Ghostscript wrapper for D:\Digitalizacja\MTS85_t23z1_4_PDF_artyku³y\mts85_t23z2.pdf M E C H AN I K A TEORETYCZNA I STOSOWANA 2, 23 (1985) T H E ST AT E O F E XT E N SI O N O F H O N E YC O M B G R I D ST R I P TOM ASZ LEWIŃ SKI Politechnika W arszawska Instytut Mechaniki Konstrukcji Inż ynierskich 1. Introduction I n th e papers [I - 4] con tin uum models of elastic hexagonal- type grid plates in plane stress state have been form ulated. Th e following Cosserat- type models have been analysed: two versions resulting from th e Wozn iak's concept of fibrous Cosserat media and so called n — models correspon din g t o R ogula- Kun in 's pseudocon tin uum description. M oreover a simple asym ptotic H orvay's model h as been recalled. In the mentioned paper [4] an „ a p rio ri" analysis of a ran ge of applicability of the Cosserat- type models has been pre- sented and several hypotheses con cern in g advan tages and disadvantages of the considered differential approach es have been pu t forward. Th e „ raison d 'et r e" of th e present work is to elucidate problems concerning accuracy of the Cosserat- type m odels of dense grid plates. Th e error analysis is carried out by an example of a grid strip with hexagonal structure in a state of extension. The problem is considered in depen den tly by m ean s of analytical approaches based on continuum descri- ption s as well as via finite element analysis which in th e considered case of the grid structure can be treated as an exact m ethod (the errors produced by com puter program are neglected). Th us a direct error analysis of th e considered con tin uum models viz. errors of evaluating displacements and rotation s of n odes as well as internal forces at th e nodes is performed. The statical problem con sidered m akes it possible to disclose scale effects following here from th e couplin g of con stitutive equation s by means of B tensor (see [1, 2]). As it has been pointed out in [2 - 4] m oduli B and C are determined non- uniquely; they depend upon the choice of a version of Cosserat- type description. These moduli can be treated as small param eters of t h e th eory. Their influence on the final analytical results is various; the aim of the presen t paper is t o analyse this ph en om en on an d, if it is feasible, to dis- , tinguish th e best differential approach which induces errors of the smallest values. The presented „ a p o st erio ri" analysis allows us t o appreciate the hypotheses of the paper [4] which have been obtain ed by „ a p r io r i" analysis of approxim ation of functions S a!l (k) in ^- represen tation. 2 * 188 T, LEWIŃ SKI 2. Formulation of the problem Consider an infinitely long hexagonal grid strip of the un it thickness in th e state of extension, F ig. 2.1. I ts height is denoted by L . Th e rods whose axes con stitute hexagons of sides equal to / are connected by rigid n odes. A position of th e hexagon s with respect to the unloaded boundary lines parallel to th e h orizon tal axis of symmetry is shown in F ig. 2.1. The main nodes (cf. [1 - 4]) are m arked by circles. Th e rods are assumed t o be made of isotropic elastic material, Young m odulus an d P oisson ratio being den oted by E _ j x1 F ig. 2.1 and v respectively. The heights of rods h (measured in th e plan e of th e strip) are assumed to be constant. The slenderness ratio v\ of th e bars is equal to P/ h2. I t is assumed t h at rods are sufficiently slender so as to classical, improved (by takin g in to accoun t transverse shear deformations) theory of bars can be applied. M oreover it is supposed t h at all con- ditions (concerning th e loads, density of t h e grid as well as wavelengths of deform ation pattern s, see [1 - 4]) being a starting poin t of the con tin uum Cosserat- type description are fulfilled here so as t o the mentioned mathematical models could be applied. The displacements an d rotation s of main nodes are approxim ated by functions tf,
' = 0, (3.5) lav' 4a
