An efficient numerical solution for the multidimensional solidification (or melting) problem using a microcomputer Y. RABINE “Mechanical Engineering Department. Ben-Gurion University of the Negev. Beer Sheva, Israel and E, KORIN ‘Chemical Engineering Department. Ben-Gurion University ofthe Negev, Bese Sheva Israel (Received 21 August 1991 and in final farm 26 Februar 1992), Abstract—The sim ofthis papers to resent simple and efficient numerical technique for solving ransient ‘multidimensional est transfer problems with melting solidification processes. The proposed tcchnigue ompeses an enthalpy-based method for solving the problems by finite dierence scheme, lump systems ‘ehuvior being astumed foreach aode. The computation techniques abet consider allkinds of boundary conditions 18. conduction, vonvection and fadition alone of in combination. The numerical method ‘neglects convection effets nthe liquid phase. The importance ofthis method lies i the fact that solations fre obtuined with + personal microcomputer ths providing 4 convenient and reliable tool for wide use Jn solving many problems of practical interest. The proposed method was verified against the two exact Solutions avaiable from the itersture for a one-dimensional semiinfine domain. one with constant temperature Boundary condition aad the second with constant heat ux. The technique was demonstrated by solving four different cases of two-dimensional problens. A comparison ofthe results obtaned with a Imicrocomputer using the technique presented in this paper with aumencal results [rom the erature ‘btansd using conventional methods... fine differences and nite elements methods, which genecaly involve the use of large computers. shows yood agreement 1. INTRODUCTION Heat TRANSFER accompanied by phase change is of great importance in many industrial applications, e.g in food processing, casting and solar energy appli- ‘cations. Solidification and melting problems are non- linear in the mathematical sense due to the existence fof a moving boundary (interface) between the two phases associated with the celease of latent heat ‘Neither position nor the velocity ofthe interface can. bbe predicted in advance. Mathematical analysis ‘becomes yet more complicated when the physical properties of the phase change material (PCM) are temperature dependent and when the boundary con- ditions are of the type that apply to convection and radiation. Phase change problems have a limited number of analytical solutions. Most of those available from the literature apply to simplified and idealized systems and are one dimensional, These analytical solutions hhave been reviewed by Lunardi [1], and some of the basie solutions may be found in the book of Carslaw and Jaeger (2). + Present address: Faculty of Mechanical Enginceriag, ‘Techaion Israel Institute of Technology. Haile 32000, Ismet ‘Numerical methods appear to offer a more practical approach for solving phase change problems assum- ing the moving interface can be traced. Reviews of this type of numerical solution have been presented bby Shamsundar {3] and by Viskanta (4) ‘The fast cunning microcomputers that have appeared in the past few years are very useful for solving engineering problems. However, they have & limited dynamic memory. Thus numerical methods using small matrices and small numbers of operations become extremely important. Unfortunately, most numerical schemes, and especially those methods that are based on finite elements, use large numbers of variables and matrices to trace the moving interface and the changes in the temperature field. Other finite difference schemes use iterative algorithms as part of ‘the numerical solution for tracing the interface. ‘The objective ofthis paper is to present an elficient numerical method suitable for a microcomputer for solving multidimensional phase change problems with complicated geometries, for systems with diverse boundary conditions with respect to fixed temperature and to convective and radiative heat transfer. The proposed numerical method can also deal with phase ‘changes caused by internal heat sources, such as an eleeiic current flowing through the