8
Three-Dimensional Numerical Simulations and Analysis of a Heat- Recovery Coke Oven Qianqian Zheng and Hongyuan Wei* School of Chemical Engineering and Technology, Tianjin University, Tianjin 300072, Peoples Republic of China ABSTRACT: Three-dimensional numerical simulations of the combustion process in a heat-recovery coke oven were carried out using the eddy dissipate concept combustion model, which give an in-depth understanding of behaviors in the heat-recovery coke oven and provide useful information for optimizing operations and design. Four dierent coal devolatilization models, including a single-step model, a two competing reaction rate model, and distributed activation energy models (k = constant, and k = ae bE ), were employed to assess the volatile matter evolution rate and temperature distribution in the oven. In comparison to the experimental data, the distributed activation energy model (k = ae bE ) gives a more satised prediction and was further extended to study the distributions of volatile species and pressure in the oven. All of the simulation results have been proven to be converged, grid-independent, and reliable. 1. INTRODUCTION The modern coke-making industry has been developed over 100 years and has tremendous success in technology and economic elds. 1 However, the application of the conventional coke oven (slot-type ovens) brought a series of problems, such as dust pollution, water pollution, and air pollution, which conicted with the new environmental and related political legislations. The clean production of cokes has therefore become the top priority in the coking industry worldwide. At present, several new and ecient technologies have been developed to control and reduce pollution, for example, the jumbo coking reactor, SCOPE21 (super coke oven for productivity and environment enhancement toward the 21st century), and heat-recovery coke oven. 2 The heat-recovery coke technology transforms the coal into a high-quality coke for the blast furnaces of the steel plant and the volatile matter into gases, which combust with air in the oven to supply the heat for the coking process. Under the negative pressure, the toxic gases and other harmful substances are burnt directly during the coking cycle. The remaining heat is then used for steam production and power generation. Considering the most energy ecient and minimum pollution that the heat-recovery coking technology provided, it is extensively regarded as one of the best available environmental control technologies. In comparison to the conventional coking technology, the heat-recovery coking process is a very complicated process because it involves several complex, simultaneous, and interdependent processes, such as turbulent ow, moisture transfer, coal devolatilization, and volatile gas combustion. Among them, coal devolatilization is the basic step in the coking process and has the dominant eect on the overall combustion behavior. 3 Accordingly, the true rate of coal devolatilization is a matter of some contention, and a number of approaches have been proposed to obtain the kinetics of the complex devolatilization process. Some approaches have been implemented in the computational uid dynamics (CFD) simulations to obtain the volatile matter evolution rate of coal by many researchers. 4-6 The simplest approach is an empirical model and employs global kinetics, where Arrhenius expressions are used to correlate rates of weight loss with temperature. These simple models can be divided into single- and multiple-step reactions. The single-step model proposed by Badzioch and Hawksley 7 is widely used for the devolatilization of coal in many research works because it contains less parameters. 8-10 However, employing the single-step model with inappropriate parameters can lead to large errors in the numerical simulations. In the work performed by Gera et al., the eect of kinetic parameters of the single-step model with dierent values of the activation energy on the ame structure has been conducted. 11 The results suggested that the errors in coal devolatilization parameters have the dominant eect on simulated results. In comparison to the single-step model, the multiple-step model considers the eect of the heating rate on the coal devolatilization and has been employed to obtain the rate of coal devolatilization satisfactorily. 12 However, the more parameters shown in the model usually cause the complexity of the simulation process. Jones et al. have used single-step and two competing reaction rate models to assess the global devolatilization rates based on a laminar ow of a drop-tube furnace. 13 The results indicated that the single-step model coupled with the functional-group, depolymerization, vapor- ization, and cross-linking (FG-DVC) devolatilization mecha- nism model can give more satised proles. Another approach is the distributed activation energy model (DAEM). The model is based on the assumption that the devolatilization occurs through several rst-order reactions, which occur simultaneously. In this model, it is crucial to estimate the frequency factor and the distribution function of the activation energy. The distribution function is generally assumed by a Gaussian distribution. As for the frequency factor, it is assumed to be a constant for all reactions to avoid the Received: February 5, 2013 Revised: May 1, 2013 Article pubs.acs.org/EF © XXXX American Chemical Society A dx.doi.org/10.1021/ef400218y | Energy Fuels XXXX, XXX, XXX-XXX