0) and — in termediate ones 0>!, y3, Ł < 0). Note that the function «*(*) does not depend on sign B so that u1 displacements of B < 0 u ' l x ) ~1 U 2 (x,y') U2(x,y3l B > 0 m2lx mMx) B < 0 m2lx) m 1 lx) P22(x) Fig. 3.1 192 T. LEWIŃ SKI main and intermediate nodes lieing in the same distance from the longitudinal axis of symmetry are identical. The function u2 consists of two term s; the first of them depends on sign.5 whereas the second one is independent of this factor. Thus longitudinal displacements u% vary accor- dingly to the choice of the sections being drawn along main or intermediate nodes (cf. Fig. 3.1). The function cp(x) is proportional to sign B. Thus rotation s of main and intermediate nodes, which lie at the same distance from the strip's axis, have same absolute values and opposite signs. Stresses p22 and m2 do n ot depend of sign.fi whereas the stress components p21 an d m1 depend upon this factor, cf. F ig. 3.1. N onlinear form of the graphs plotted in F ig. 3.1 results from coupling of constitutive equations (B <Ł 0). The results obtained can be devided in to two groups (a) an d (b). To t h e first group these quantities belong which do n o t vanish if one substitutes B = 0; whereas the quantities vanishing in the case of B = 0 constitute (b) group. The quantities of (a) type have nonvanishing values along transverse cross- sections. The quantities of (b) type take essential values in edge zones only. These areas which can be treated as effec- tive carriers (i.e. domains where values of functions cannot be neglected in com parison with the values of (a)- type functions) will be called further 5- effect zones. D espite apparen t imprecision of this definition we do not see any need t o give a precise one allthough such a definition can be formulated. In the subsequent section a numerical test of the theoretically obtained qualitative results and corollaries will be carried out. M oreover the errors induced by Cosserat- type models as well as by asymptotic model (in which B = C = 0) will be examined. REMARK 1 N ote that in the problem considered a density of strain energy e = - - (p^ y ̂+ m'xj (3.10) does not depend of the choice of main nodes, i.e. of sign B. REMARK 2 Consider analogous problem of extension of a lattice- type strip of hexagonal structure rotated at an angle n/ 2 from the position considered above (as in F ig. 3.1). I t can be pro- ved that in this case an assumption of one- dimensionality of a stress state leads t o a con- tradiction; the boundary conditions of the form of (3.1) cannot be fulfilled. REMARK 3 Examine rotations of the boundary nodes lieing at a distance of L / 2 from th e longitu- dinal axis of the strip, subjected to stretching stresses a yy = N / L , in t h e limiting case L - * co at / = const. By inserting x = 0 into (3.6)3 and taking into account th at L - > co the following formula ; i v i * 1 ( fr+ *)fr+ «) X12 is obtained. The R H S of Eq. (3.11) involves effective elastic moduli only. I n the subsequent section an accuracy of the derived formula will be examined. HONEYCOMB GRID STRIP 193 4. Numerical analysis by displacement method The subject of the numerical analysis are plane grid structures A and B (see Fig. 4.1 where the quarters of the structures are shown). The rods are assumed to be made of a steel with Young modulus E = 2.106 107 N/cm 2 . Transverse shear deformations of the rods are neglected. Crosssections of bars are rectangular \xh where h = \ cm or h = 2 cm. The internode distance / is equal to 10 cm. The both grids are subjected to stretching longitudinal forces P = 8660.250 N (cf. Fig. 4.1) hence the mean stress of tension reads a = — £ — = 1000 N/cm2. /1/3/2 The state of extension is realized by various ways (exemplary loads are shown in Fig. 4.1). For further analysis only these results are important which do not vary under various, statically equivalent systems of loads. The aim of the numerical tests is to create a one dimensional state of deformation. Thus at some distances from the loaded boundaries of the strip the displacements, strains and stresses (apart from the displacements u2 parallel to strip's horizontal axis) ought to assume stable values, viz. independent of the distance of the section from the loaded ends. Numerical computations (performed with the aid of the program STRAINS 75, computer ODRA 1305) confirm this supposition which can be interpreted as „discrete analogy" of Saint Venant principle. However, transverse forces in horizontal bars (which occur due to the fact that the strips A and B are of finite length) do not satisfy this condition; their values are not periodical. Nevertheless it should be stressed here that these forces are negligible in comparison with transverse forces in bars whose axes are situated at angles ± TC/3 from the horizontal symmetry line. Moreover it should be emphasised that periodicity od some quantities occurs in some boundary layer only, e.g. the moments in horizontal bars, reaching the greates values in the vicinity of the unloaded edges, vanish rapidly towards the strip's horizontal axis; the greatest, moments only (in B stripat the first, say, six nodes lieing at the edge) satisfy the desired onedimensional state of stress condition whereas the other (negligible) values vary at random. The complete set of numerical results will not be reported; indeed not they are of essential importance here. In the subsequent section selected results will be given together with analytical results obtained by approximate differential models discussed in the paper. 