PCM. ‘The proposed method was tested against two exact onors YY. Raat and E, Kors NOMENCLATURE 4 ea xyz Cartesian coordinates 3 Cofficients matrix Bio Blot number Grek symbols ©) peat capacity matrix D thermaldisivity G,spenie est fy tee Cheat capacity ofan element P thermal ditasivity ratio, 2, @ Coordinate T boundary 1D agonal ength A imerval E* energy per nit volume © misty F radiation shape factor © nondimensionl temperature fe Convetve bet transfer coefiient ss thermal conduct rato, kik, he taditve heat transfer coefiient, © ondimensional time, 3-1/2 fh iertace location A thermal conductivity other sybot K conductivity matrix Given of ZL tatent heat |e umber of grid points Subserins | | format Pai, surounding | beat source. heat flux per unit of mo wvenee volume 8 index oF equation (12) 4 ineteat source € convection |e Charette length } Gkeulted element Bs nondinensionlraivs neighbor element tothe element |r Coordinate i ua | s coordinate m melting Si Stetan umber Predation time 5 tela 7 temperature wal > lume 3 nan constants vector 1 ottime solutions of phase change problems and against (wo ‘other numerical methods based on finite diflerences and finite elements, in four different cases. 2, MATHEMATICAL FORMULATION ‘The mathematical modeling presented below assumes that heat transfer by convection in the liquid phase of the PCM can be neglected. The heat balance equations that describe the transfer of heat during phase change can be broken up into three equations describing each ofthe three regions, the liquid phase the solid phase, and the solid/liquid interface. In the liquid phase. div (ky grad 7) +4 In the solid phase div (k, grad 7) +4 where k is the thermal conductivity, T the tempera: ture, gan internal heat source, p the density. C, the specific heat and ¢ the time. The subscripts $ and | denote solid and liquid phases, respectively, and m ‘denotes the melting front At the solijliquid interface: where isa direction normal to the interface, L the latent heat and f the interface location. The (+) in ‘equation (3) is applicable to solidification and the (~) to melting. The boundary conditions of equations (1) and (2) can include conduction, convection and. radiation, singly or in combination. The heat source can describe radiation at the boundary or into the PCM, or any other heat source/sink, such as an elec- tric current flowing through the PCM. For a constant thermal conductivity coefficient, equations (1) and (2) can be written as, . or eV THO= Gy Se TH TatOTn (4) or KVTHG= IGG TS Ta OTe (8) In order to simplify the mathematical model, theAn ficient numerical solution physical fact that in nonpure materials the phase ‘change takes place ina temperature interval Ty AT can be used. The PCM then acquires a new set of physical properties in the region of the melting tem- perature interval (subscripts m), and equations (3)~ (5) can be written as by PTHG* Oa Cue TETtDT. (6) ‘The new set of PCM properties (subscripts m) have to be chosen in such a way as to ensure that the heat balance integral on the interface of the PCM will not ‘change. In order (0 find the new set of properties, the heating process for a small unit of volume of PCM was analyzed (Fig. 1). Let us assume that the density and the specific heat are L Pn’ Cre o ee [" “pe care | G,tdT Jooyare [" Pe CAT n teh Tie Fic. |. Time-dependent temperature of unit of volume of PCM duning heauag from T, (below the melting tem- perature) to 7 fabove the melting temperature). The dot End-lash line represents the melting temperature and the ‘dove lines represent the boundaries ofthe phase change region ors Equation (9) may be simplified to ere {7 Equation (10) describes the amount of energy stored in a unit volume as if there were no need to use the new set of PCM properties. The density p on the right-hand side of equation (7) should correspond to 1 solid in the case of melting and to a liquid in the case of solidification. ‘The thermal conductivity coetficient can be esti mated as an arithmetic average kg = +h) TeTetdT. 