empl oyi ng an i mpl i ci t - expl i ci t al gori thm M. S a mo n d s , K. Mo r g a n a n d R. W. Le wi s The f i ni t e el ement met hod is used t o model t he sol i di f i cat i on process i n a sand casti ng. Wi t h a r el at i vel y coarse mesh, t her e is good agr eement bet ween numer i cal and exper i ment al resul ts. An i mp l i c i t - e x p l i c i t t i me- st eppi ng al gor i t hm demonst r at ed si gni f i cant c omput er t i me savings when t he appr opr i at e sol ver was i mpl ement ed. Consi der at i on is also gi ven t o t he pr obl em of capaci t ance mat r i x l umpi ng when quadr at i c el ement s are empl oyed. Key wor ds: mat hemat i cal model s, f i ni t e el ement , sand casti ngs, i mp l i c i t - ex pl i c i t al gor i t hm There has been increasing interest over the past t wo decades in the use of numeri cal met hods t o model solidification processes in metal castings. ~-3 With an accurat e predictive model , the f oundr y will be able t o dramat i cal l y reduce shrinkage por osi t y, det ermi ne the opt i mum pl acement and size of risers, minimize trial runs and scrap met al , and even cont rol the mechani cal properties of the finished castings t hr ough modul at i on of t hermal gradients and solidification rates. Whereas the f'mite difference met hod has been used by a number of researchers, 4' s and in many appl i cat i ons uses less comput er time t han finite el ement s, it has the severe disadvantage of being very unwi el dy when it comes t o compl ex geometries such as those t ypi cal l y f ound in casting. It is shown here t hat a t wo-di mensi onal finite el ement program can give excel l ent results for a t apered slab sand casting, even wi t h a fairly coarse mesh. Concern has been voiced in some quart ers about the effort t hat is wasted by model l i ng the t emper at ur e field in the sand f or m .6 The aut hors, however, f ound it adequat e t o use a single layer of quadrat i c el ement s in the sand. I f an i mpl i ci t -expl i ci t f or mul at i on is used wi t h an appropri at e solver, and the sand el ement s t aken as explicit, substantial savings in comput er time can be realized in compari son wi t h a fully implicit scheme. Ma t h e ma t i c a l f o r mu l a t i o n The finite el ement f or mul at i on of the heat conduct i on equat i on is well d o c u me n t e d ] ' a Basically, st art i ng wi t h the conservat i on of energy equat i on in a field g2, assuming no mass t ransport and wi t h t emper at ur e- dependent t hermal propert i es: d H. V. ( KV0 ) = p - - 0 (1) dO wi t h Dirich]et and Neuman boundary conditions on Fl and F: respectively: O ( x ) = O(x)f (t ) on F l (2) KO .ti + q + h(O - - 0,4) = 0 on 1" 2 (3) Discretizing the spatial domai n by finite el ement s and empl oyi ng t he Galerkin procedure, 9 the fol l owi ng syst em of differential equat i ons is obt ai ned: CO + KO = F (4) where t ypi cal component s of the mat ri ces C and K and the vect or F are defi ned by: Cii = p - ~ N i N ] d a (5) I2 Ki]= f V Nj . ( KV Ni ) d ~ + f hNi Ni dF 2 (6) i2 F 2 F i = -- f ( q- - hO A) U i dr= (7) F: Here N i denot es the finite el ement shape f unct i on associa- ted wi t h node i of the finite el ement mesh. Lat ent heat evol ut i on is account ed for by the t erm dH/dO, and march- 0307-904X/85/03170-05/$03.00 170 Appl. Math. Modelling, 1985, Vol. 9, June 1985 Butterworth & Co. (Publishers) Ltd i ng i n t i me is a c c o