5. Accuracy analysis of the continuum Cosserat-type and asymptotic approaches This section is devoted to comparison of results analytically obtained in Sec. 3 with results produced by computer analysis (outlined in Sec. 4) of the A and B structures, cf. Fig. 4.1. Such comparison can be carried out, because a) honeycomb grids A and B satisfy the desired regularity and density conditions, b) striptype forms of A and B structures as well as the loads subjected ensure (as it has been pointed out in Sec. 4) a onedimensional state of stress so that the fundamental assumption of the presented in Sec. 3 analytical approach is fulfilled, c) deformations of structures vary smoothly except for the B zones lieing at unloaded. < o. bo s [194] HONEYCOMB G RID STRIP 195 edges. Th us t h e m ath em atical models, based on th e assum ption t h at strain energy density depen ds on th e first deform ation gradien ts only, may be applied. P rior t o perform in g an accuracy analysis of the considered differential models of h on eycom b grids, let th e effective elastic m oduli of the structures A and B, un der the assum- ption of slenderness of bars (rj « rf), be c o m p u t ed 1'. T h e following models are examined: Klem m an d Woź n iak version (see [2], Sec. 3) with th e set of con stan ts (A, ft, a, Bv, C), second Woź n iak — type (I I ) version (see [2], Sec. 4) with the moduli {X, \ i, oc, B*, C A) a n d x — models (see [4], Sec. 5) leading to th e con stan ts (A, fj,, a B°, C° X) ). Th e values of th e effective m oduli are set up in Table 5.1. 5.1 Approximation of displacements of nodes. 1. Displacements u 1 perpendicular to the strip's horizontal axis T h e displacem en ts u1 of th e n odes 188, 189, . . . , 198, lieing alon g the lines perpendicular t o th e h orizon tal axis of t h e strip A (cf. F ig. 4.1), obtained via displacement m ethod (de- n o t a t io n s: V) as well as by th e con tin uum m odels, are shown in F ig. 5.1. Relevant graphs of relative errors in duced by t h e latter approach es are plotted in F ig. 5.2. These errors are com puted by assum in g th e results of th e displacement m ethod as „ exact" ones. Table 5.1 U 10cm h l c m ] a b 1 2 n 100 25 M N / c m 2 ] 563178 109S390 p [ N / c m 2 ] 11781 91532 alN/ c m 2 ] 1487 11099 Bv[ N/ cm] 58907 457661 B"[N/ cm] 29453 228831 EPtN/cm] 5U70 398165 l=10cm hlc m ] a b 1 2 CV[ N] 393694 3081580 C A [ N ] 99160 7 9 3 2 7 9 C,°0)[ N] 542434 4271500 C(> ] 319324 2486630 q > ] 96215 701747 Th e „ exact " results confirm t h at behaviour of ul function is nonlinear. H owever, th is effect is so un con siderable t h a t it can n ot be shown in F ig. 5.1. I t can be n oted th at the greatest errors are in duced by: t h e zero- order approxim ation an d th e un stable (x = 1) m odel. Apart from th is a very good (0.5% - 0.6% error) approxim ation of u1 by (x = 0) version is worth em phasisin g. Th e second (II) version provides a slightly better results t h an th e first (I) on e. On th e basis of com pu t at ion s which are n ot reported herein it can be stated th at in th e case of h — 2 cm relative errors are greater t h an in the case of h ~ 1 cm. Ań analysis of th e u1 displacem en ts in B strip do n ot lead to new conclusions. N ever- theless th e analysed relative errors are smaller in th is case: th e (/ ) version induces ca. 1.4% errors (in A case —2 . 3 % ) ; th e (II) ver sio n —1. 07% (in A case —c a . 1.75%); (x = 0) m odel —c a . 0.34% (in A case —0. 60% ) ; (« - 1) version- ca. 4.4% (in A case —7.35%) . Therefore th e m ore dense a lattice is th e better are results. 2. Displacements u 2 parallel to the horizontal symmetry axis I n order t o exam ine u2 displacem ents a slightly m ore complex procedure should be applied since these displacem en ts change their values along the strip's axis. Apart from *' For the definition of i\ , see [2]. v results obtained by displacement method with the aid of the program STRAINS 5.0 2 . 