Louvai n- l a- Neuve ( Bel gi um) Summary - A f ul l y aut omat i c pr ogr amgener at i ng t wo- di mensi onal Fi ni t e El ement (FE) gr i ds has been de- vel oped wi t h t he mai n pur pose t o deal wi t h under gr ound cabl e geomet ri es. The user needs onl y t o speci f y t he l ocat i on of t he va- r i ous obj ect s i n t er ms of ci r cl es and st rai ght - si ded pol ygons. A f ewgl obal par amet er s cont r ol t he whol e process. Node and el ement r enumber i ng f aci l i t y al l ow near- opt i mal bandwi t h or f r ont al sol ut i on i n t he f i ni t e el ement sol ver. STATI NG THE PROBLEM The ext ensi ve use of Fi ni t e El ement (FE) t echni - ques has demonst r at ed i t s per f or mance i n sol vi ng Par- t i al Di f f er ent i al Equat i ons i n mul t i mat er i al domai ns wi t h compl ex boundari es. Thi s makes FE met hod part i - cul ar l y sui t abl e f or t he st udy of t he t her mal f i el d wi t hi n and ar ound power cabl es buri ed i n non- uni f orm soi l . The pri ce t o pay f or t he ver sat i l i t y of t he met hod l i es i n t he amount of i nput dat a r equi r ed t o descr i be t he pr obl emproperl y. pr oper t i es of t he mat er i al s i nvol ved, t he t her mal l oads, t he boundar y condi t i ons and t he geomet r i cal l ayout i n t he f or mof a FE mesh. The mesh pr epar at i on, i f hand- coded, i s by f ar t he heavi est part of t he j ob. Al t hough t he same gri d can somet i mes be re- used f or a set of cal cul at i ons, t her e ar e cases wher e some part of t he geomet ry i s i t sel f unkwown ( l i ke : howmuch con- t rol l ed backf i l l i s needed t o i ncr ease cabl e ampaci t y by 10 %). An ent i r e f ami l y of si mi l ar gr i ds must t hen be used. buri ed power cabl es have char act er i st i cs t hat make t he use of common gener al purpose mesh gener at or s i mprac- t i cal or uneasy : hi gh number of subdomai ns, hi gh l evel of i ncl usi on, hi gh l engt h r at i o (up t o l o4) bet ween smal l est and bi ggest el ement s. mesh gener at or t hat i s especi al l y t ai l ored f or t he t her mal st udy of buri ed power cabl es. Among t hese dat a ar e t he t her mal Fur t her mor e, FE meshes f or t he t her mal st udy of Thi s paper pr esent s a dedi cat ed t wo- di mensi onal CHARACTERI STI CS OF THE FE MESH Some pr oper t i es of t he pr obl emhave a di r ect i mpact on t he FE mesh t hat must be gener at ed : - i t can be l i mi t ed t o 2-D pr obl ems - ci r cul ar cabl e boundar i es i mpl y 2nd- order i sopara- met r i c el ement s t o avoi d l ar ge number s of f acet s i n t hei r r epr esent at i on - t hi n ci r cul ar l ayer s ar e best covered by 8- node i so- par amet r i c quadr i l at er al s. IEEE Transactions on Power Delivery, Vol. 3, No. 1, January 1988 A 2-D FI NI TE ELEMENT MESH GENERATOR FOR THERMAL ANALYSI S OF UNDERGROUND POWER CABLES D. MUSHAMALI RWA N. GERMAY J .C. STEFFENS UNI VERSI TY OF LOUVAI N LABORELEC LABORELEC 36 SM400-6 A paper r ecommended and appr oved by t he TEFF. I nsul at ed Conduct or s Comni t t ee of t he I ZEE Power Engi nzer i ng Soci et y or Dr esent at i on at t he lF,EE/PES 1985 Summer Yeet i ng, Mexi co Ci t y, ' qexi co, J ul y 20 - 25, 1986. Febr uar y 3 , 1986; made avai l abl e f or pri nt i ng i pri l 23, 1956. Pr i nt ed i n t he U. S. A. Yani i scri pt submi t t ed B. P. l l Rhode- St - GenSse ( Bel gi um) Addi t i onal l y, a good mesh gener at or shoul d have a f r i endl y and power f ul user i nt er f ace, r educi ng i nput dat a t o a mi ni mumas cl ose as possi bl e t o t he physi cal probl emat hand. Fami l i es of si mi l ar meshes shoul d be obt ai ned by ver y smal l var i at i on i n dat a. The use of some wel l chosen gl obal par amet er s and act i ons shoul d