1D ‘The new set of PCM properties makes it possible to express the phase change process in terms of a single equation : er koPT HE = GS 12) T TatdTa 0. The material CaCl, 61,0 was taken 4s the PCM. It has the following physical properties St= 0.175 when Se= CJL (Tu—To), f= 2.1 and v= 0.725, Figure 2 presents the numerical and the exact solutions. An infinite slab was taken as the sem infinite body forthe numeral solution (length of 0.2). Itcanbeseen that thereis good agreement between the analytical and the numerical solutions as long as the slab acts as a semi-infinite body from the thermal Point of view (+ = 0.054) Case 2. A one-dimensional solidification problem in an infinite body with constant heat fux. This type of problem is found in solidcation around pipes in very large media due to coolant flow. The intial temperature is uniform and above the melting temperature Oy = | when 0 = (T—T.)/(Ty— Ty). A uniform tine ‘heat source (with a negative sign) q'= 100 Wm stars acting and thereater Stays constant for «> 0. The material Ma(NO,),* 64,0 (61.5 wt%), NHANO; (38.5 w1%) was taken as the PCM. It has te folowing physical properties Sem 05125 where St = Cy/L*(Ty~ Tq), B= 0898 and x = 078, Figure 3 presents the numerical and the analytical temperature solutions as function of the dimensionless radius Rs, which isthe position of the thermal influence (9= 001) at ¢ = 0.3. The Solution of @ melting problem with electric current flow would be similar that for the postive line heat Incases | and 2, the one-dimensional phase change problems with constant temperature and constant heat flux boundary conditions were tested in Cartesian and cylindrica coordinate systems Itcan be sen that there is @ good agreement between the analytical and the numerical solutions of these one-dimensional problems Two-dimensional analytical problems are necess- arily limited to simplified cases. The numerical solu- tions of the two-dimensional problems for various boundary conditions were tested against the numeri cal solution of Lazarigs(6} based on finite ferences and that of Hsiao and Chung (7) based on nite cle- rents. The two-dimensional phase change problem ofa long prism is presented in Fig. 4, Because of the symmetry of the geometry, only the temperature feld in one eighth ofthe prism had tobe solved. Cases 23-6 present phase change problems of that prism for different boundary conditions and different initial temperatures, Case 3. A solidification problem with constant wall temperature. The initial state of the PCM is liquid av its melting temperature = 1 where 9 = (T=T.)(Tq—T-). The prism is suddenly exposed {© a dniform wall temperature 9=0 for ¢>0 Sea 0681 where S¢=C,/L"(Tq—T.). Figures 5 and 6 show the interface positions as a functionAn ficient numerical solution (r-ToyTm-T9) “2.00 on SeOI1s _w072s Bar 010 0.5 020 Distance From The Wall (2 Fic. 2, Melting problem due to constant wall temperature ‘of time along the diagonal and the center lines, espee- tively, On the assumption that the phase changes ‘occur in the melting temperature interval, the use of the new set of physical properties pm Cm enables a ‘200d estimation of the temperature field in the PCM, but makes estimation of the interface location more dificult. Here interface location is predicted by para- bolic interpolation. In order to take into account all the latent heat effects, the initial temperature was 10 os oof t = é Se0SIS € 078 e095 00 om chosen as T,-+ATx instead of Ty. From the agree ment between the numerical solutions, one can see that T,-+AT, is @ good choice for the initial tem- perature Case 4. A solidification problem with constant wall temperature, The initial temperature of the PCM is 9 =9/7, where 0 = (T—T,)|(Tq—T). The prism is suddenly exposed to a uniform wall tem- perature = 0 for + >0. Sr=2 where Si + =0.6 Analytical Solution 0.12 Analytical Solution +0030 Analytical Solon + =006 Numerical Solon © e012 Numerial Sotason © ¢=030 Numerical Solution 02 03 Dimensionless Radius os Fa. 3. Soldication problem due toa constant heat sink at the enter of un infinite eytinde:os YY. Rann and E, Kons x Fic. 4. Pictorial representation of the eros-secton of the pase involved inthe two-dimensional phase change prob: fem solved in cates 3-6, The solution referred to one eighth ofthe