mp l i s h e d b y a t wo s t e p p r e d i c t o r - c o r r e c t o r a l g o r i t h m, wh i c h is gi ven as f ol l ows : Pr e di c t or [ C ( O n ) + A t " Y K ( O n ) ] o n + l = [ C ( O n ) - - A t n ( 1 - - 7 ) x K( On) ] O n + At nF( O n) ( 8) Co r r e c t o r [C(0~,) + at%K(O~)] ." +~ = ~ p . l [ C ( 0 ~ , ) - At"(1 - - 7 ) x K(0~,)] 0" + at" F (O~) ( 9 ) whe r e 0~, = / 3 0 ? . * 1 + ( 1 - 13) O n ( 1 0 ) and On+l = On+l ( 11) i . e. t he c or r e c t i ve i t e r a t i o n begi ns wi t h t he p r e d i c t e d t e mp e r a t u r e ve c t or . The s u p e r s c r i p t n r ef er s t o t he t i me - s t e p n u mb e r a nd p is t he n u mb e r o f c or r e c t i ve i t e r a t i ons . Wi t h 3' = 13 = 1/ 2, one ha s an u n c o n d i t i o n a l l y s t abl e Cr a n k - Ni c ol s on s c h e me wh i c h is s e c o n d - o r d e r a c c ur a t e i n t i me ) De p e n d i n g on t he n u mb e r o f i t e r a t i o n s r e qui r e d t o r e a c h a s pe c i f i e d c onve r ge nc e l i mi t , t he t i me - s t e p ma y be in- cr eas ed or de c r e a s e d. Pr oc e e di ng a l ong fines si mi l ar t o t hos e o f Hughe s I l. ~2 a nd l i u , t3 t he i mp l i c i t - e x p l i c i t f o r mu l a t i o n f o r t he f i r st - o r d e r s y s t e m gi ven i n e q u a t i o n ( 4) is as f ol l ows . The e l e me n t s o f t he me s h ar e cl assi f i ed as expl i ci t (3' = 0) or i mpl i c i t ( 0. 5 _~ 3' K 1. 0), e i t he r a pri ori or as a c o n s e q u e n c e o f s o me cr i t er i a, a nd t he i r c o n t r i b u t i o n s t o t he gl obal K a nd C ma t r i c e s ar e t r e a t e d a c c or di ngl y. I n t he case o f e xpl i c i t e l e me n t s , f o r e x a mp l e , t he c a p a c i t a n c e ma t r i c e s mu s t be l u mp e d . T o t hi s e nd, one wr i t es : K = K I + K E ( 1 2 ) C = CI + CE ( 13) F = F I + F E ( 14) Cons i der e q u a t i o n ( 4) as i t appl i es t o a pa r t i c ul a r el e- me n t . Ta ki ng V = 0 and -On+l as an i n t e r me d i a t e v e c t o r o f n o d a l t e mp e r a t u r e s , ba s e d u p o n t i me l evel n, one can wr i t e: CV n+l + KI On+l + KE ~n+l = F n+l ( 15) ~n+~ = 0 n + &t" Vn( l _ 7) ( 1 6 ) O n+l = 0 n +l + At "Vn+13" ( 17) or r e wr i t i ng e q u a t i o n ( 17) : v " * ~ = ( 0 " 1 - ~ " + ~ ) / A t " 7 ( 1 8 ) "= Su b s t i t u t i n g ( 1 8 ) i n t o ( 15) gives: C(On+ 1 _ ~n+ 1) + At n TKl On + 1 + A t " T KE ~ n + 1 ~_ = A t " T F n+l ( 19) "~ o a nd r ear r angi ng: x2_ [ C + A t " T K I ] O n + 1 = [ A t n T ( F n * 1 _ K e ~ n + l ) + c # " + ~1 ( 2 o ) I f t he e l e me n t is e xpl i c i t , i . e. 7 = 0, t h e n ( 16) b e c o me s : ~n +l =On + A t n v n ( 21) F i g u r e l b u t usi ng t he f a c t t h a t : CV n + KO n = F n ( 22) one sees t h a t : v" = c- ' [F " - K 0 " ] ( 2 3 ) a nd: c ~ " ' = c o " + Lxt " I F " - K O" ] (24) Thi s is e x a c t l y t he same as t he r esul t t h a t w o u l d be pr o- duced i f t he p r e d i c t o r ( 8 ) was used w i t h 7 = 0 t o gi ve On+ 1 F u r t h e r , w i t h 7 = 0 i n ( 20) : , co " *1 = c ~ "*1 = c o" + at" [F " - K O" ] ( 2 5 ) whe r e n o w C, K a nd F ar e e va l ua t e d a t t he ne w t e mp e r a - t ur e O h = ~ n * l + (1 - - 13) 0 n. Agai n, t he s a me r es ul t oc c ur s i f t he c o r r e c t o r ( 9) is used I n t he f ul l y i mpl i ci t case, i t c a n be s h o wn t h a t t he a l gor i t hm gi ven i n e q u a t i o n s ( 1 5 ) - ( 2 0 ) c a n be i mp