CHAPTER 4 NUMERICAL INTEGRATION AND COORDINATE TRANSFORMATION30 4 NUMERICAL INTEGRATION AND COORDINATE TRANSFORMATION InChapter 3, the principle ofminimum potential energy was used to derive the element stiffness equations for a simple rod element. ‘To deduce these equations it is necessary to perform integrations over the volume of atypical element. Even forsimple elements these {integrations become quite complex. In this Chapter we describe how analytic integration may be avoided by the use of coordinate transformation and numerical integration. [Numerical integration is now standard in most finite element codes and will be used throughout the remaining Chapters. The advantages of using numerical integration, as ‘opposed to analytic integration, are as follows: ‘+ All integrations are carried out by the computer. For many types of elements, especially those involving two or three dimensions, analytic integration is both tedious and impractical + Elements with curved boundaries can be developed. Analytic integration for these ‘ypes of elements is usually intractable in two or three dimensions, + Properties such as Young's modulus and cross-sectional area may be allowed to vary throughout each element. Because numerical integration permits the material ‘properties to vary from point to point inside an element, itisalso possible to deal with nonlinear problems ina straightforward manner. Although there area number of methods avilable for evaluatingintegrals numerically, the ‘one most often used in finite element analysis is due to Gauss. This method will be used exclusively in the remaining Chapters. 4.1 One-dimensional Gauss Integration Gauss’s formulae are writen for finite integrals over the interval (—1, 1] and specify @ set of coordinates and weights. In one dimension these rules are ofthe form. | Fea = Speeym, (aay ‘where nis the numberof integration (or Gauss) points ,isthe-coordinate for integration point i, and ws the weight for integration point “umber of points coordinate weight exact for polynomial a x om of degree T 0 z T We 1 a Ue 1 0 59 3 59 5 6 5/9 ‘Table 4.1 Data for Gaussian integration The rules ae designed for cases where f(a) is polynomial. Indeed, a rule with n points is exact fora polynomial integrand of degree 2n~1 of less. For example, a one-point rulea gives no error for a constant or linear function, Whereas a two-point rule gives no error for «polynomial up to and including a cubic. Gauss rules may also be used for cases where the integrand isnot spolynomialbut the result wil, n general, be inexact. The sampling points and weights for various Gauss formulae are shown in Table 4.1. Other formulae for higher order polynomial integrands are also available and are given in many numerical analysis tems ‘illustrate the use of these rules, consider the third degree polynomial $0) = 48 +32 +21 ‘The exact analytical solution forthe integral of f(x) Over [—1, 1s ["Wosatrnetans Since f(a) is of degree three, a two-point Gauss rule is needed to evaluate this integral without error. Using the two-point rule we see that J fede = > Fea)w; = 0/8) x 1+ FCA 3) x1 («07 y+ 3.8) + 21/08) + 1) (4-218) + 347.8) + 264.8) + 1) “[F)o (Fe) land the Gauss rule performs the integration exactly, 42 Numerically Integrated Elastic Rod Element To evaluate the integrals that are necessary to form the element stiffness equations it is convenient to introduce a coordinate transformation, 1n this procedure we map the “real” clement onto a “model” element using a standard transformation. ‘The model clement is chosen so that all of the required integrations can be done nuerically in a SMraightforward manner. The shape functions and integrations, which were previously expressed in terms of the real coordinates, are now evaluated in terms of the model coordinates. To illustrate the process, consider the linear two-noded rod element that was, considered in Section 3.2, real coordinates model coordinates Figure 4.1: Coordinate transformation for numerical integration32 ‘The model coordinate, is chosen to range from —1 19 +1 so that Gaussian integration ‘may be applied easly. To derive the shape functions in terms ofthe model coordinate, we again use Lagrangian interpolation. For the shape function associated with node one we we know N, = Oat = 1 andthus guess that My = Ny) = AG -1) ‘where 4 isa constant to be determined. Using the condition that Ny = 1 when & gives 4 = -1/2, and thus 46-1 (42) Similarly, forthe shape funetion associated with node wo, weknow Ny = Oat § = ~1 and sess that Ny = NG) = 1G +1) In this case Ny = Lwhen § = 1 and hence 2 = 1/2, This gives N,=3E+1) (43) The displacement at any point inside each rod clement is again given by the equation an + Nat = [Me safe] Nu= INT “s where u, and ware the nodal displacements, N, and Nare respectively given by (8.2) and (43), and no [Mi A] {in addition to using the shape functions to define the displacements, we also use them to define the coordinates. The 3-coordinate at any point within the element can be written in a similar form tothe displacement according to NT (as) =n n= [Mi maf} x where xTe fs *] and and 3p ae the nodal cordnates. Substituting for the shape fictions and rearranging, we see that the model coordinate & is related to real coordinate x according.8 Bonn by my fe gee ‘When the same shape functions are used for both the coordinates andthe displacements, as in equations (4,4) and (4.5), the element is said to be isoparametric. Numerically integrated elements are often called isoparametric elements in the literature and the two terms are used synonymously. 