Three-Dimensional Numerical Simulations and Analysis of a Heat-Recovery Coke Oven

Embed Size (px)

Citation preview

Three-Dimensional Numerical Simulations and Analysis of a Heat-Recovery Coke OvenQianqian Zheng and Hongyuan Wei*

School of Chemical Engineering and Technology, Tianjin University, Tianjin 300072, People’s Republic of China

ABSTRACT: Three-dimensional numerical simulations of the combustion process in a heat-recovery coke oven were carriedout using the eddy dissipate concept combustion model, which give an in-depth understanding of behaviors in the heat-recoverycoke oven and provide useful information for optimizing operations and design. Four different coal devolatilization models,including a single-step model, a two competing reaction rate model, and distributed activation energy models (k = constant, andk = aebE), were employed to assess the volatile matter evolution rate and temperature distribution in the oven. In comparison tothe experimental data, the distributed activation energy model (k = aebE) gives a more satisfied prediction and was furtherextended to study the distributions of volatile species and pressure in the oven. All of the simulation results have been proven tobe converged, grid-independent, and reliable.

1. INTRODUCTIONThe modern coke-making industry has been developed over100 years and has tremendous success in technology andeconomic fields.1 However, the application of the conventionalcoke oven (slot-type ovens) brought a series of problems, suchas dust pollution, water pollution, and air pollution, whichconflicted with the new environmental and related politicallegislations. The clean production of cokes has thereforebecome the top priority in the coking industry worldwide. Atpresent, several new and efficient technologies have beendeveloped to control and reduce pollution, for example, the“jumbo coking reactor”, SCOPE21 (super coke oven forproductivity and environment enhancement toward the 21stcentury), and heat-recovery coke oven.2

The heat-recovery coke technology transforms the coal into ahigh-quality coke for the blast furnaces of the steel plant andthe volatile matter into gases, which combust with air in theoven to supply the heat for the coking process. Under thenegative pressure, the toxic gases and other harmful substancesare burnt directly during the coking cycle. The remaining heatis then used for steam production and power generation.Considering the most energy efficient and minimum pollutionthat the heat-recovery coking technology provided, it isextensively regarded as one of the best available environmentalcontrol technologies.In comparison to the conventional coking technology, the

heat-recovery coking process is a very complicated processbecause it involves several complex, simultaneous, andinterdependent processes, such as turbulent flow, moisturetransfer, coal devolatilization, and volatile gas combustion.Among them, coal devolatilization is the basic step in thecoking process and has the dominant effect on the overallcombustion behavior.3 Accordingly, the true rate of coaldevolatilization is a matter of some contention, and a numberof approaches have been proposed to obtain the kinetics of thecomplex devolatilization process. Some approaches have beenimplemented in the computational fluid dynamics (CFD)simulations to obtain the volatile matter evolution rate of coalby many researchers.4−6

The simplest approach is an empirical model and employsglobal kinetics, where Arrhenius expressions are used tocorrelate rates of weight loss with temperature. These simplemodels can be divided into single- and multiple-step reactions.The single-step model proposed by Badzioch and Hawksley7 iswidely used for the devolatilization of coal in many researchworks because it contains less parameters.8−10 However,employing the single-step model with inappropriate parameterscan lead to large errors in the numerical simulations. In thework performed by Gera et al., the effect of kinetic parametersof the single-step model with different values of the activationenergy on the flame structure has been conducted.11 Theresults suggested that the errors in coal devolatilizationparameters have the dominant effect on simulated results. Incomparison to the single-step model, the multiple-step modelconsiders the effect of the heating rate on the coaldevolatilization and has been employed to obtain the rate ofcoal devolatilization satisfactorily.12 However, the moreparameters shown in the model usually cause the complexityof the simulation process. Jones et al. have used single-step andtwo competing reaction rate models to assess the globaldevolatilization rates based on a laminar flow of a drop-tubefurnace.13 The results indicated that the single-step modelcoupled with the functional-group, depolymerization, vapor-ization, and cross-linking (FG-DVC) devolatilization mecha-nism model can give more satisfied profiles.Another approach is the distributed activation energy model

(DAEM). The model is based on the assumption that thedevolatilization occurs through several first-order reactions,which occur simultaneously. In this model, it is crucial toestimate the frequency factor and the distribution function ofthe activation energy. The distribution function is generallyassumed by a Gaussian distribution. As for the frequency factor,it is assumed to be a constant for all reactions to avoid the