l e me n t e d wi t h t he p r e d i c t o r - c o r r e c - t o r s c h e me wi t h 7 = 1. I t is t h e r e f o r e ne c e s s a r y o n l y t o s pe c i f y t he a p p r o p r i a t e val ue o f 3' f or e a c h e l e me n t a nd t he r es ul t i ng gl obal l i near s y s t e m: KEFFOn+ 1 ___FEFF ( 26) wi l l have t he des i r ed pr ope r t i e s The e xpl i c i t e l e me n t s wi l l c o n t r i b u t e t e r ms o n l y t o t he di agonal o f t he ma t r i x KEFF. Thi s s t r u c t u r e ma y be mo s t f ul l y e x p l o i t e d wi t h an act i ve c o l u mn pr of ' de s ol ver , H wh i c h onl y car r i es o u t t he Gaus s i an e l i mi n a t i o n o n t hos e p a r t s o f t he ma t r i x whe r e of f - di a gona l t e r ms oc c ur . Th u s , wi t h ma n y e xpl i c i t e l e me nt s , a con- s i der abl e r e d u c t i o n i n s ol ut i on t i me ma y be r eal i zed. E x p e r i m e n t a l a n d n u m e r i c a l r e s u l t s A t a p e r e d sl ab o f a l u mi n i u m b r o n z e was cas t i n a resi n b o n d e d si l i ca s and mo u l d . Th e r mo c o u p l e s wer e pl a c e d d o wn t he cent r e- l i ne o f t he c a vi t y a nd t he t i me t o r e a c h t he sol i dus t e mp e r a t u r e a f t e r p o u r was p l o t t e d as a f u n c t i o n o f pos i t i on ( Fi gure 1) . Thi s e x p e r i me n t a l wo r k wa s c o n d u c t e d a t t he Br i t i sh No n - Fe r r o u s Met al s Te c h n o l o g y Ce nt r e , Want age, UK. The t e mp e r a t u r e - d e p e n d e n t t h e r ma l p r o p e r t i e s o f t he me t a l wer e f o u n d f r o m r e f e r e nc e 14, a nd ar e s hown in Tabl e 1. The l a t e n t h e a t was c a l c ul a t e d f r o m t hos e o f t he c o n s t i t u e n t p u r e me t a l s b y a we i ght e d average The AIBz t aper ed slab 9 0 - Numerical resul ts using l ow sand conducti vtti es 8 0 . . . . Numeri cal resul t s using high send conducti vi ti es Experi ment al r esul t s 70 / 60 / . / " 5(3 . / s 40 30 ~ ' ~ 20 ~ ' ~ _ 1(3 0 I I I I I I I l I I I I 0 100 200 300 400 500 600 Di stance f r om end of casting ( mm) Experimental and numeri cal results A p p l . M a t h . Mo d e l l i n g , 1 9 8 5 , V o l . 9 , June 171 Tabl e I Mat er i al pr oper t i es f or a l u mi n i u m- b r o n z e Composi ti on: Copper 80. 5% Al u mi n i u m 9. 5% I ron 4. 5% Nickel 5. 5% Freezing Range: 1 0 8 0 - 1 0 5 0 C Lat ent Heat : 2. 37 X 10 s J/ kg ( 56. 7 cal / g) Specific Heat : 452. 2 J/ kg C Density: 7600 kg/ m ~ Co n d u c t i v i t y as a f u n c t i o n of t emper at ur e: Temper at ur e ( C) k (J/s m C) 200 5 4 . 4 2 4 0 0 46. 05 6 0 0 43. 95 8 0 0 41. 86 1000 41. 86 1200 41. 86 Tabl e 2 Mat er i al pr oper t i es f or sand Density: 1444. 7 k g/ m ~ Hi gh c o n d u c t i v i t y cur ve as a f u n c t i o n of t emper at ur e (C): k ---- 0. 6606 - - 2. 084 X 10 -4 T + 7. 741 X 10 -~ T a J/s m C L o w c o n d u c t i v i t y cur ve as a f u n c t i o n of t emper at ur e (C): k = 0. 2828 - - 1. 681 X 10 - " T + 1. 352 X 10 -~ T = J/s m C Specific heat as a f u n c t i o n of t emper at ur e: Temper at ur e (C) Cp ( J/ kg C) 200 975. 7 400 1092. 9 600 1151. 5 800 1159. 9 1000 1176. 