Assessment of numerical sirnulationof industrial glass melter

M.c. CARVALHO, M. NOGUBIRA and WANG JIANMechanical Engineering Departmentlnstituto Superior Tdcnico I Technical University of LisbonAv. Rovisco Pais, 1096 Lisboa Codex, Portugal

Abstract :

A mathematical model for the integrated prediction of the performance of whole glass rank furnace is described. itcomprises sub-models for the combustion cbamber, thc batch melting and the glass tank. The three sub-models were stuciied, bya cyclical iterative way. Useful information abour lurnace can be supplied by the model as long as the inler and boundary

condirion are given. In this paper, model tesring againsr data available under the scope ol the ICC-TC2l activities is described.

Using of 3D modelling in design is discussed.

IntroductionModel assisted design\ /f odel based integrated designIYI tools may handle withgeometrical, operating and controllingparameters as a basis to analyse theperformance, in a rapid prototypingapproach, of a large number of tech-nological solutions, Optimisation offurnace operation conditions (mi-nimising the cost of poliutant emissionscontrol and maximising product quality,Ioad flexibility, energy efficiency,process controllability, equipmentreliability) is not possible in reai timewith real costs in basis of theconventional methodologies used at thedesign stage, New geometrical anange-ments, new control strategies, newoperating targets and variables have tobe arranged in an integrated approach.Integrated design of the industrialiurnaces will benefit from recentdevelopments in heat transfer 3D anddynamic modelling allowing toovercome conventional techniques,empiricalrules and very simplified heattransfer analysis which continue to beused in industrial environments.

The use of modelling for glassmelting fumaces may provide detailedknowledge of complex phenomena

occurring in this industrial equipment.Application of this knowledge to thestudy of new design and oPerationconcepts will allow oPtimisation,integration and intensification.

Model based integrated design is a

powerful tool enabling a detailedevaluation of the equiPmentperformance reliabilily, maintenanceand product quality at design stage'Physicaliy-based modelling wiil allowgeometry, controI and oPerationparameters unified treatment, without

neglecting the decisive effects oi theinteraction of several equipmentcomponents. This will allow the studyof a very large range of originalconception strategies enabling the set-upof non conventional solutions andstrategies.

In the present paper a 3D modelli,rgtool testing is reported, The:inalobjective of the present work is to makethis modelling tool capable to be rncludedin the integrated design automatic ioopproposed in figure l.

Figure 1 : Proposed integrated structure for model assisted conceptionof glass melting furnaces

Jndu!!rl6lFurnaco

,@lEoouona ISbdond Md ConEalns l)abPhysloal b.bFlcitd Mos.uo! Doh

doutauoh d t{oswm€n! Daq B6t.6--;;;;; to.iG.Gti lqulr tv' Efnclcncv' s efcrvSrl€lAcadon ot tho Ps's6Ert w dauont

Dvnrrnlo ^tod.l!'dB PcE.xFEn SYtEh

Ind.lorilon! altou! d6tl gnbdffi66t rnd

!Ebla otFradorr domaln

M.G. Camalho, M. Nogueira & Wang Jian

Modelling techniques, based inCFD concepts, have been developed inrecent years. Physicaily-basedmathematical prediction of fluid flowand heat lransfer inside industriai scalefurnaces is nowadays possible with anaccuracy level suitabie to be used indesign of new geometry arrangements,operating conditions and controlstrategies. Fundamentally, two phy-sically based modelling approaches willbe employed :

- 3D modelling procedures in whichfluid flow and heat transfer arepredicted in detail for centralcomponents of the system, forstabilised stationary operatingconditions;

- zero/one-dimensional dynamic mo-deliing in which the time-dependenrbehaviour is calculated for the wholefumace system.

To support the utilisation of rhesemodelling techniques, parallelcomputing systems can be used to makepossible fast processing and consequentiarge number of optimisationparameters. Strong man-machineinterfaces usable in industrialenvironment and the installation ofautomatic procedures to control themodel based optimisation process maymake modelling tool exploitable by non-research engineers. The use of thesemodelling procedures to test geometricaldimensions, operating set-points andcontrol parameters, requires the presenceof optimisation algorithms to handlewith such multi-input/multi-outputproblem, The complexity of the problemis enlarged by the number oftechnological constraints whichnormally have to be considered,Dedicated optimisation algorithms will

be specially developed to take profit fromthe capacities of parallel processing.Those, coupled wirh strong man_machine interface s may turn possible thedevelopment of model based designstation for optimised furnace conception.

Developed 3D moclelling toolsCombustion chamber model, barch

melting model and glass rank modelhave been individually developed I I -5 ] .

Each of these model componentsmay not correctly simuiate indusrrialprocess of glass melting. A newgeneration model able to predict rhemain thermal physical phenomenaoccurring in whole glass tank furnace isrequired. The present paper describes apackage model which includes three sub-models: the combustion chamber, thebatch melting, and the glass rank, ableto simulate the combustion process,

Figure 2 : Predicted temperature field in thecombustion chamber

Ternperature (C)

Figure 3 : Predicted flow pattern in the combustionchamber

FLOW PATTERN1961.t 490.1770.l66l,r 63e.t 4t2.1u 93,tt74.I 0t5.s35.g7c.