Received: February 5, 2013Revised: May 1, 2013

Article

pubs.acs.org/EF

© XXXX American Chemical Society A dx.doi.org/10.1021/ef400218y | Energy Fuels XXXX, XXX, XXX−XXX

complexity of the analysis or correlated with the activationenergy based on experimental data. In comparison to the coaldevolatilization model described above, an additional modelparameter, the standard deviation of the activation energy, isintroduced in this model.Besides, very recently, a new coal devolatilization model,

named the tabulated devolatilization process (TDP) model, hasbeen proposed by Hashimoto et al.14 In their work, variousdevolatilization models, such as the single-step model, twocompeting reaction rate model, and TDP model, have beenused to assess the global devolatilization rates. The resultsindicated that the coal particle and gas-phase behaviorspredicted by the TDP model were better than those by theother models.15 However, the TDP model was based on thedevolatilization database obtained from experiments or fromother models, such as the FLASHCHAIN model. Moreover,considering the complex calculation of the TDP model, thesimulation process needs more computational efforts thanthose of conventional devolatilization models. Therefore, therange of application and extrapolation is limited.As mentioned above, the heat-recovery coke oven is one of

the available environmentally friendly technologies for thecoking industry. However, up to now, a limited number ofstudies address the combustion process in the heat-recoverycoke oven using the numerical simulation method. In thepresent work, three-dimensional numerical simulations forheat-recovery coke oven were performed using a commercialcode, ANSYS FLUENT 12.1. Meanwhile, the influence of fourdifferent devolatilization models, including the single-stepmodel, two competing reaction rate model, DAEM (k =constant), and DAEM (k = aebE), was assessed on the accuracyof simulation results of the combustion behavior. This study isto provide a wide range of information on future studies in theheat-recovery coke oven and to optimize the designs of thecombustion chamber and sole flues as well as the cokingprocess operations.

2. COMPUTATIONAL DOMAINA heat-recovery coke oven consists of five main sections, coal/coke bed, combustion chamber, downcomers, sole flues, anduptakes, as shown in Figure 1. During the coking process, coalwas charged into the oven and heated to a high temperatureuntil substantially all of the volatile matters have been driven offand the remaining solid residues were coke. The air and volatilegas flow directions in the oven are also illustrated in Figure 1 byarrows in blue and red, respectively. The volatile gas releasedfrom the coal/coke bed mixed with the air from the primary airinlets at the top of the coke oven and burnt directly in thecombustion chamber above the coal/coke bed. The heat fromcombustion was then transferred into the coal/coke bed in theform of convection and radiation. However, only part of volatilegas combustion took place in the combustion chamber. Theremaining portion of fuel-rich gas flowed through thedowncomers in the oven side walls into the sole flues, whereflammable species remaining in the gas flow continuously burnout with air from the secondary inlets. Finally, the exhaust gaspassed through the uptakes in the oven side walls into thewaste-gas-collecting system.The configuration and dimensions of the computational

domains, which were designed to exactly match the actualconfiguration and dimensions of a typical heat-recovery cokeoven, are shown in Figure 2. Grids were created in a computer-aided design (CAD) program called GAMBIT 2.4.6 and

exported into ANSYS FLUENT 12.1. The coking coal used inthe present study is a medium-volatile coking coal from thenorth of China, the properties of which are summarized inTable 1.

3. MODELING3.1. Coal Devolatilization Models. In this study, four coal

devolatilization models, i.e., a single-step model, a two competingreaction rate model, a DAEM (k = constant) model, and a DAEM (k =aebE) model, are employed to obtain the devolatilization rate.

Figure 1. Structure and gas flows in a heat-recovery coke oven.

Figure 2. Configuration and dimensions of the computational domain.

Table 1. Properties of Coking Coal

proximate analysis (wt %)

moisture ashvolatilematter

fixedcarbon

high heating value(kcal kg−1)

1.65 9.76 25.45 63.14 7559

Energy & Fuels Article

dx.doi.org/10.1021/ef400218y | Energy Fuels XXXX, XXX, XXX−XXXB

3.1.1. Single-Step Model. The single-step model was commonlyemployed for modeling the evolution of volatile matters from coalduring the pyrolysis process. The Arrhenius expression was used tocorrelate rates of mass loss with temperature. The overall process ofdevolatilization can be represented by a single-step first-order reaction

→ + −V Vcoal (volatiles) (1 )(char)k0

(1)

The devolatilization rate of coal pyrolysis can be then expressed by

= * −Vt

k V Vdd

( )0 (2)

where V and V* represent the mass of volatile matter evolved from thecoal at specific time t and the total volatile content of the coal,respectively. The rate constant k0 can be obtained by the Arrheniuslaw.