prism. (Ty~T), B = 0.9 and x = 0.9. Figures 7 and 8 show the intertace positions asa function of time along the diagonal and the center lines, respectively Case 5. A solidification problem with convection from the wall of the prism to the surroundings. The initial temperature of the PCM is 8 = 9/7 where 0 = (T—T,)\(Tm~T,). The prism is suddenly exposed by convection (0 a surrounding region at a tempera ture 8 =0. St=2 where St = C,iL* (Tm—T,), B= 0.9, «= 0.9 and Bio = h-s/k, = 3. Figures 9 and 10 & 2 show the interface positions as a function of time along the diagonal and the center lines, respectively. Cases 3-5 show good agreement between the three different numerical solutions. One can see that the present work agrees better with the work of Hsiao ‘and Chung (finite elements) along the diagonal of the prism and with Lazaridis' work (finite differences) along the centerline. ‘The last case (case 6) presents a similar problem but with thermal radiation at the boundary. Sach problems are solved by using the heat source gin each. of the boundary elements. Assuming a gray body, the hheat source ¢ ean be expressed by: gj = a" Fre: (T!~T%). The validity of this scheme is shown by the good agreement between the different numerical sl tions of the convection boundary problem (case 5) ‘The term (T/T?) can be estimated by writing the term 4°72,-(T,=T), where Ty isthe average tem- perature between 7, and T,. Using the new term, fone can write the heat radiation coefficient as sh,= 40° Fee- Ty. One can then compare the differ- ent solutions of the same radiative phase change prob- Jem obtained using the heat source and the radiation coefficient h, Case 6. Solidification due to thermal radiation from the wall of the prism to the surroundings, The initial temperature of the PCM is @-= 9/7. where 8 = (T=T,)/(Tq~T.). The prism is suddenly exposed by convection to a surrounding region at a tempera ture 8 =0. St=2 where St= C,/L*(Tm=T,) B= 0.9, «=09. The initial temperature of the prism is 363 K and the prism is exposed to the surround- ings at a temperature of 273 K. In that temperature ‘Along The Diagonal x2 Dimensions Ineeface Location & ‘resent Numerical Solution Finite Elements Solution By Hsiao etal. (7) Finite Diferences Solution By Lazare (6) 01s 030 04s Fi. 5. terface location along the diagonal ofa square repion of solidieation problem dve toa constant wall temperature, iil at the melting temperatureAn efficient aumerial solution 9 Dimensionless Interface Location ‘Along The Cente Line 1 Present Namereal Solon Finite Elements Solution By Hsiao etal (7) Fit Ditferences Solution By Lari (6) 01s 030 04s 1 Fic, 6, laerce location along the center line of a square region ina solidifiation problem du to & Constant wall temperate, initially atthe melting temperature. range the use of the term 4+73,-(T,—T) instead of the term (T3—T) involves an error range of 2%, The radiation coeficient ‘, is calculated for teach boundary element in each time interval. In the ‘numerical solution, the Biot number was changed. from 9.15 at the beginning to an average of 5.25 at + = 0.225, Figure 11 shows the interface positions as 1 function of time along the diagonal and the center lines, respectively. ‘The agreement between the two methods of solu- tion of the same problem, ie using A, oF 4, is of the order of 97%, We conclude that the translation of thermal radiation at the boundary to heat sources at the boundary elements is good. Use of the radiation Dimensions Inerfece Location ‘Along Te Digonal wX2 Present Numerical Solation Finite Elements Solution By Heino et Al. (7) Finite Differences Solution By Lazard (6) 005 0.10 ous: 020 02s Fig 7. Interface location along the diagonal ofa square cegion in solidifieation problem due to a constant ‘ell temperature, nally at aiguié temperature.80 Y. Ranw and E. Kons Dimensionless Inurface Location ‘ong The Center Line XL Present Numerical Solution Finite Element olution By Hao et A. (7) Finite Difleeace Solution By avai (6) 005 0.10 015 020 02s Fic. 8 Interface location along the ceter line of square region ina solidifiation problem due to a constant wall temperature, ntally 2 liquid temperature, coefficient