be pref erred t o a post er i or i manual cor r ect i ons of t he mesh t o sol ve l ocal di f f i cul t i es. near l y si ngul ar or t wi st ed el ement s and hi ghl y con- nect ed nodes. The el ement and node number i ng shoul d be ar r anged as t o keep f ur t her FE cal cul at i ons t o a mi ni - The f i nal mesh must be wel l shaped, avoi di ng mum. The met hod must al so be ahl e t o t ackl e t he probl em of subdomai n i ncl usi on and t o r educe mul t i pl y connect ed r egi ons t o si mpl y connect ed ones. Speci al i nt er f ace management i s needed t o assess t he r egul ar i t y of t he mesh acr oss subdomai n boundari es. Because el ement si zes var y f r omf r act i ons of mi l l i met er s i n t he cabl es t o sever al met er s i n t he f ar soi l r egi on, t he mesh gr adi ng pr obl emmust r ecei ve spe- ci al at t ent i on. MAI N STEPS OF THE METHOD The r egi on under consi der at i on i s descr i bed as a set of embedded obj ect s (i .e. cl osed domai ns or hol es def i ned wi t hi n an uni que boundary), t hat can be ci r- cl es, ci r cul ar sect or s or gener al pol ygons ( curve- si ded pol ygons under study). each ot her but t hey can have common edges or si des ( t angent domai ns). di vi ded i nt o t went y- t hr ee such obj ect s. These domai ns may not over l ap Fi gur e 1 shows a backf i l l r egi on - Fr omt he "i ncl usi on t ree" ( see f i gur e 2) and t angen- cy i nf or mat i on, mul t i pl y connect ed subdomai ns ar e det ect ed and di vi ded i nt o si mpl y connect ed zones by st r ai ght cut l i nes. r i t hm[l ] i s used t o keep t he number and t he l engt h of cut s t o a mi ni mumand make sur e t hat al l r esul t i ng A "mi ni mumspanni ng t r ee" al go- subdomai ns are si mpl y connect ed. At t hat moment , t he domai n under st udy i s r epr esent ed by a mosai c of gener al i zed pol ygons wi t h st r ai ght or ci r cul ar si des. 2. CONCRETE SLAB 3. BACKFILL I r 6 PROTECTING APE 7 9 CABLE SURFACE 12 GWBbL INSLATION 1 \ PHASE INSULATION Fi g. 1 : Si mpl y connect ed subdomai ns i n t he t her mal backf i l l regi on. Thr ee- phase cabl es wi t h f or ced cool i ng pi pes. The soi l i s not repre- sent ed here. 0885-8977/88/0100-0$01.0001988 IEEE 63 dar i es ar e t hen t r i angul at ed usi ng an al gor i t hmdeve- l oped by Al l an Geor ge [Z]. Ot her al gor i t hms are appl i ed i n speci al cases as : - The si mpl y connect ed subdomai ns wi t h segment ed boun- Far yrl rqlron Concrete slab Ther m1 backfill . mooncrescent - l i ke gaps bet ween eccent r i c ci r- cul ar domai ns ( cabl es i n duct s) bl es) conductors). Heot wnks . concent r i F ci r cul ar l ayer s ( si ngl e- core ca- Protecting pipes . mul t i - core cabl es ( round or sect or- shaped Gable surface blobol insulotion The r egul ar i t y of t he mesh accr oss t he subdomai ns i s guar ant eed by t he uni que r epr esent at i on of t he sub- Pho.se ,nSulat,On domai ns boundari es. Fi g. 2 : I ncl usi on t r ee f or a mul t i - core cabl e conf i - gur at i on l ai d i n backf i l l , as shown i n Fi g. 1 A geo- - Each si de- ar c i s t hen di vi ded i nt o segment s. met r i cal pr ogr essi on bet ween segment si zes i s al l o- wed, wi t h l i mi t ed val ues f or t he r at i o and maxi mum segment si ze. t o keep t he endi ng segment s of connect ed si des to compar abl e si zes. t he l owest possi bl e number of segment s i s generat ed. Speci al adapt at i on al gor i t hms are used Wi t h t hese const r ai nt s i n mi nd, - A qual i t y f act or [3] of t he t r i angl es i s comput ed t o eval uat e t he qual i t y of t he gener at ed gri d. l engt hs a, b and c, t he qual i t y f act or qf i s def i ned by : For si de qf =8 (s- a) (s- b) (s-c)/(a.b.c) wher e s = 0.5 (a +b +c). Smoot hi ng al gor i t hms can be appl i ed t o t he gener at ed gr i d, ei t her t o mer ge t wo adj acent t r i angl es and re- di vi de t he quadr i l at er al obt ai ned al ong t he ot her di agonal , or t o move a node t o t he "cent er " of t he pol ygon f or med by i t s nei ghbour i ng t r i angl es (Fi g. 4) Fi g. 3 : Some exampl es of gener at ed gr i ds f or si mpl e regi ons. (a) Backf i l l around 6 si ngl e- core cabl es ; 4 st r ai ght cut l i nes di vi de t he r egi on i nt o 3 subr egi ons whi ch ar e (b) Mooncr escent - l i ke ai r gap bet ween ext er nal cabl e sur f ace and i nt er nal pi pe surf ace. (c) Si ngl e- core cabl es ( concent ri c ci r cul ar l ayers), hal f and whol e cabl e regi on. (d) 3- core bel t ed cabl e wi t h round conduct or s. (e) 3- core bel t ed cabl e wi t h sect or shaped conduct or s ( one i nsul at i on l ayer around each conductor). separ at el y t ri angul at ed. Fi g. 4a : Redi vi si on of a quadri l at eral . Fi g. 4b : Move of a node t o t he "center' * of t he sur r oundi ng pol ygon. - Once t he gri d i s f i xed, an el ement / node r enumber i ng scheme [4], [ 5] i s appl i ed t o mi ni mi ze bot h t he band- wi t h and t he f r ont wi dt h i n t he FE sol ver program. Cal l i ng "adj acent nodes" t hose whi ch shar e a common el ement , t he r eor der i ng al gor i t hmwhi ch uses t he mi ni mumf r ont gr owt h cr i t er i on may be summar i zed as f ol l ows : - St ar t wi t h one node whi ch wi t h i t s adj acent nodes def i nes t he set of act i ve nodes ( current front). Al l ot her nodes are i nact i ve. - Exami ne t he cur r ent l y act i ve nodes and sel ect t he one whi ch has t he f ewest i nact i ve adj acent nodes. Rel abel t hi s node, del et e i t f r omt he set and add al l i t s i nact i ve nei ghbours. Repeat t hi s st ep unt i l al l nodes have been rel abel l ed. - Sel ect anot her st art i ng- node and repeat t he pro- cess f r omt he begi nni ng. - Each st art i ng- node pr oduces a number i ng st r at egy and t he best r eor der i ng i s t hat whi ch gi ves t he l owest maxi mumsi ze of t he set of act i ve nodes. - Af t er t hi s node r enumber i ng, t he el ement s are r eor der ed i n an ascendi ng sequence of t hei r l owest numbered nodes. Thi s preserves t he opt i - mumel i mi nat i on or der as cl osel y as possi bl e. To choose t he st ar t i ng- nodes, Sl oan [ 5] uses an aut o- mat i c met hod t aken f r omgr aph t heory ; t hese are at t he ends of t he l ongest pat hs (i n t erms of el ement si des) cr ossi ng t he gri d. GOVERNI NG PARAMETERS ---- As was sai d bef or e, aut omat i c mesh generat or gover ned by a f ewpar amet er s, chosen i ni t i al l y by t he oper at or , has been pref erred t o an i nt er act i ve met hodo- l ogy wher e t he oper at or woul d modi f y l ocal l y t he gri d. I n t hi s way, t he user needs not even t o know t he exact node or el ement number s and t he gri d pr epar at i on can even be made i n bat ch mode. i n sever al di r ect i ons : User' s act i ons can proceed - The si ze of el ement s i n a domai n can be l i mi t ed by set t i ng a maxi mumval ue t o t he segment si ze and/ or geomet r i cal pr ogr essi on r at i o al ong i t s boundar y (Fi g 5). The Al l an Geor ge al gor i t hmwi l l propagat e t hi s si ze f r omt he boundar y i nt o t he i nt er i or of t he domai n. Bot h segment and r at i o l i mi t s can be gi ven separ at el y f or each obj ect but f or a gi ven si de, t he most const r ai ni ng l i mi t s of adj acent subdomai ns are used. Fi g. 5 : Tr i angul at i on of t he backf i l r egi on of a symmet r i cal si x- core cabl e conf i gurat i on. The maxi mumr at i o was f i xed t o 2, t he maxi mum si de l engt h t o 7 cmand t he t r i mmi