= −⎜ ⎟⎛⎝

⎞⎠k A

ERT

exp0 00

(3)

where R is the universal gas constant in J K−1 mol−1 and T is theabsolute temperature in K. Referring to the study by Arenillas et al.,the pre-exponential factor A0 of 114 s

−1 and apparent activation energyE0 of 7.44 × 104 J mol−1 are used in this study.16

3.1.2. Two Competing Reaction Rate Model. Kobayashi et al.proposed the two competing, irreversible, and first-order reactionswith different rate parameters and volatile yields to describe thethermal decomposition of coal17

α α→ + −coal (volatiles I) (1 )(char I)k

1 11

(4)

α α→ + −coal (volatiles II) (1 )(char II)k

2 22

(5)

where the coal is taken to be on dry and ash-free (daf) basis, k1 and k2are the rate constants to determine the rate of conversion of the coal,and α1 and α2 are the mass stoichiometry coefficients. α1 = 0.2139 isobtained from proximate analysis of the coal. With regard to α2, it wastaken as 0.8, referring to the literature.18 Although the two reactionsoccurred simultaneously, the thermal decomposition of coal wasdominated by the first reaction (eq 4) at lower particle temperaturesand the second reaction (eq 5) at higher particle temperatures.Similar to k0 shown in eq 3, k1 and k2 can also be represented by the

Arrhenius equation shown as follows:

= −⎜ ⎟⎛⎝

⎞⎠k A

ERT

exp1 11

(6)

= −⎜ ⎟⎛⎝

⎞⎠k A

ERT

exp2 22

(7)

where A1 and A2 are pre-exponential factors and E1 and E2 are theactivation energies. Values of A1 and A2 and E1 and E2 taken fromliterature are 3.70 × 105 and 1.46 × 1013 s−1 and 7.37 × 104 and 2.51 ×105 J mol−1, respectively.18

Therefore, the rate of volatile matters is expressed by the followingequation:

∫ ∫α α= + − +VV

k k k k t t( )exp( ( )d )dt t

coal 01 1 2 2

01 2

(8)

where Vcoal is the daf mass of the raw coal.3.1.3. DAEM. The DAEM, originally developed by Vand19 in 1942,

has been frequently used to represent the change in overallconversion.20 In this model, the total amount of volatile materialreleased up to time t is given by

∫ ∫−*

= −∞

−VV

k t f E E1 exp( e d ) ( )dt

E RT

0 0

/(9)

where f(E) is a distribution function of the activation energy and k isthe frequency factor. According to the assumption that the activationenergy had a continuous distribution, the function f(E) must satisfy

∫ =∞

f E E( )d 10 (10)

Generally, the distribution curve f(E) is assumed by Gaussiandistribution,21−23 with mean activation energy E and standarddeviation σ. In this work, E and σ are 2.427 × 105 and 4.105 × 104

J mol−1, respectively.24,25 As mentioned above, the frequency factor kcan be treated as a constant (1.67 × 1013 s−1) or variables dependsupon the activation energy E26−29

=k aebE (11)

where a and b are constants and set as 2.8905 and 0.0001,respectively.25

3.2. Governing Equations in CFD and Radiation Model. Toestablish three-dimensional numerical simulations of the heat-recoverycoke oven, except for the previous four devolatilization models, othermodels, including the governing equations for fluid dynamics, speciestransport equations, the turbulent model, the moisture-transfer model,and the radiation model, are also needed.

The conservation equations for mass, momentum, energy, andspecies can be typically represented by the following general form:

ρϕ ρ ϕ ϕ∂∂

+ = Γ +⇀

tu S

( )div( ) div( grad )

(12)

where ρ is the density in kg m−3, u is the mean velocity in m s−1, andΦ, Γ, and S in eq 12 are the generalized variable, diffusion coefficient,and source term, respectively. The descriptions of the model in detailcan be found in the literature.30

The effect of the turbulence on fluid flow can be represented by thestandard k−ε model with default parameters,31 which was the simplestbut reliable complete model to describe the turbulence and has beenwidely adopted in the industrial flow and heat-transfer simulations.