7 propert i es of the particular sand used coul d not be de- t ermi ned, but reference 15 gives a range of t ypi cal values (see Table 2). Numerical results were obt ai ned wi t h the highest and lowest sand conduct i vi t y curves, whi ch bracket ed the experi ment al solidification times as shown in Figure 1. The vertical cross-section of the moul d and casting was enmeshed wi t h 56 quadrat i c i soparamet ri c el ement s (see Figure 2). The ext ernal boundar y of the sand el ement s was fLxed at 20C, wi t h the initial t emper at ur e in t he sand t he same. The t op boundar y of the riser was hel d at 1156C while t he riser sides were per f ect l y insulated. The initial t emperat ure of the met al was t aken as 1156C t hr oughout , and the initial t emper at ur e at t he i nt erface bet ween met al and sand was 1080C, whi ch is the liquidus t emperat ure. A condi t i on of perfect conduct i on was assumed at the i nt erface, i.e. no t hermal resistance. For want of bet t er i nf or mat i on, the l at ent heat was assumed t o evolve linearly over the freezi ng range. Because t he slab is relatively thin, convect i on in t he mel t woul d not be significant except in the riser, so no effect f r om convect i on was i ncl uded. Fr om an initial value o f 1.0 s, t he time-step was doubl ed i f onl y one corrective i t erat i on was requi red t o reach a convergence t ol erance of 0.5C. Three or mor e i t erat i ons woul d cause t he time-step t o be halved. Originally, 3' and/3 in equat i ons ( 8) - ( 10) were t aken as 0.5 for all the el ement s. Figure 3 shows the hi st or y o f t he solidus (1050C) cont our s at 90 s intervals for t he case when the highest sand con- duct i vi t y curve was used. Even wi t h such a coarse mesh, t he shape and progression o f sol i di fi cat i on cont our s is in accor d wi t h physi cal expect at i ons. A slab casting wi t hout t aper was also model l ed, wi t h all ot her condi t i ons i dent i cal t o t he previ ous casting. Figure 4 depi ct s t he sol i di fi cat i on cont our hi st or y, again at 90 s intervals. The fact t hat a large por t i on o f t he slab solidifies in one time interval, unlike the even progressi on f ound wi t h the t apered casting, suggests t hat there woul d be feedi ng probl ems in this region, resulting in shrinkage por osi t y. This is, of course, what occurs in pract i ce. I f mor e exact i nf or mat i on about the relationship bet ween fract i on solidified and t emper at ur e were available, it woul d be possible t o make a more definitive st at ement about likely areas of underfeedi ng. Using t he i mpl i ci t -expl i ci t al gori t hm, t he sand el ement s were t hen all t aken t o be explicit, leaving 3' = 0.5 f or t he Fi gur e 2 Fi ni t e el ement mesh (sand el ement s shaded) ] > Figure 3 S o l i d i f i c a t i o n f r o n t h i s t o r y f o r AI Bz sl ab ( t a p e r e d 1 : 2 0 ) 1 7 2 A p p l . Ma t h . Mo d e l l i n g , 1 9 8 5 , V o l . 9, J u n e Figure 4 Solidification front history for untapered slab highest sand conductivity metal. When a frontal solver was used, there was no im- provement in the cpu (central processing unit) time per iteration, as this solver could not take advantage of the changed structure of the global stiffness matrix. In fact, the global mat ri x is never actually formed. The frontal solver has the i mport ant features t hat it makes node numberi ng arbitrary and it requires less core space. When an active column profile solver was used, however, the solution time was reduced by 75% from t hat of the frontal solver. This is with a mesh where 65% of the elements are explicit. A good portion of this reduction is due simply to the fact t hat the