Figure 4 : Predicted flow pattern in the combustionchamber

FLOW PATTERN

Figure 5 : Predicted NO distributioncombustion chamber

Mass Fraction

@NO

casl to.st.m0.qtltffitr.6Hl.€lu

in the

4 /' Z '/ ' /==-.-1:/

'. '-

:==r-=.-_:\-\ \ \+=\\rl---\ \ \. L

-*-\ \ \t l, ./.

%r'r'->l2/,< -_ - /, Ii->=:-Z I

Ntrrtericul sirttttlarion of industtial g,lass mett(l

Irigure 6 : Predictedc0mbustion chamber

Ot distribution in the

Fraction

l'igure 7 : Predicted heat flux lrom the combustionchamber to the top surface of batch blanket andfree surface of glass melt

n,l:!31

{r:1. ..,r1 0.tll:a

ffiil*

\{ass Net Rad. Flux k\l/rn?,.,,,:.;,i":iilli*l\'{l r: r

ffiti

Figure 8 : Predicted temperature field in theglass tank

,jemperature (C)r0lt-l(Jt ...rN-r...,:.::.ri ll.t. .Of..XN I'r:-6:+ylj tu ,'. >t{N\,.F\+rli4idii, r rn,ffii#, <--iir

ffi#'

Figure 9 : Predicted temperature field in theglass tank

Temperature (C)t5a2-I 539.r 495.| 45?.r409.r 366.I 323.t260.l ?3?.L l94.tr7l.

Figure 11 : Sketch ofend'fired glass tank furnaceFigure l0 : Predicted flow field in the glass tank

FLO1Y PATTIRN TI

M.G. Canallto, M. Nogueira & Wang Jian

batch melting process, and the flow andhear transier of the molten glass in glasstank furnace. The three sub-models werestudied by a cyclical iterative way.

Numerical simulationMathematical modelling

Individual mathematical models(combustion chamber, batch meltingand glass tank) have been alreadydescribed by previous publication.tl,4,5.| The present paper willconcentrate in the coupling procedure,A package modeI, a three-dimensionalmathematical model simulating thephysical phenomena occurring in aglass tank furnace, has been developed.The model is based on the solurion ofconservation for mass, momentum,energy and combustion relatedchemical species, and comprises threemain coupled sub-models: thecombustion chamber, the batch melting,and the glass tank.

The combustion model incorporatesphysical modelling for the turbulentdiffusion flame, soot formarion andoxidation, NO formation anddissociation ard radiative heat transfer.The time-averaged equations for theconservation of momentum were used aswell as the equation for the conservationof energy. The two equation turbulencemodel k-e was considered appropriated.The combustion model is based on theideal of a single step and fast reaction,together with a statistical approach ableto describe the temporal nature of theturbulent fluctuations. The discretetransfer radiation model was used in thisstudy. []

The batch melting modelincorporates physical modelling for theheat transfer and melting down process.The chemical reaction was assumed totake place at a delined temperature. Thephase change was presumed to occurover a temperature range betweenchemical reaction temperature andmolten temperarure. 14,6,'71

The glass tank model incorporatesphysical modelling for the flow and heattransfer of the molten glass. Theradiation transfer inside the molren glassbulk was handled by using an effectivethermal conductivity in the 6nergyequation. [51

Coupling algorithm andnumerical solution

The simulation of the whole glassfurnace was performed in an integratedmethod. The three main sub-modelswere coupled by a cyclical iterative way,matched by the relation between rhe heatflux from the flame to the molten glasssurface and the top surface of the batchblanket, and the heat fiux from rhemolten glass to the bottom surface of thebatch blanket. The whole procedure iscalculated until <convergence> of thecoupled process is achieved, Interfacetemperature between the flame and thebatch, the flame and the molten glass,the molten glass and the batch werecalculated by a cyclical iterative way.

The finite difference/finite volumemethod was used to solve the equations.As the upstream conditions determinethe downstream ones on the batchblanket, a matching downsrreamtechnique in its finite difference formwas used to the batch model. [4]

Results and discussionThe above referred model was

applied to an end-port container glasstank furnace - ICG-TCZl Robin TestCase. Melting area of glass tank was6l square meter. The pull of the tankwas 175 ton each day. The flow rate offuel oil was 755 kg per hour and inlettemperature of the combustion air was1180 "C. For the batch blanket, rawmaterials constitute 55 o/oby weight andcullet 45 Vo by weight, Thermalphysical properties of the glass batchwere calculated using the contentsavailable in [8].

Resuhs for rhe combusrion chamber.the batch melting rate and rhe glass rankare presented in figure 2 ro figure 10.The results obtained from rhe modelwere in good agreement with rhoseobtained from the measuremenl on anop€rating fumace ( rable I and figure I I ).