In most previous work, for the simplification of the simulation,moisture transfer in the coking process was normally ignored orexpressed by the Arrhenius equation. In the present study, themoisture nonlinear evaporation model was employed to simulate thewater evaporation in the coal/coke bed during the coking process.32

Gaseous combustion between volatile matters and air was calculatedusing the eddy dissipation concept (EDC) combustion model,33 whichconsiders detailed chemical mechanisms in turbulent flows. The ratesof volatile combustion are controlled by the Arrhenius law. Thevolatile matters consist of 10 species (methane, oxygen, carbonmonoxide, carbon dioxide, water vapor, ethane, ethylene, benzenevapor, hydrogen, and nitrogen), whose fraction compositions aredetermined on the basis of previous research.34

In the combustion chamber of the heat-recovery coke oven,radiation, which has been ignored in the study by Jones et al.,13 isactually the dominant mode of heat transfer. The discrete ordinates(DO) radiation model,35,36 which has generally been chosen in theCFD applications to simulate the industry process because of higheraccuracy,37,38 was used in this study.

4. NUMERICAL APPROACH4.1. Numerical Algorithm. Differential equations men-

tioned in section 3 were discretized by a first-order upwinddifferencing scheme and solved by a finite volume method byPatankar39 in the ANSYS FLUENT 12.1. The moistureevaporation model and coal devolatilization models werespecified by user-defined functions (UDFs) written in the Cprogramming language and compiled to the FLUENT solver.The semi-implicit method for pressure-linked equations(SIMPLE) algorithm was used for the pressure−velocitycoupling and correction. The normalized absolute residualsfor all of the variables in each cell were limited to be less than10−6. The simulations were started in transient mode of 60 h,which was counted as a coking cycle. All cases of thesimulations were carried out in the “TH-1A” supercomputer,which is the one of the fastest speed computers (2566 trillion

Energy & Fuels Article

dx.doi.org/10.1021/ef400218y | Energy Fuels XXXX, XXX, XXX−XXXC

calculations per second) in the world, and a total of 100processors have been employed for parallel computing in thisstudy.4.2. Boundary and Initial Conditions. All of the

boundary conditions in the simulation were set at the sameconditions as the actual operations. The coal/coke bed wastreated as a porous media, and the porosity of the bed wasmeasured in the previous experiments with the value of 35%.To simplify the calculation, the value of porosity was assumedto be constant during the coking period, although it changedwith the coking process. Two types of inlets (primary andsecondary inlets) were designated as the pressure inlet, wherethe direction of fluid flow was normal to the boundary. Theoutlets for the computational domain on the top of uptakes ofthe coke oven were specified as the pressure outlet, and therelative pressure was −150 Pa. The surface on the coal/cokebed was set as interior boundary conditions, so that the volatilematter can flow from the bed to the combustion chamber.The computational domain was predefined by a set of initial

conditions. According to the actual operating condition, theinitial relative pressure and temperature for coal/coke bed andfluid zone were set to be 0 Pa and 298 and 1273 K, respectively.The measured composition and concentration of gas remainingin the combustion region after the coking batch were taken asthe initial conditions for the gas species.

5. RESULTS AND DISCUSSION5.1. Grid Independency. The dependence of the grid size

on the numerical simulation results has been investigated.Three different non-uniform grid sizes (1.42 × 106, 2.76 × 106,and 4.94 × 106) were employed for 1 h of coking time. Theinstantaneous temperatures of three mesh cases in the center ofthe oven are shown in Figure 3. It can be noticed that thetemperature distributions obtained using the three grid sizesshowed similarly. Table 2 presents the prediction of temper-ature values at the measuring point, as shown in Figure 2, forthree grid size cases. The results show that the temperaturevalue obtained using the coarse-mesh case was higher thanthose from the medium- and fine-mesh cases. The temperaturevalues obtained using medium- and fine-mesh cases were very

close to each other, which indicates that the numerical solutionis essentially not sensitive to the number of cells and both ofthe mesh cases can provide the same results. Moreover, thesimulation time of medium grid size was half of that of the finecase. Thus, the medium grid size was employed for all of thesimulations discussed in this study.