1t should be pointed here that itis not necessary to use the same shape functions to model the displacements and the coordinates. For example, with ' quadratic rod element, itis possible to use quadratic shape functions to model the displacements bu linear shape functions to model the coordinates. Such an element is said tobe subparamettic. Thistype of element i often used where the geometry ofthe problem is simple and a high order is not required to model the coordinates, Its also possible to use superparametric elements, where the expansion used to model the coordinates is of higher order than the expansion used to model the displacements. Both ofthese types of elements, however, ae much less common than isoparametric elements. Differentiating (4.4) with respect tox, the axil strain i the od i given by (46) an where axial strain in rod cy 2 8 nip ma ut = (1; u;) = nodal displacements for rod Inordertocomputethestran-dsplacement matrix andhence te strain at any point inside am clement, we need to evaluate partial derivatives of the shape functions wth respect to the real coordinate x. Since the shape funetios are now expressed in terms ofthe model coordinate these derivatives ere found using the chain rule according 0 ANE _ d (1, de Ba ebes)g - aN, NodE | d (1, us Ze Ea ~ Aa +o)s 2 BE In the above, the quantity dx/d is found by differentiating (45) to give a, ay, fay, avy] f) an, em sty = Me, -[# ehh) 3s as) = Ee be "Note that dx/a is nown as the determinant of the Jacobian matrix and is usualy denoted 4s detJ. For a one-dimensional case the Jacobian is a 1x1 matrix and hence its {determinant isequal:o the matriitset. Substituting forthe partial derivatives ofthe shapeu functions in (4.8), we see that det J isin fact equal to L/2, where Lis the length of the rod element. (Once the strain has been computed, the ail stress inthe rod is given by the stress-strain relation o.= Ee where F is Young’s modulus. This equation is often written in the more general form of o=De a9) where stress-strain mati, oT = (on) axial sures vector ‘Te derive the stiffness equations for a numerically integrated element, we again invoke the ‘principle of minimum potential energy. Using the Jacobian transformation to change the Integration variable the strain energy stored in atypical rod element may be written as SHAY onde 4f ecteccersat Substtingequstions (7) and earning es 54] cuyepe ons a 4[ erwrene cers “ “ 4o'[ won aaa a0) ‘The external work done by the body force on the element is “ waa [Rude af aNee af xnecua , a 4 Defining b? (aan) ‘The element stiffness equations are again found using the principle of minimum potential energy, This states that35 a a aly Er =0 Where a is the potential energy for the element. Inserting (4.10) and (4.11) and differentiating we obtain the element stiffness equations as where k 4f BDB dets df an) J NT b dots af (a3) and fe (44) Equations (4.12) and (4.13) may be integrated numerically using the Gauss rule to give Sere, oets,v, (as) SINT b dots, (416) k f ‘here is the number of integration points, the value of the weight for integration PointiB; = BCE, isthe strsin- [ee] xk kx? (423) Lm) ay fio] a2 au ef} 425) ‘This is the same as the element force vector given in equation (3.17), which was derived by exact integration. 44 Derivation of Stitfness Equations for Rod Element using Virtual Work ‘The governing equations for finite element analysis may also be derived from the principle of virtual Work. Indeed, for many problems in structural mechanics, the virtual work principle is simpler to apply than the principle of minimum potential energy and is also more general ‘The principle of virtual work states that if@ and d--dd are any two displacements fickis satisfying the displacement boundary conditions, then equilibrium is automatically satisfied provided the following equation holds [steer [ora [sas 25 {In the above we note that e isthe strain vector whose length equals the number of strain ‘components, isthe stress vector winose length equals the numberof stress components, bis the imposed body force vector whose length equals the numberof foree componeats, 1 is the applied surface pressure vector whose length equals the number of force components, Vis the volume ofthe body, and Sis the area of the body over which suface pressures are applied. It should be stressed that, provided the changes in geometry ae infinitesimal, equation (426) is applicable to materials which have either linear or nonlinear stress-strain ‘elations. This means it can be used to derive the governing finite element equations for avery general class of materials. The principle of minimum potential eneray, on the other ‘hand, is only applicable to linear elestc or nonlinear elastic materials which obey Hooke's law. ‘ollustrat how the principle of virtual work canbe used to derive the element stiffness {equations in structural mechanics, we again consider the rod element discussed in Sections 3:2and 42. In this ease the virtual work equation becomes A [ ee (62n38 ‘where the strain vector is ¢ = {e,}, the stress vectors 0 ‘vector is 6d = {u), and the body force vector is b = [R}. (og), the virtual displacement Now from equation (4.7) it follows that T= Bul = WaT and hence eT = duTBT (428) ‘Substituting (47), (4.9) and (428) into the lft hand side of (4.27) we see that * et af area =f de BB ude owtaf BIDBdets dw (629) After isertng (4) ito the right hand sie of (427), we lo obtain ‘| Combining (4.27), (4.29) and (430) gives saToac =f aetwtnds = aota [" nr vaesae (420 asta f BTDBdet J dE w= owt [ NT b det dg ‘Since the variation duin the nodal displacementsis arbitrary andthe above equation must hhold forall such variations, it follows that af wopaudu =a [ NT bets df ‘This expression may be written in the usual form ose ‘where the element stifiness matrix and force vector are defined as, xeaf" oon aeaae toa f Wrause and the nodal displacement vectors These stiffness equations are identical to those shown in (4.12), (413) and (4.14), which were obtained using the principle of minimum potential energy.