5 0.5 asymptotic model " version I version II 186 169 190 151 192 Fig. 5.2 193 194 195 196 [196] HONEYCOMB GRID STRIP 197 this an approximation of main and intermediate nodal translations u2 should be considered separately, see Fig. 3.1. The displacements u2 of the nodes 188, ..., 194 (A strip) are shown in Fig. 5.3. Non linearities of functions / i , f2 interpolating „exact" translations of main and intemediate nodes respectively can be noted, but, on taking into account that their deviations from the straight lines are very small, it could not be shown in Fig. 5.3. However, differential theories produce curvings of u2 functions much stronger. _ J | _ . I Strip A I h=1cm 194 The displacements w2 of main nodes are approximated in a different manner than ana logous displacements of the intermediate nodes. An error analysis of u2 along crosssections perpendicular to the horizontal strip's axis proves (cf. Fig. 5.3, see also [5], where approp riate graphs were plotted) that main nodes'displacements are over estimated whereas the intermediate ones are underestimated. Approximation errors of the main nodes'trans lations u2 decrease in the edge 5zones while the analogous errors of computing the dis placements of the intermediate nodes increase in this zone. A behaviour of relative errors along the horizontal strip axis from the loaded edge to the transverse symmetry axis is worth examining. The relevant graphs are plotted in Figs. 5.4, 5.5. Absolute values of relative errors grow rapidly in the vicinity of the symmetry axis where u2 tends to zero; then their values go down and again grow at the loaded boundary. Similar diagrams concerning B structure are plotted in Figs. 5.6, 5.7. In this case a stabilization of errors along the strip axis is readily seen. 55 77 99 121 U 3 165 187 209 231 n o d es F ig. 5.4 20 " " • — ^ = j 1—«_ 1 — • _ - Jt=0 ~ r— asymptotic approac Str h= -~ — — • i — * — - . P A cm - I Ł — . — . — — . — — ^ - — ^ - ^ , — . ^ \ A ) \ i 22 tt 66 88 110 132 154 176 198 220 nodes F ig. 5.5 20 KT 10 1O 0 - 10 L asymptotic :p p r jach / Stl h ip E • 1cr i i n —̂' ——- 17 51 85 119 153 187 221 255 289 323 357 391 425 459 493 527 561 n o d e s F i g. 5.6 [198] HONEYCOMB G RID STRIP 199 10 5 0 - 5 ^ - 20 o I x=1 * — = Strip B h=1cm_ = asyr - m npto ic c Jt=Q pprć ach" — • ~ _ = . 1\ 34 68 102 136 170 204 238 272 306 340 374 408 442 476 510 544 578 nodes F ig. 5.7 It is worth stressing that approximation errors of displacements of nodes of the B strip are less than analogous errors in the case of the A strip. Thus the errors vary together with the density ratio l/ L . Moreover, it was tested that the errors grow if the slenderness ratio r\ increases. 3. Rotations of nodes q> Rotations of the nodes lieing along the cross- section lines perpendicular to the strip horizontal axis are set up in Tables 5.2a, 5.2b; the values of rotations are increased 1000 times. The results of the fable 5.2a are shown in F ig. 5.8. A similar behaviour of
T = s i n a c o s a ( / 1 ^ 2 2 ) s i n 2 a / ? 1 2 + c o s 2 a J p 2 1 , m = sin a • m1 + cos a • m2, where a denotes an angle between t and x1 axis. On the basis of the above formulae the stresses at the sections perpendicular to the bars joint in main nodes of the grid can be found. The longitudinal force, the transverse force and the moment at the nodes are 2 V « c r Z ] / 3 , r = T / | / 3 , M = m / ] / 3 , respectively. 3 Mech. Teoret. i Stos. 2/85 202 T. LEWIŃ SKI it) approximate slope- deflection equations'method. A main idea of th e m eth od consists in applying equation s which express in tern al forces in term s of strains, i.e. by m ean s of th e form ulae (3.3), [2] in (I) version or (4.1), [2], in (II) version. Thus only in th e two m entioned cases of con tin uum descriptions this m ethod can be applied. F ig. 5.10 Hi) exact slope- deflection equations'method I n tern al forces can be calculated by substituting th e values of n odal displacem ents (translations and rotations) in to slope- deflection equation s well kn own from th e classical theory of bars. H owever, it occurs th at such apparen tly n atural procedure results in com- pletely incorrect outcomes. The appropriate negative examples are presented in [5]. Th e- refore this method will n ot be applied in further analysis. Strip B h = 1cm 3,6 3.7 3,8 3,9 4,0 Zj V 43 V- %& .6 (longitudinal forcesM O^N ) i 4,6 4,7 4,8 - r3r - * F i g. 5.11 HONEYCOMB GRID STRIP 203 1. Longitudinal forces Sloping bars. The graphs of functions which interpolate values of longitudinal forces n sloping bars 513 512, ..., 524 525 (viz. in bars which are inclined from the horizontal symmetry axis of the strip at angles ±7t/3, see Fig. 4.1) are plotted in Fig. 5.11. By means of a prime (F, IF) and the lines —. —. —., the results due to (ii) method are distinguished. The best results are produced by (« = 0) version, the worst — by the asymptotic and unstable (x = 1) models. Horizontal bars. Appropriate graphs are shown in Fig. 5.12. Similarly to the preceding case the best results yield from (I), (II) and particularly from (u = 0) versions; the worstare produced by {x = 1) and the asymptotic versions. The latter model does not describe the jBeffect, of course. 