5.2. Volatile Matter Evolution Rate. Four simulationcases were carried out in the present study, and the detailinformation is listed in Table 3.The total volatile matter evolution rate of coal/coke bed,

which is the volume integral of the instantaneous volatilematter evolution rate in the whole coal/coke bed, predicted byfour devolatilization models versus the coking time is shown inFigure 4. It can be seen that different devolatilization modelsgave different results. Among them, cases 3 and 4 gave thefastest and slowest evolution rates, respectively, although theDAEM model was used in the two cases. It indicates that theselection of the frequency factor k in the DAEM model has asignificant influence on prediction results. Many researchershave suggested that k = constant cannot reflect thedevolatilization behavior of coal and the compensation effectsbetween k and the activation energy E must be considered.40,41

Also, the volatile matter evolution rates predicted in cases 1 and2 showed an apparent difference. The rate predicted by the twocompeting reaction rate model in case 2 was relatively rapid,which was in coincident with the results obtained by Jones etal.13

The comparison of the experimental value of total volatilemass according to the proximate analysis to the simulatedvalues obtained by four coal devolatilization rates is list in Table4. The percentage error in total volatile mass for foursimulations and experimental value was within 1%, whichindicated that the simulations were converged and all of thenumerical results were mass-balanced.

5.3. Temperature Distribution. Figure 5 shows thetemperature profiles of different coal devolatilization modelsat measuring points with the comparison to the temperaturemeasured by the thermocouple probes in the production. It canbe clearly noticed that the trend of temperature profilespredicted in cases 1 and 4 are in good agreement with themeasured data. However, the temperature values obtained incase 1 show different values at the initial stage (t < 10 h) andthe end stage (t > 55 h) of the coking process. The former isdue to the combined effect of the lack of volatile matter and theinflow of cold air. The latter is caused by the faster volatilematter evolution rate, which can be clearly observed in Figure4. For case 4, it gives more accurate temperature values, which

Figure 3. Instantaneous temperature profiles at three mesh cases (left, 1.42 × 106; middle, 2.76 × 106; and right, 4.94 × 106).

Table 2. Grid Independency Results

mesh 1.42 × 106 2.76 × 106 4.94 × 106

temperature at the measuring point(K)

1284.17 1237.11 1228.81

simulation time (s) 7339 10160 16328

Energy & Fuels Article

dx.doi.org/10.1021/ef400218y | Energy Fuels XXXX, XXX, XXX−XXXD

indicates that the application of case 4 is more appropriate forthe simulation of heat-recovery coke oven in this study. Theresults from cases 2 and 3 do not adequately predict thedistribution of temperature over the operating period. Althoughthe predictions of temperature distribution at the early stage (t< 30 h) from cases 2 and 3 were reasonable, they gave a totallywrong trend after 30 h of operating time. It can be explained bythe overestimation of volatile matter evolution rates predictedby the two competing reaction rate model and DAEM (k =constant) model at the early stage and then only few volatilematters maintained to react at the late stage. On the basis of theprevious discussion, the DAEM (k = aebE) model was applied asthe coal devolatilization model to simulate the combustionbehavior in further studies.The temperature contours in the heat-recovery coke oven

predicted in case 4 at different coking times are shown in Figure6. It can be noticed that the temperature in the combustionchamber was higher than that in other parts of the oven. Thetemperature in the middle part of the combustion chamber washigher than that near both oven doors. The higher temperaturein the combustion chamber denoted that the combustionmainly occurred in the combustion chamber under the presentoperating condition. The maldistribution of temperature in thecombustion chamber can be attributed to the location ofprimary inlets. The absence of primary inlets near the bothoven doors results in the slightly lower temperature. Moreover,

the heat loss caused by the oven door was also taken intoaccount in the simulations.

5.4. Volatile Species Distribution. Figure 7 shows thedistribution of the predicted methane mass fraction in thecenter of the oven at different coking times. The mass fractiondistributions of other components in the volatile matterbehaved in a similar way. It can be found that the location ofthe enrichment area of methane gradually shifted from thesurface of coal/coke bed toward the bottom, which iscoincident with the temperature distribution in the oven andindicates the difference from the conventional oven.42 Withrespect to the high methane mass fraction near the surfaceagain at the end stage of the coking period, it can be explainedby the accumulation of methane evolved from the bottom ofthe coal/coke bed. In addition, it can also been noticed inFigure 7 that the methane mass fraction in the combustionregion at t = 35 h is much higher than that at t = 5 and 60 h,which is in coincident with the fact that a large amount ofvolatile matters released from the coal/coke bed at t = 35 h.