becomes less accurate as the temperature difference between the boundary and the sur roundings increases. In order to evaluate the efficiency of the proposed. ‘numerical solution from the standpoint of application ‘with a microcomputer with a limited memory capacity, let us compare it with other common ‘numerical solutions based on finite elements Basically, to solve heat transfer problems by using finite element methods, one has to solve the following equation: CTHK-THG=0 as) ‘where C and K denote the heat capacity and the con- ductivity matrices, respectively, Gis the vector of heat (To-Tay(Tm-Ta}=9/7 Lt) ii é y Blo ‘Present Numerical Slaton Finite Elements Solution By Hsia etal, (7) + Finite Dilfeece Solution By Lazar (6) 0.02 004 0.06 0.08 010 Fic. 9. Interfuce location along the diagonal of a squace repion in a solidification problem due to a ‘convection atthe boundary, initially ata liquid temperatureAneficient numerical solution at ‘Present Numerical Salton Finite Elements Solution By Hsiao tal (7), Fite Differences Solution By Lazaiis (6) 0.02 0.08 0.06 0.08 0.10 ration along the centerline ofa square region in a solidification problem due to a ‘convection a the Boundary initially ata bquid temperature sources and boundary conditions, and T denotes the vector ofthe calculated temperatures ‘A detailed comparison of different numerical methods based on finite elements for solving heat transfer problems with phase changes is given by Dathuijsen and Segal (8). As suggested by Dalhuijsen and Segal, we will focus this discussion on the linear elements. There are several ways to accomplish dis cretization of equation (19) in time. The algorithm of Leas (9] was chosen for the present discussion, as was suggested by Morgan er a. (10), CO (TET ARTE ET ET of a +6°=0. (20) ‘This schemes straightforward and involves three time levels, where C, K and G are evaluated at the time (ToTaNT TH g sea DIAGONAL, q a9 4 sas i Biota a CENTERLINE i z a ‘Neal Slain Ug Heat Trautr Cael By Raion Numeral Slaton Using An nal Het Source 005 ou aus 020 o2s Fic U, Interface location ina square region in a sliication problem du to radiation at the boundary. ‘eiially a3 quid temperature,co level p. For the numerical solution, the last scheme willbe in the form Ber en where Bae Pack) 2) Vn Fear RTE (Ct pea ke) Te! Harr 23) The matrices K and C and therefore B are of order ‘n-n, where isthe number of arid nodes. During the computer simulation matrix B and vector V can be calculated directly, saving the unnecessary evaluation of K, Cand G. In cases of one-dimensional heat trans- fer problems (using linear elements) matrix B has three non-zero diagonals, and therefore equation (21) can be solved by a simple algorithm. In fact, in one~ dimensional problems, as in cases | and 2 above, itis preferable to save only 3+n of the nn coefficients of ‘matrix B in the computer memory. In two-dimen sional problems there are more than three non-zero diagonals. In simple two-dimensional problems, such ascases 3-6 above, there ae five non-zero diagonals in mateix B, however, solution of equation (21) requires saving at least (1%1)=n/2 coefficients of matrix B in-memory. However, the solution of the general three-dimensional problems requires saving all nm coefficients of matrix B. When estimating the amoun of dynamic memory required for numerical solution, ‘one must bear in mind that there are addtional vec- tors of orders nto be placed in the computer memory Tet! Tr, Te! and V. (In some schemes other than Lees" the vector T?-" is omitted but the remaining vectors and matrices do not change.) In contrast, in the proposed straightforward ‘numerical scheme, represented by equation (13), only ‘wo vectors of order n need be placed in memory: T**! and T*, The sizeof the dynamic memory required for this scheme is only a function of the number of grid points. The heat capacity C,and the thermal resistance to heat flow R,, can be calculated for each node at the relevant stage. The computer memory capacities needed for the finite elements and finite difference solutions are summarized in Table | To obtain some idea of the running time required for