefore a (4 element that bends alse develops shear suain, (This trouble does no: appear in the LST in pure bending: when 8, in Eqs. 3.3-2 is nonzero, jy assumes # value such that A, + 2x is zet0 in the shear steain expres- sion.) Clearly, arguments of the preceding paragraph apply in similar fashion when bending moments are applied (© top and bottom edges of the element instead of to sides, The physical consequence of these defects is that the Q4 element is too stiff in beading be- ‘cause az applied bending inoment is resisted by spurious shear siess as well as by the ex- » fact eA L, @ i» © Fig. (a) A ope-element cantilever hear under transverse tip loading. (b) Correct deformae Udon mode of a rectangulee biocic in pure bending, (c) Deformation mode of the bilinear quadila- ral under bending Toad, 0 Plane Problems 34.3, Shape function N. of the bilinear quacie Iatecal, (For visualization only, imagine that dis- placement oocurs aermal to the xy plane.) pected flexural stresses, Further discussion of this problem, as well as a remedy for it, ap- eats in Section 3.6 If the fi, in Eq. 3.4-1 are expressed in terms of nodal d.o.£, we obtain the displace ‘ment field in the form of Eq, 3.1-7, where py GEDDY) dab fab | y, OTEK) yy, _ Goat yd G43) * 4ab * dab A -representative shape function (N, ) is plotted in Fig. 3.4-3. Note thar Wy = 1 at node 2 and A, = 0 ui every other node. This is true of shape functions in general, for any element ‘ype; that is, N= L at node ¢ and al node j where, j¥ f In the format of Eq. 3.3-8, the element susia field is wh fe] fer 2 ee fe) = yl 0 orn 0 ers) Sh aay Pal Aaex) -b-y) atx) -9) We can again deduce that the deformation mode of Fig. 3.4-2c contains spurious shear strain by svbstizting the nodal d.o. of this mode into Eq. 34-4, Equilibrium (Buys. 3.1-1) is not satisfied at every point in the Q4 element unless fi, = By Gia Bas. 3.4 1, in which case a stave of constant strain prevails. Despite this and colher criticism of the OM element, it converges properly with mesh refinement and most problems it works heuter chan the CST element (which ulways satisfies Eqs. 3.1-11) Examples of element behavior appear in Seotion 3.1. Equations stated in this section restriet the Q¢ element to rectangular shape, but this restriction van be overcome; see Section 3.8. 3.5 QUADRATIC QUADRILATERAL (Q8) “The QS element is shown in Fig. 35-1. In terms of generalized coordinates its dise placement fiald is b+ Bax + Bay + Bac + Br Bo? + Boy + Bes p= f+ Bott Buy + Box + Bony t Bio* + Bis + Bie” 5-1 3.6 Improved Biltwear Quadrilateral (Q0) 31 fee tle at Fig. 351. (a) A quadretic quadsilateral.(b.e} Shape functions NV, and N, . (For wisualization onty, Imagine that displacsreat occurs noc a Ue sy plane.) In more compact fom than Ey, 3.1-7, che displacement fied in terms of shape Functions Nis Yow SNe 352 whore index i runs from £ 10 8, whit ‘of the eight shape funetions are explains the “S” in the name Q8. As examples, two Ny = PSO lm) $0 By H+ Bd) tein) whore E= 4/0 and 1) = y/. By looking ata typical edge, for example, the edge x = a, we Sco from either Eqs. 35-1 or 2.5.3 that displacements are quadratic in», which means tbat the edge deforms into a parabola when any single des. om that edge is nonzero, From Eqs, 31-9 and 3.9, the element stam tela 1s = B+ 2x + Bay + 2Pay + Bay 5 Bit Buk + Bart Brot + Pio B5-4) Yor = (Bs + Bin) + (Bs + 2Biale + 2B + Bysly + Bot +206, + Bader + Bias? Each of the three strains contains all linear werms and sore quadratic terms (e.., there is 20 2° term in the ¢, expression}. The Q& element can represent cxactly all states of con- stant strain, and states of pure bending if itis rectangulas. Nonrectanguler shapes are per- mitted; see Section 3.8. Examples of element behavior appear in Section 3.21 3.6 IMPROVED BILINEAR QUADRILATERAL (Q6) ‘The principal defect of the Qa clement is its overstfness in bending, which can be iiles- trated by comparison of the bending moments in Figs. 3.4-2b and 3.4-2c, Let the rectan- gular block and the element have the same dimensions, elastic modulus £, and Poisson ratio v. Then apply wiretover bending moments Af, and M, are necessary lo make vertical 52 Plane Problems ner (4 stamens r Fig. 36-1. Quolitstive variation of axial sess and average transverse shear stress in 2 cuntlever beam modeled by rectangular Q¢ ele- sides of the block and the element include the same angle, 8, = @,. Moment M, is the cor. rect value. It can be shown that M, is etl) where @ and & are dimensions shown in Fig, 34-1. if aspect ratio afb increases without limit, so does ifs, which means that the Q4 element becomes infinitely stiff in bending. This phenomenon is eatled “locking” [3.2]. In practice we avoid elements of large aspect ratio, and & FE mesh does not “ioe” bat rather is overly stiff wher bent, a explained in Section 3.4. Qualitative rosults appear in Fig. 3.6-1, Defleetions and axial