5.5. Pressure Distribution. Generally, the pressuredistribution was influenced by the flow field. Figure 8 showsthe distribution of velocity vectors at different coking times. Itcan be noticed that the velocity vectors at different coking timesexhibit similarly. With respect to the vortex at the coking timeof 35 h, it can be attributed to the higher volatile matterevolution rate (Figure 4). Further, the volume-weighted averageof velocity at different coking times was calculated. The

Table 3. Models Used in the Four Simulations

case devolatilization model turbulent model combustion model radiation model grid

1 single-step model

standard k−ε model EDC model discrete ordinates model 2.76 × 1062 two competing reaction rate model3 DAEM (k = constant)4 DAEM (k = aebE)

Figure 4. Total volatile matter evolution rate of coal/coke bed duringthe coking process.

Table 4. Comparison of the Simulated Total Volatile Mass to the Theoretical Value

simulated value

experimental value case 1 case 2 case 3 case 4

total mass of volatile matter (kg) 1.3082 × 104 1.3136 × 104 1.3141 × 104 1.3172 × 104 1.3089 × 104

relative error (%) 0.4142 0.4519 0.6854 0.0531

Figure 5. Comparison of the temperature at the measuring pointpredicted by the simulations to the experiment.

Energy & Fuels Article

dx.doi.org/10.1021/ef400218y | Energy Fuels XXXX, XXX, XXX−XXXE

relatively lower values (3.475, 3.842, and 3.576 m s−1 at t = 5,35, and 60 h, respectively) indicate that the impact of the flowfield on the pressure distribution can be negligible.To quantify the differences in the pressure distribution in the

oven, five typical locations at the center part of oven wereselected (Figure 9). Moreover, the points at the primary inletand outlet were used as references. In Figure 10, the pressureprofiles of the typical points at different coking times arepresented. It can be noticed that, for the given time, thepressure dropped gradually from the primary inlet to point 5,which reflected the flow direction. Namely, the gas flowed fromthe primary inlet to point 5 in the oven. With respect to thehigher pressure at outlets, it can be attributed to thermalbuoyancy, which is the density difference between the hot andcold gas. The thermal buoyancy results in less density of hot gasflow upward in the uptakes.For the pressure at the specific point, it can be noticed in

Figure 10 that, for all examined points, the pressure increasedwith the coking time and then dropped. It needs to bementioned that the pressure difference between the primaryinlet and point 1 determined the air input and then affected thecombustion of volatile matters.

6. CONCLUSION

The three-dimensional numerical simulations of the heat-recovery coke oven were performed using ANSYS FLUENT12.1. The effect of the four coal devolatilization models,including the single-step model, the two competing reactionrate model, and two DAEM models (k = constant, and k =aebE), on the distributions of the volatile matter evolution rateand gas temperature have been investigated. Although thepredicted temperature distributions using the single-step modeland the DAEM (k = aebE) are in good agreement withexperimental data, the DAEM (k = aebE) gives more accuratetemperature values. On this basis, behaviors in the heat-recovery coke oven were studied. The results show that thecombustion mainly occurred in the combustion chamber andthe maldistribution of temperature in the combustion chambercan be attributed to the absence of primary inlets near bothoven doors. The predicated distribution of the methane massfraction reflects the characteristic of the coking process in theheat-recovery coke oven. With respect to the pressuredistribution, it reflects the flow direction of gas and affectsthe combustion performance in the oven.

Figure 6. Temperature distributions in the oven at different coking times (left, 5 h; middle, 35 h; and right, 60 h).

Figure 7. Distribution of the methane mass fraction at different coking times (left, 5 h; middle, 35 h; and right, 60 h).

Figure 8. Distribution of velocity vectors at different coking times (left, 5 h; middle, 35 h; and right, 60 h).

Energy & Fuels Article

dx.doi.org/10.1021/ef400218y | Energy Fuels XXXX, XXX, XXX−XXXF

■ AUTHOR INFORMATIONCorresponding Author*Telephone: +86-22-27405754. Fax: +86-22-27400287. E-mail:[email protected] authors declare no competing financial interest.

■ ACKNOWLEDGMENTSThe study was carried out at the National SupercomputerCenter in Tianjin, China, and the calculations were performedon TianHe-1 (A).