prof'de solver is faster in general than the frontal solver. When the profile solver is used with 3' = 0.5 for all ele- ments, however, a reduction of 39% still results due to the implicit--explicit algorithm alone. Ca p a c i t a n c e ma t r i x l u mp i n g As not ed above, the capacitance matrices of the explicit elements must be lumped to give KEF F the desired struc- ture. With linear elements, this presents no problem: the terms of a row may be simply summed to the diagonal. This will be referred to as ' st andard' lumping. If this procedure is adopt ed with the 8-noded serendipity element, however, ill-conditioning is introduced 16 which causes severe stability problems and loss of accuracy. Hinton t7 and Donea Is proposed two different lumping schemes to restore stability. In Hinton' s met hod, the total heat capacity is preserved and the terms of the lumped mat ri x are proportional to the diagonal terms of the consistent matrix. Donea introduces a new set of weighting functions with which to form the lumped capacitance matrix. Both of these schemes were used with the tapered slab probl em and did, in fact, improve the stability. The accuracy was degraded in bot h cases, however, with the Hinton met hod being slightly better. In order to regain the accuracy of the consistefit mat ri x, a modified predi ct or- corrector algorithm was adopt ed for explicit elements: Predictor T n + l = [CD-- At K] T n Co, Corrector CDT n+l = [CD-C] Tn+t + [C--ArK] T n =, CD is the capacitance matrix diagonalized by the Hinton met hod. The effect of the consistent mat ri x is introduced in the correct or step. With the tapered slab probl em, when the convergence tolerance was taken as 4.0C, this scheme converged to within 1.0C of the consistent results, an error of less than 0.1%. What is more, it required fewer steps and iterations to reach a given point in time. This added to the comput er time savings due to the implicit-explicit algorithm alone. Nine-noded Lagrangian elements were then used with the same probl em. Standard and Hinton lumping, when applied with the unmodified time-stepping algorithm, gave nearly identical accuracy, with an error of less than 0.1% when compared with the consistent results. The modified predi ct or-correct or gave a slight i mprovement in accuracy, but at the cost of more time-steps, iterations, and mat ri x assembly time. Since standard lumping is somewhat faster than Hi nt on' s met hod, it is most advantageous when Lagrangian elements are used. In fact, even though a mesh with Lagrangian elements requires solution for more degrees-of-freedom, the total comput er time is less than for 8-noded elements with the modified predi ct or-correct or scheme described above, because fewer iterations are required to obtain comparable accuracy. Concl us i ons In transient field problems involving mixed media with widely different properties, such as metal casting in sand, a substantial reduction in comput er time may be obtained with an imphcit-explicit algorithm coupled with an active column profile solver. When 8-noded serendipity elements are used, a modified predi ct or-correct or scheme will preserve accuracy. With 9-noded Lagrangian elements, only standard lumping is required for good results. No me n c l a t u r e C capacitance mat ri x Co diagonalized capacitance matrix F right-hand side vector h convective heat transfer coefficient H enthalpy K conductivity