stress in the FE ‘model are smaller than the exact values, and (ranswerse shear stress is greatly in error ex cept along the y-pareltel centerine of each element ‘A remedy for the Uouble is fairly simple and produces an clement sometimes called the Q6 element (2.2, 3.3] ls displacement expansions for w and w each coniain six shape Funetions: that is, Die +0688 + Coes B62 De + Oe + mes where 2 = va, = y/b, and the JV, in the summations are shape functions of the Q4 ele ‘ment, Eq. 34-3. In Eq. 36-2.we have simply augmented the displacement field of the Q4 element by modes that describe a state of constant curvature. This #s easy to see for modes associated with Lo. g. and gy! as shown in Fig. 36-22, they allow edges of the element to become curved. Accordingly, the Q6 element can mode] bending with either an x-parallel neutral axis or 4 y-parallel neutral axis; indeed gy und g, can be nonzero sis 3.6 Improved Bitinear Quuleiaserat QO} 3 o ior @ Fig. 3.6.2. (a) Displacemem modes a= (1 — Py and ify In the QS element (6) Incompatibility berween adjacent Q6 elements. (6) No incompatibility ewween acs ‘con’ Q4 element, smultaneously. Modes associated with d.o.f. g, and g, allow the existence of strains nor- ‘ma to a beam axis that uppear because of the Poisson effect. From Bgs. 3.1-5 and 3.6-2, shear strain in the Q6 element is - 2, - Bs 653) Jn pure bending, the negative testes (2y#6%)g, ancl Qxfag, are equal ia magainude fo pos itive wumms produced by she summations, thus permitting shear strain to vanish, as is proper. The Q6 element can represent pure bending exsetly, but only if the element is rectangular. (This point is discussed farther in connection with Fig, 3.8-2.) Qualitative re suks appear in Fig. 3.6-3, Axial stress is exact slong the y-parsliel centerline of each ele meat, and average transverse shear stwess is exact everywhere. Further examples of ele- mont behavior appear in Section 3.1). The Q6 element, or a differendy formelsted | lncanetinls (25% sermons F hems | Fig. 36:3. Quolitaive variation of bg aia tions and average transac Ft ans exact shear sioss 2F/A = 24/t in a can tilever beum modeled by rectangular QSelemenss, 54. Plane Problons clement of comparable behuvior, 18 usually the defuule option for a four-node quadritat- cral in commercial software, The dof. g; shrough 2, are intemal do. Unlike nodal dof. u, and vy . they are not connected to corresponding do. in adjacent elements. Modes associated with dof. g, tre incompanible. That is, under some (hut act ail) losdings, an overlap or a gap moy ap pear between adjacent clements. This point is made in Fig. 3.6-2b, No gap appears with (Q4 elements under similar loading (Fig, 2.6-2c). Indeed no gaps or overlaps appear in a physical continaum; why then is the Q6 clement acceptable? It is hecause elements ap- [proach a state of constant strain as a mesh is repestedly refined, In a state of constent ‘rain all initially straight lines, including element edges, remain straight efter deforma. tion, Then there is no incompatibility between elements, Thus mesh refinement produces convergence toward comect results, 3.7 ELEMENTS WITH “DRILLING” D.OF. A “Grilling” do. is a rotations .0,f. whose vector is nownal 10 the plane of sn element ‘Thus @, isa drilling do.f, at node # for ap elentent ia the ay plane. Elements with drilling, dog, a8 not yet in common use, so we discuss them only briefly An element edge thar has 2 midside node can deform into either @ straight or @ para bolic shape, As shown below, translational é.o.f. at midside can be expressed in terms of translational and deilling dof. at corners. This permits an exchange: translational 4.0.8. at ridsides are traded for driling dof. at comers. Consider the LST element in Fig. 3.7-1a, Let n, sepsesemt displacement normal o side 2-3. Norma} displacement at nade 3 ys writ- ten tips = Hoga ya) $893 ~ Boa en) where 8: and 6, are drilling do, at nodes 2 and 3 and Loy is the length of side 2-3, ‘Tangential éisplacement at midside is written tg? Hus +) 137-2) “Toese equutions mean, for example. that ug, ty =D and y= 02s = @, side 2-3 6is- aces into a patabola wih end rotations @ und conterdispuccrm ty = Glaahs IE Be 12 dos. set te © to Fig. 37-1, (a) Displacements normal to one sive of a LST element. (b) The do. in a LST ele rent, (e) The dog, ina riangular clement sith dling dod. 4.8 Blements of More Generel Shope 35 6, the edge remains staight, Equation 3.7-2 constrains tangential strain ¢, 10 be constant slong edge 23 Equations analogous to gs, 37-1 snd 37-2 are also written for the remaining twa sides. When these six equations are swbstituted inca the shape funetions of the J.ST ele- ment, along with sine and cosine terms to relat discctions m and s to directions x and y. mmidside nodes disappear and their d.o.l. ase replaced by drilling do. at comer noses. Because the three edge strains ¢, are constrained

    You might also like