The time averaged predicredtemperarure field of the gas [urbulenrflow field is ploued in figure 2. The hightemperature region, near the llame fronr,is well visible in this figure. Oprimaltime averaged location of the gastemperature zones may be found usingvaiidated and coupled 3D modellingprocedures as the presenrly proposed.For instance, further atrenuation ofeventually excessive temperature peaksis possible acring on the fuel distriburionamong burners or on design paramerers.A significanr feature of rhe calculatedfield is rhe iocarion and magnirude oftemperature gradients, apparent in rheair/fuel flows shear layer. The calculatedtime averaged flow field is plorted infigure 3 and figure 4. The horse-shoeshape of flame is well apparent in figure4 as well as the effect of the secondaryrecirculations caused by rhe vertical andhorizontal inclinations of the air in.ierduct and burner, In figure 5 NO massfraction in the hot gases inside rhecombustion chamber is plotted. In rhisfigure the NO dissociarion is wellevident. In fact, a NO concentration peakmay be observed between the air inlerport and the side wall, and imporranrdecreasing gradients towards the outlerregion may be observed. In rhis regionstrong oxygen concentrations may befound together wirh high remperarurelevels due ro the flame presence. Thisaspect may be illustrated by rhe

Table 1: Measured temperature and calculated results

CalculationTemp6rature

AT

TI t525 "C 1512 0C t3 0c

T2 1595'C t5B9 "C 60c

T3 1340 0c i33l oc 90c

Numerical simulation of industrial glass melter

distribution of the oxygen, plotred infigure 6. In this ligure a horizonral planecutling the air entrance is used.

The batch is fcd into glass tank bybatch charger from the doghouses whichlocated at both side walls. The batchblanket is heated from top surface byheat radiation from thc combustionchamber and heated from bottom surfaceby heat transfer from hot molten glassunderneath, When the temperature isabove molten temperature, the batch isconverted to molten glass and meltingdown into glass tank. The heat flux fromthe combustion chamber to the surfaceof the molten glass and the top surfaceof the batch blanket is shown in figure7. The heat ilux was calculated bycombu.stion model by a cyclical irerativeway. The values of the heat flux are verylarge in the surface oI the batch blanket,because the temperature of the topsurface of the batch blanket was muchlower than the temperature of the lreesurface of the molten glass. This effectbalances the energy requirements of thecndothermic melting reactions.

The laminar glass melt flow patternis shown in figure B ro 10.In these figuresthe temperature field and the flow fieldare plotted. Thc temperature of the moltenglass is the essential property for the batchmelting, glass homogenisarion andrelining which is dependent on hearradiation from the combustion chamber,heat transfer from the glass melt to thebatch blanket and inside glass tank, alsoact on the convective effects and theproductive flow. The weir forces themolten glass flowing near free surface,

which has the function of homogenisationand refining. Figure 9 and 10 show thetemperature field and flow field of themolten glass at, the centre plan of the glasstank with the weir. The effect of the<cold> molten glass getting in the glass-melt buoyant flow causes significantdeformation of the flow pattern andtemperature field in the vitrificationregion as shown in figures 6 to i0,

ConclusionsThis paper reports a test case used

to evaluate 3D modelling capabiiitiesand to study its effective applicabilityinside fumace design loops. Predictionsof the temperature lield, the flow fieldof the glass tank and the combustionchamber, the chemical speciesconcentration distribution, the radiativefluxes, the batch blanket area floatingon the surlace of the molten glass andthe batch melting process were shown,Useful information about whole glassfurnace can be supplied by the modelwilh a very acceptabie deviation rangemaking possible modelling use foroptimal furnace conception. Detailedanalysis of thc lumace performance istherefore possibie from the predictionof spatial distribution of relevan!properties inside the fumace enclosure,Taking into account, during the furnaceconception process, very detailedknowledge about the physical andchemical process occurring inside thefurnace environment is an openpossibility.

AcknowledgementsThe authors acknowledge FundagaoOriente for the financial supporr of thiswork through rhe Ph.D. scholarship ofMr. Wang Jian. The valuable supportfrom Saint Gobain Recherche, rheimportant contributions from M.Khalil and F. Anterion (SGR) and thehelp of SORG in the validation of thepresent model are also kindlyacknowledged.

ReferencesI I ], Ce.nvar-Ho M.G., LocKwooD F.C.,Proc. Instn. Mech. Engnrs., 199(1985),113-120.[2]. Uttcr,N A., VtSKANTA R., GlasrechBer. 60(1987), No.3, 71-78, N0.4, ll5-r24.[3]. Uwo,q,N A., VTSKANTA R., Glastech.Ber. 59(1986), No. t0,279-29t.[4].CanvnlHo M.G., Compurersimulation of a glass furnace. Ph. D.Thesis,lmperial College of Science andTechnology, London, 1983.

[5].Cenveluo M,G,, NocuerRa M.,Proceedings of the first conference ofthe European Sociery of Glass Sciettceand Tec'hnology, 169- 177, Sheffield,England, 9-12, September, 1991.[6].Mase H,, Ona, K., Reports of theresearch laboratory, Asahi Glass Co.ttd. 30(1980), No.2, 69-78.[7]. Kunul M. Private communication.May, 1994.[8],Knocen C,, EltceHnusEN H.,G last ec h. B e r. 32(1959), No.9, 362-3'7 3.