ng angl e t o 70" . - Pr oxi mi t y of smal l and l ar ge segment s i n a subdomai n boundar y can happen wi t h ver y t hi n subdomai ns. user can add cut s bet ween obj ect s ei t her bef or e or af t er t he connect i vi t y r educt i on ( f i gures 6). These cut s "pr opagat e" smal l segment s to t he opposi t e boun- dar y and avoi d l abyr i nt hi c subdomai ns, i mpr ovi ng t he subsequent t r i angl e qual i t y. The (b) Fi g. 6 : Addi ng cut l i nes (b) i mpr oves t r i angul at i on of t hi n subr egi ons (a) - The maxi mumangul ar span of ar cs al ong a cabl e boun- dar y can be speci f i ed t o ensur e t hei r proper repre- sent at i on wi t h par abol i c curves. Thi s par amet er can be separ at el y speci f i ed f or each cabl e. - Wi t hi n each subdomai n, t he Al l an Geor ge al gor i t hmcan be i nf l uenced by modi f yi ng t he "t r i mmi ng angl e" l i mi t : at any st age of t he t r i angul at i on, a cor ner i n t he cur r ent boundary showi ng an angl e bel ow t he t r i mmi ng l i mi t causes a t r i angl e t o be f or med i mme- di at el y ( f i gure 7). I ncr easi ng t he t r i mmi ng l i mi t decr eases t he over al l number of el ement s. Fi g. 7 : Tr i mmi ng oper at i on causes f or mat i on of t r i angl e P2 P3 P4 f r omt he cur r ent boundary sect i on Pi , P2, P3, P4. 65 I - After the generation, the grid smoothing procedure is controlled by specifying : . the admissible triangle quality factor above which the smoothing procedure is in any case stopped vided. In practice however, this procedure stops when quadrilateral redivision does not improve the grid any more . the minimum displacement between a corner node and the center of its surrounding polygon. This avoids endless microscopic corrections in the re-centering process. . the maximum number of quadrilaterals to be redi- I I 1 I t NUMERICAL EXAMPLES k$TD APPLICATIONS The procedure described above is an interesting interface to a general purpose thermal FE package accepting second order isoparametric triangles and quadrilaterals. of the thermal field inside and outside power cables in many configurations. We report here the results of two cases that were already used :or reference calculations in Working Group 02 of CIGRE Committee 21 [6, 71. They refer to two superimposed layers of 3 cables in flat formation (Fig 8a), and two trefoils placed side by side in the same trench (Fig 8b). Both cases concern single-core oil filled cables. For the double layer, the cables are placed in ducts. Because of symmetry, the grid covers only one half of the domain, which is limited to a rectangle of 15 m depth and 40 m width. isothermal, and the other boundaries are adiabatic. Experience has shown that these boundary conditions and domain dimeqsions result in negligible error on the cable temperature. Physical and geometrical data of the cables It has been used for the computation The soil surface is kept We used oil-filled (OF) cables, 1 x 930 m 2 , 400 kV with circular copper conductor. gathered in the following tables : Table 1 : Physical and geometrical data of the cables, The data are the thickness of the pipe has been neglected. Element I External dia- meter (mm) Thermal re- sistivity (K.m/W) Oil duct Copper conductor Semiconductive conductor screen Paper dielectric Semiconductive screen Lead sheath Polyethylene protective covering Pipe interior (air gap) 18 39.3 40.5 81.75 82.75 89.75 100 120 6.00 0.00268 5.00 5.00 5.00 0.02857 3.50 0.20 I 1 - Table 2 : Heat losses in W/m, for a rating current I =114 A for the double layer system and 261 A for the trefoil system. I Superposed systems I Trefoil systems I I Conduct or Dielectric Lead sheath 1.26 1.86 10.66 1.20 I -- 1 Fig. 8a : Cable layout : two layers af