n time-step t~ unit outward normal to surface p . mmber of corrective iterations q specified heat flux t time x spatial coordinates 13 weighting factor for determining t emperat ure at which to evaluate matrices 3' weighting fact or in two-level time-stepping scheme F~ boundary with Dirichlet condition F2 boundary with Neuman condition p density 0 t emperat ure intermediate t emperat ure Appl. Math. Modelling, 1985, Vol. 9, June 173 0 . pr e di c t e d t e mp e r a t u r e 0,4 a mbi e nt t e mp e r a t u r e s peci f i ed t e mp e r a t u r e ~2 spat i al d o ma i n References 1 Henzel, J. G. and Keverian, J. ' Predicting casting solidification patterns with a comput er' , Foundry 1964, 50-53 2 Weatherwax, R. B. and Riegger, O. K. ' Computer-aided solidifi- cation study of a die-cast aluminium piston' , AFS Trans. 1977, 85, 317-322 3 Sahm, P. R. ' Applications-properties-microstructure-solidifi- cation technologies: Challenge for computer simulation and modelling foundry processes', Solidification Processes: Computer Simulation and Modelling Workshop, CIATF-Cairo, Nov. 1983, pp. 2-18 4 Jeyaraian, A. and Pehlke, R. D. 'Casting design by comput er' , AFS Trans. 1975, 101, 405- 412 5 Hansen, P. N. ' Numerical simulation of the solidification process' , Solidification and Casting of Metals, Proc. Conf. Metal Society, Sheffield, July 1977, (1979), pp. 350-356 6 Wei, C., Hansen, P. N. and Berry, J. T. ' The Q-dot met hod - A compact technique for describing the heat flux present at the moul d-met al interface in solidification programs' , Num. Meth. Heat Transl. 1983, 2, 461- 472 7 Zienkiewicz, O. C. and Parekh, C. H. ' Two and three dimensional isoparametric elements for transient field prob- lems', Int. J. Num. Meth. Engng 1970, 2, 61-71 8 Morgan, K., Lewis, R. W. and Zienkiewicz, O. C. ' An improved algorithm for heat conduction problems with phase change' , lnt. J. Num. Meth. Engng 1978, 12, 1191-1195 9 Zienkiewicz, O. C. and Morgan, K. ' Finite elements and approximations' , John Wiley, New York, 1983 10 Hughes, T. J. R. ' Unconditionally stable algorithms for non- linear heat conduction' , Comp. Meth. App. Mech. Engng 1977, 10, 135-139 11 Hughes, T. J. R. and Liu, W. K. ' Impl i ci t -expl i ci t finite elements in transient analysis: Implementation and numerical examples',ASME J. AppL Mech. 1978, 45, 375-378 12 Hughes, T. J. R., Pister, K. S. and Taylor, R. L. ' Impl i ci t - explicit finite elements in nonlinear transient analysis', Cornp. Meth. in AppL Mech. and Engng 1979 17/18, pp. 159-182 13 Liu, W. K. ' Development of mixed time partition procedures for thermal analysis of structures' , lnt. J. Num. Meth. Engng 1983, 19, 125-140 14 Metals Handbook, 8th edn, Vol. 1, Properties and Selection of Materials (ed. Lyman, T.), ASM, Novelty, Ohio, 1961, pp. 1007, 1049, 1197, 1203-4 15 Pehlke, R. D., Marrone, R. E. and Wilkes, J. O. ' Comput er simulation of solidification' , AFS Monograph, 1976, p. 69 16 Zienkiewicz, O. C. ' The finite element met hod' , 3rd edn, McGraw-Hill, 1977, pp. 535-539 17 Hinton, E., Rock, T. and Zienkiewicz, O. C. ' A note on mass lumping and related processes in the finite element met hod' , Earthquake Engng Struct. Dynamics 1976, 4, 245-249 18 Donea, J. and Laval, H. ' An improved formulation of the parabolic isoparametric element for explicit transient analysis', Earthquale Engng Struct. Dynamics 1979, 7, 23-29 174 Appl . Math. Modelling, 1985, Vol . 9, June