2. Transverse forces Sloping bars. The functions interpolating transverse forces in bars 5 1 3 5 1 2 , . . . , 524 525 of B strip are plotted in Fig. 5.13. The remarks concerning approximations of 6 7 8 9 10 11 12 13 U 15 16 17 (longitudinal iorces)iO"3[Nl Fig. 5.12 18 19 3* 204 T . L E WI Ń SKI 51 5 51 7- 51 9- 51 8 5 2 1 - 5 20 - 5 22 S 23 >j IT) c sym ptot Strip B h = 1a n c m 5dGl \ \ t i1 • n' \J \ !rl70 7.1 7.2 7.3 7.4 7.5 7.6 7.7" 7.8 '/ es1 / 9 8.0 (transverse forcesl- 10"3 F ig. 5.13 longitudinal forces in sloping bars can be transferred to th e considered case of transverse forces. I t can be mentioned yet t h at the m ethod (ii) becomes in correct in 5- zones. Horizontal bars. Transverse forces in the horizon tal bars are equal to zero provided the considered strip is infinitely long. Th e analytical solution found in Sec. 3 (which is based on the assumption of an one- dimensional stress state) is obtain ed by virtue of th e approximate way of fulfilling the boun dary con dition s. Specifically th e con dition p2i = 0 on the loaded boundary is substituted by an in tegral expression ( 3.1) 8. Th us p 21 ^ 0 in each Cosserat version whereas the m entioned con tradiction does n ot hold in t h e zero- order theory where the condition p21 = 0 is n ot in con trast t o th e assum ption of onedim en- sionality of the state of stress. A dom ain of essential values of p2X is a 5- effect zon e. Th us th e solution T = p21- l]/ 3 = 0 is approxim ated exactly only by th e asym ptotic m odel. 3. Approximation of bending moments Sloping bars. The functions approxim ating m om en ts at th e n odes of th e ba r s: 513 - 512, ..., 527 - 526 in B strip are plotted in F ig. 5.14. One can poin ted ou t t h a t th e second version I I ' , method (ii) provides a very good approxim ation whereas th e ordin ary m eth od (i) leads (in the case of the same version) t o errors of ca. 50%. Two times increased results of (II) version are shown in F ig. 5.14, th e appropriate graph being den oted by 2 M n . Th e coefficient 2 results from th e ratio I? / 2?A = 2 when v\ — r\ . T h e physical in terpretation of nf in (II) version especially an d also in other versions is n o t clear an d is retain ed in this paper as open question ; some theorems on this problem are form ulated in [5]. Horizontal bars. Appropriate graphs standing for m om en ts at n odes of t h e bars 511 - 528, . . . , 527- 544 in B strip are displayed in F ig. 5.15. Th e considered m om en ts are 5 1 3 51 5 5 16 5 18 51 7 ui Si! t in DC 5 20 5? 1 a T 5? 3 I f \ Strip h=1cm 30 M i V L, I j I i j i 35 * ! 40 iN'Cm] -3 Fig. 5.14 i n , Strip B h=1cm.. 1000 5000 10000 bending moments Fig. 5.15 [205] 206 T. LEWIŃ SKI negligibly small comparatively with m om en ts in sloping bars except for the one m om en t (in th e bar 511 - 528) which is of th e sam e order as m om en ts in sloping bars. H owever, exactly this maximal m om en t is approxim ated in th e best way, specifically by (I ) and (IF ) m ethods. C on tin uum models supply qualitatively correct description of the ben din g m om en ts in horizontal bars. H owever, quan titative errors are con siderable; the best results yield from (I I ') m ethod. 6. Concluding rem arks 1) N umerical analysis confirms an existence of ^- effect zones which have been predicted previously by the analytical considerations based upon Cosserat- type models (the asymp- totic model does n ot describe this effect). N um erical test proves t h at th e effective height of this zone is evaluated with various degrees of accuracy, th e m inim al errors yield from the (II) Wozniak's version and from the unstable (x = 1) m odel. Both analytical consi- derations and finite element com putations show th at th e Ł zon e's height H varies alm ost independently of the strip's height L so th at a relative height of B zon es: H/ L decreases if L grows. 2) Quantities characterizing states of stresses an d strains of th e grid can be divided (as has been done in Sec. 3) into two groups (a) an d (b). C om puter analysis confirms that such a division is reasonable. Quantities of (a) type take essential values in every point of the strip's cross- sections whereas (b) — quantities take inconsiderable values outside the B — effect domains. I n the .S- zones relative errors induced by th e con tin uum models are extrem e since exactly in these zones gradients of if and
i