single-core cables in flat formation. I Fig. 8b : Cable layout : two trefoils of single-core cables in the same trench. Finite element grid To generate the FE grid, one must give : . the X, Y coordinates of the polygonal ver- tices (far soil, backfills) and the centers of the cables tices) and their layers) the tangency and cutting information (speci- fying only the code number of the objects concerned) the physical and thermal data of materials (thermal resistivity and losses) the global parameters to control the genera- tion (maximum geometrical progression ratio, trimming angle, ...). . the definition of polygons (sequbnce of ver- . the radii of all circular contours (cables . the inclusion tree of all objects . . . Fig. 9a and 9b show the generated grid of the far soil, backfill and cable zones for the configuration of Fig. 8a. The whole grid has 826 elements and 1983 nodes (including the mid-side nodes). The "near- optimal" front has 112 nodes. The far soil grid of the trefoil system installation is quite similar to that of Fig. 9a. The backfill and cable regions are shown in Fig. 10. In this case, 771 elements and 1946 nodes have been generated and the front length is 81 nodes. 66 Cabl e nr Doubl e l ayer Doubl e t r ef oi l 1 2 65. 369 - 3 63. 542 4 70. 429 5 66. 935 68. 660 6 65. 164 69. 278 - - - - * Fi g. 9a : Tr i angul at i on of t he f ar soi l r egi on doubl e l ayer of cabl es. of t he Fi g. 9b : Backf i l l and cabl e mesh, doubl e l ayer syst em. Fi g. 10 : Backf i l l and cabl e mesh, doubl e t ref oi l syst em. Ther mal cal cul at i ons The t her mal f i el d ( t emper at ur e at each node of t he gri d) has been comput ed supposi ng t hat al l t he cabl es i n t he conf i gur at i on di ssi pat e t he same heat as spe- ci f i ed i n t abl e 2. The t emper at ur e val ues of t he cabl e cent er s and copper conduct or ar e r epor t ed i n t abl e 3. Tabl e 3 : Cor e t emper at ur es of t he cabl es ("C) ( mi ssi ng val ues ar e obt ai ned by symmet ry) Fi g. 11 : Ther mal f i el d, doubl e l ayer syst em. - Fi g. 12 : Ther mal f i el d, doubl e t r ef oi l syst em. COMPUTER PERFORMANCES Tabl e 4 gi ves r epr esent at i ve t i mi ngs (i n seconds) f or t he var i ous st ages i n t hi s pr ocess, f or t he t wo appl i cat i on exampl es, i n t he act ual r un of t he program ( approx. 7800 l i nes), on a I BM4381 (VM/CMS) t i me shared syst em, act i ng i n bat ch mode. The memory si ze needed i s near t o 400 Kb f or gri d si ze up t o 2000 nodes and 1000 el ement s, al l i nt eger var i abl es bei ng 2 byt es l ong. 67 N. GERMAY, D. MUSHAMALI RWA : A compar i son of f i ve met hods t o cal cul at e t he t her mal f i el d ar ound a gr oup of under gr ound cabl es, SRBE - KBVE Revue E ( Act a Techni ca Bel gi ca), Vol . 99, no 3/4 (1983), 156 - 180 The cal cul at i on of t he ef f ect i ve ext er nal t her mal r esi st ance of cabl es l ai d i n mat er i al s havi ng di f - f er ent t her mal r esi st i vi t i es, El ect r a Nr 98 (J an. 1985), 19 - 42 Tabl e 4 : CPU t i mes i n seconds f or t he di f f er ent st ages of t he mesh gener at i on Two superpo. sed syst ems f or mat i on Boundar y di vi si on Far soi l t r i angul at i on Backf i l l t r i angul at i on Cabl es (+ev. ai r gap) Node h el ement smoot hi na Node compact i on Node r esequenci ng El ement r esequenci ng 1.38 1.58 2.03 4.39 1.63 1.42 10.30 0.88 Two hor i zont al t r ef oi l syst ems 0.82 2.41 1.63 3.32 1.78 1.32 8.08 0.81 1 I Tot al I 23.61 I 20.17 I I I 1 Most of t he t i me (* 50 %), i s consumed by t he node r enumber i ng pr ocess, but t hi s st ep i s perf ormed onl y once and when t he gener at ed mesh i s accept abl e. CONCLUSI ON We exper i enced t hat t hi s pr ogr amhas cut down f r om many days to