solving this kind of problem using micro- computers, the solution of ease $ was tested for two different meshes on three versions of IBM-PC com- ‘Table [. A compaston ofthe computer memory capacities required for solving the heat transfer problem using Bite ‘ement and finite eiference methods, where nis the number of ged points Proposed Fite solution elements aa Bngen One-dimensional Sn (thine Twosdimensional “Threedimensional Y. Raa and E, Kori ‘Table 2. A compariton of the cunaing times required for solving cae 3, wing tree versions of IBM-PC compatible microcomputers and with diferent numbers of snd nodes, 20x20 10x10 {ed points end points Microcomputer version 14s XT,8 MH, 8086-8087 2083 AT,25 Milz, 80386730387 63s 'AT.33 Mie, s0486 patible computers. The physical properties were & TW! Ko! a, = 10" ms! and R= | m. Case $ was solved using a mesh of 10% 10 and 20% 20 grid points (Fig. 4). The stability eriteria (17) required rnondimensional time intervals of Ax < 2.5% 10"? for the fist mesh and A < 6.25 x 10 for the second. In order to eliminate the possiblity of the temperature changing from below to above the melting tem- perature interval in one time step, ie, to enforce ‘energy conservation, the time intervals were reduced to r= 20x10 for the first mesh and to ‘Ar = 6.0% 10" for the second. Table 2 gives the results ofthe running time test using a turbo PASCAL 5.0 compiler. 5. CONCLUSIONS ‘The proposed numerical scheme is capable of solv- ing phase change problems with boundary conditions ‘of constant temperature and convection and/or radi- ation, Radiation and convection boundary conditions ccan easily be combined in the same problem. The resulting Scheme is simple and straightforward and requires neither iteration nor tracing of the interface location. Since the physical properties of the PCM— the heat capacity and the thermal resistance to heat flow of each element—are determined by its tem- perature, the proposed numerical method can deal ‘with more than one liquid/soli interface. Since the scheme is based on two temperature vec~ tots only, the properties of each node being deter- ‘mined atthe beginning of each time interval according to the previous temperature field, itis suitable for solving phase change problems using a micro- ‘computer with a limited dynamic memory. ‘Comparison of the proposed numerical solution with exact solutions and with other numerical solu- tions based on finite differences and finite elements showed good agreement, Thus, the proposed numeri- cal scheme is a simple and reliable tool for obtaining fast solutions to many phase change problems. ‘All the numerical solutions inthe present work were obtained with an IBM PC microcomputer using a turbo PASCAL compiler. The running time for each ‘numerical solution was afew minutes, REFERENCES. 1. V. J, Lunaedini, Heat Transfer in Cold Climates, Vas. [Nottrand Reinhold, New York (198).An efficient numeriel solution HLS. Carslaw and C.J, Jaeger, Conduction of Heat in Solids, Chap. {1- Caeendon Press. Oxford (1959), N'Shamsundat, Comparison of Numerical Methods for Diffaion Probleme with Moxing Boundaries in Moving Boundirs Problems (Edited by D.G. Wikon, A.D. Solomon and P.T. Boggs). Academic Pres, New York U7) R, Viskanta, Solar Heat Transfer Storage: Latent Heat Maveral (Edited by G. A. Lane), CRC Press, Flonda (1983), B. Carnahan, H. A. Luther and J. 0. Wilkes. Applied Mumercal Methods. Wily. New York (1969). Ar"Lazaridis, ® numerical solution of the multi- dimensional soldieation (or melting) problem. Int. J “eat Mass Transfer U3, 1859-1497 (1970), 83 1. J,S. Hsiao and B. T. F. Chuas, An efficient algorithm for finite clement solution t0 two-dimensional heat transier with melting and frezing, ASME S4¢HT-2 (388) 8. A.J Dathuifen and A. Sesal. Comparison of finite ‘lements techniques for Soliifeation probiems. fn ‘Numer. Meth. Engng 23, {807-1829 (1986). 9. M. Lees, A linear three-level diference scheme for quasilinear parabolic equations, Maik Comp. 20, 16-622 (1966) 10. K. Morgan, R. W. Lewis and 0. C. Zienkiewice. An improved algorithm for heat conduction problems with phase change, Int J. Numer. Merk. Engng 12, 1191- 1195 (1978)