a f ewhour s t he del ay of handl i ng f i ni t e el ement gr i ds f or underground cabl e probl ems. Thi s one i s now essent i al l y r educed t o t he t i me needed f or col l ect i ng t he physi cal and geomet r i cal basi c parame- t er s of t he pr obl emand ent er i ng t hemi nt o t he com- put er. The pr ogr amcan al so be used f or many 2-D f i ni t e el ement pr obl ems, ot her t han cabl e heat i ng. I mpr ovement s ar e pl anned i n t wo di r ect i ons : - t o al l owt he mi xi ng of r ect i l i near and cur vi l i near ( parabol i c) boundar i es i n a gi ven pol ygon - t o al l owt he i nser t i on of pre- meshed subdomai ns ( t y- pi cal l y t he i nsi de of cabl es) t hat coul d be st or ed i n a dat a bank. The advant age of t hi s l ast t ool must be bal anced wi t h t he speed of t he pr ogr am, as i t i s now, and t he possi bl e non- r epeat abi l i t y of t he cabl e geomet r i es t o whi ch i t i s appl i ed. REFERENCES A. NI J ENHUI S and H. S. WI LF : "Combi nat or i al Al gor i t hms", Academi c Pr ess, 1975, pp. 232 - 237 A. GEORGE : An Aut omat i c Two- Di mensi onal Domai n Tr i angul at or , i n : "Comput er i mpl ement at i on of t he f i ni t e el ement met hod", Ph. D. di ssert . , Depar t ment of Comput er Sci ence, St anf or d Uni ver si t y ( March 1971), 222 pp. D. A. LI NDHOLM : Aut omat i c Mesh Gener at i on on Sur f aces of Pol yhedr a, I EEE Tr ans. on Magnet i cs, V o l . MAG- 19, no 9 (Nov. 1983) J .P. KI NG : An Aut omat i c Reor der i ng Scheme f or Si mul t aneous Equat i ons der i ved f r omNet wor k Syst ems, I nt. J . Num. Met h. Engng, Vol . 2 (1970), 523 - 533 S.W. SLOAN, M. F. RANDOLPH : Aut omat i c El ement Reor der i ng f or Fi ni t e El ement Anal ysi s wi t h Fr ont al Sol ut i on schemes, I nt . J . Num. Met h. Engng, Vol . 9 (1983), 1133 - 1181 ' D. MUSHAMALI RWA was born i n Nyakal onge, Za' i re, on J ul y 30, 1953. He gr aduat ed i n El ect r i cal Engi nee- r i ng at t he Uni ver si t y of Za' i re i n 1978. Af t er a t wo- year t r ai ni ng cour se on Power Syst emAnal ysi s at t he Cent r o El et t r ot ecni co Speri men- t al e I t al i ano, CESI , Mi l an, I t al y, he j oi ned t he Techni cal Depar t ment of t he Nat i onal El ect r i c Company, %i re wher e he was concer ned wi t h power syst emt r anspor t and di st ri - but i on. Pr esent l y he i s pr epar i ng - a PhD t hesi s at t he Uni ver si t y of Louvai n about var i ous t her mal anal ysi s met hods f or under gr ound power cabl es. N. GERMAY was bor n i n Namur , Bel gi um, on Apri l 14, 1925 and re- cei ved hi s El ect r i cal Engi neer de- gr ee at t he Uni ver si t y of Li ege i n 1948. He wor ked successi vel y i n nucl ear engi neer i ng and cabl e manu- f act ur i ng at t he At el i er s de Cons- t r uct i ons El ect r i ques de Char l er oi , f r om1950 t o 1963, wi t h an i nt er- medi at e af r i can mi ssi on i n Za' i re, f r om1956 t o 1960 as del egat e of t he - bel gi an cabl e manuf act ur er s. Si nce 1963, he i s head of t he Mat hemat i cs Depar t ment of t he l abor at or y of t he bel gi an el ect r i cal ut i l i t i es, LABORELEC. He was part - t i me pr of essor at t he Uni ver si t y of Ki nshasa, i n 1959-1960 and i s now l ect ur er i n el ect r i cal engi neer i ng at t he Uni ver si t y of Louvai n si nce 1968. J . C. STEFFENS was bor n i n Mal mgdy, Bel gi um, on Sept ember 4, 1947. He gr aduat ed i n Appl i ed Mat hemat i cs at t he Cat hol i c Uni ver si t y of Louvai n, Bel gi um, i n 1971. He j oi ned t he mat hemat i cal depar t ment of LABORELEC, wher e he i s concer ned wi t h a great var i et y of pr obl ems, r angi ng f r omt her mal anal ysi s of power cabl es t o t or si onal f at i gue of Turbo- Generat or shaf t s. Discussion G. Luoni (Pirelli, Milan, Italy): Field calculation, and particularly ther- mal field calculation with f i i t e element techniques, is nowadays widespread and generally available for research purposes, and it is not very demanding on computers. Yet, in order to transform it from a scien- tific tool to an everyday engineering instrument, we need a development in input data techniques, that is, we need an automatic mesh generator. In this respect the paper by Germay and colleagues is certainly a major step forward. I have noticed that many important practical techniques have been incorporated into the algorithm, and I suppose that node renumbering is a very important factor in the performance of the program. a) The programmer enters some data, such as cut lines, which require an interchange of information back from the mesh generation Some clarifications may be of interest: 68 algorithm. How is this obtained? Only with graphical output or also D. Mushamalirws: Wethank the discussers for their enlightening obser- with other moresynthetic means? vations and comments. Answers to Mr. Luoni are as follows. b) Which is the criterion for accepting a given quality factor? Is it some a) Becausethe program was designed to work in batch mode as sort Of Overall quality Of a mesh, Or the minimum quality fat- as interadvity, no real feedback other than the final graphical out- put is provided to the user. Data for userdefied cut hes are entered tor of a singletriangle? An apparent limitation of the programis that it is dedicated to a par- ticular geometry. I believethat this is not a real limitation sincestraight lines and circles constitute a good deal of our engineering shapes. The data lacking is the inclusion tree for applications different fromcables, but this is, in my view, morea problemof data than a problemof programming. I congratulate the authors for the very interesting work. Manuscript receivedAugust 4, 1986. George J. Anders (Ontario Hydro, Toronto, Canada): Generation of a finite element grid has always been one of the most difficult problems when this advanced numerical technique has been applied to solvecom- plex partial differential equations. Usually, the user is required to pro- videcoordinates of all the nodal points and designation of element con- ductivities, boundary conditions for each boundary element, and heat generation per element. Preparation of this kind of data is very timecon- suming and prone to errors. Theauthors are to becongratulated fpr attempting to provide the user with an easy-to-usemesh generator. For the programof this typeto beuseful for practicing engineers the output files should providean easy interface with a solver routine. Could the authors briefly describethe format of the output produced by the pro- gram? In the examples presented in the paper the authors showed quite elaborate mesh generated in the air space between a cableand the re- taining duct. Oneof the problems associated with the useof a f i te ele- ment techniquefor a study of cables in ducts is that the thermal resistivity of air between the duct and the cableis temperature dependent. Could the authors explain how they envisageentering, at the beginning of the programrun, the thermal resistivity data for this part of the thermal network? Another problemis related to the fact that the finite element technique, as it is used today for the study of thermal fields in the vicinity of cables in ducts, assumes that the heat conduction is the only mode of heat transfer in the entire region of study. Th i s assumption isobviously not correct for the air space between the cableand retaining duct. This implies that it is probably moreaccurate to represent ducts as heat flux boundaries with suitableheat flux values without a need for an elaborate representation of the internal duct structure. Author?comments on this subject will beappreciated. And f dy , a few comments on the numerical examples presented in the paper. This discusser disagrees with the state- ment on page4 of the paper that the isothermal boundary condition at the soil surface leads to accurate computational results. As a matter of fact, the soil surface is usually not isothermal and convectiveboundary provides a much better representation of the actual situation. Someof the inputs required for the numerical examples presented in the paper are heat losses in the conductor, dielectric losses, and sheath losses. Could the authors explain how they obtained the values shown in Table 2 with such accuracy? Also, the thermal resistivity of air shown in Table 1 is an order of magnitude too low. Manuscript receivedAugust 8, 1986. in a simplemanner: only the code numbers of objects to belinked must beprovided. Theprogramlocates automatically cut ends to keep its length to a minimum. b) A quality factor is associated with each individual triangle. The user defines a minimal valueunder which no triangle should beaccepted. Thequality factor is then used to control the grid smoothing pro- cess: the worst triangles are corrected fiist, until all of themhavea factor abovethe prescribed criterion. c) Concerning the last remark, weagreethat the concepts of object and inclusion are quite general and can beapplied to many other two-dimensional engineering shapes. Extension to generalized polygons with a combination of linear and curved sides is presently investigated. Concerning Mr. Anders questions, our answers are as follows. Because the generator has been developed in conjunction with LABORELECs FE programfor thermal analysis, mesh data are for- matted for that program. However, conversion to any other format is no real problem, becausedata are similar for any FE solver. Material properties, boundary conditions, and heat loss values are not needed by the generator, which works only with code numbers. In this respect, thermal field with soillair convection is as simpleas changing the data associated with the soil/air boundary codein the FE solver. Actual parameter and heat loss values for the calculations presented in the paper are derived fromthe IEC 287 normand cable manufac- turers data, with the aim of comparing the results with other simplified methods. Weagree that the statement in page4 is am- biguous: in fact, accuracy is preserved by adiabatic conditions on the far soil boundary when compared to infinite soil, not by the isother- mal soivair interface. Cabletemperatures haveindeed been calculated to belowered by about 3 OK when computed with isothermal soil sur- facecondition instead of taki ngaccount of realistic convectivetransfer coefficient ( 5 W/K.m2). Theair gap in ducts poses a difficult problem, which has been final- ly circumvented in a rather crude manner: using the resistanceof the air gap as defied in the IEC 287 norm, the resistivity was adapted to obtain the corresponding mean valueof temperature difference between cableand duct. This procedure led to an actual valueof air resistivity much higher than shown in Table 1 (13 K.m/W). In our opinion, replacing the air zoneby heat flux boundaries does not really simplify the problemdue to the difficulty of determining the boun- dary flux values if the cableis represented. Thesituation becomes even totally untractable for transient calculations. Wethi nkthat much work remains to bedone in that field. Theconductiveapproach re- mains totally valid, however, when the air is replaced by semiliquid filling (bentonite) with known thermal properties. Manuscript receivedSeptember 11, 1986.