■ REFERENCES(1) Toll, H.; Worberg, R. The current development and latestachievements in cokemaking technology. Rev. Metall. 2003, 100, 243−250.(2) Polenske, K. R.; McMichael, F. C. A Chinese cokemakingprocess-flow model for energy and environmental analyses. EnergyPolicy 2002, 30, 865−883.(3) Guruz, G. A.; Uctepe, U.; Durusoy, T. Mathematical modeling ofthermal decomposition of coal. J. Anal. Appl. Pyrolysis 2004, 71, 537−551.

(4) Vuthaluru, R.; Vuthaluru, H. Modelling of a wall fired furnace fordifferent operating conditions using FLUENT. Fuel Process. Technol.2006, 87, 633−639.(5) Sadhukhan, A. K.; Gupta, P.; Saha, R. K. Modeling andexperimental studies on single particle coal devolatilization andresidual char combustion in fluidized bed. Fuel 2011, 90, 2132−2141.(6) Shen, Y. S.; Yu, A. B.; Austin, P. R.; Zulli, P. CFD study of in-furnace phenomena of pulverised coal injection in blast furnace:Effects of operating conditions. Powder Technol. 2012, 223, 27−38.(7) Badzioch, S.; Hawksley, P. G. W. Kinetics of thermaldecomposition of pulverized coal particles. Ind. Eng. Chem. Proc. Des.Dev. 1970, 9, 521−530.(8) Al-Abbas, A. H.; Naser, J.; Dodds, D. CFD modelling of air-firedand oxy-fuel combustion of lignite in a 100 kW furnace. Fuel 2011, 90,1778−1795.(9) Alvarez, L.; Gharebaghi, M.; Pourkashanian, M.; Williams, A.;Riaza, J.; Pevida, C.; Pis, J. J.; Rubiera, F. CFD modelling of oxy-coalcombustion in an entrained flow reactor. Fuel Process. Technol. 2011,92, 1489−1497.(10) Fang, Q.; Musa, A. A. B.; Wei, Y.; Luo, Z.; Zhou, H. Numericalsimulation of multifuel combustion in a 200 MW tangentially firedutility boiler. Energy Fuels 2012, 26, 313−323.(11) Gera, D.; Mathur, M.; Freeman, M. Parametric sensitivity studyof a CFD-based coal devolatilization model. Energy Fuels 2003, 17,794−795.(12) Tang, Y.; Liu, D.; Liu, Y.; Luo, Q. 3D computational fluiddynamics simulation of natural coke steam gasification in general andimproved fluidized beds. Energy Fuels 2010, 24, 5602−5610.(13) Jones, J. M.; Patterson, P. M.; Pourkashanian, M.; Williams, A.;Arenillas, A.; Rubiera, F.; Pis, J. J. Modelling NOx formation in coalparticle combustion at high temperature: An investigation of thedevolatilisation kinetic factors. Fuel 1999, 78, 1171−1179.(14) Hashimoto, N.; Kurose, R.; Hwang, S.; Tsuji, H.; Shirai, H. Anumerical simulation of pulverized coal combustion employing atabulated devolatilization process model (TDP model). Combust.Flame 2012, 159, 353−366.(15) Hashimoto, N.; Kurose, R.; Shirai, H. Numerical simulation ofpulverized coal jet flame employing the TDP model. Fuel 2012, 97,277−287.(16) Arenillas, A.; Rubiera, F.; Pevida, C.; Pis, J. J. A comparison ofdifferent methods for predicting coal devolatilisation kinetics. J. Anal.Appl. Pyrolysis 2001, 58, 685−701.(17) Kobayashi, H.; Howard, J. B.; Saroflm, A. F. Coaldevolatilization at high temperature. Proceedings of the 16th Interna-tional Symposium on Combustion; Cambridge, MA, Aug 15−20, 1976.(18) Ubhayakar, S. K.; Stickler, D. B.; Von Rosenburg, C. W.;Gannon, R. E. Rapid devolatilization of pulverised coal in hotcombustion gases. Proceedings of the 16th International Symposium onCombustion; Cambridge, MA, Aug 15−20, 1976.(19) Vand, V. A theory of the irreversible electrical resistance changesof metallic films evaporated in vacuum. Proc. Phys. Soc. 1943, 55, 222−246.(20) Pitt, G. J. The kinetics of the evolution of volatile products fromcoal. Fuel 1962, 41, 267−274.(21) Burnham, A. K.; Braun, R. L. Global kinetic analysis of complexmaterials. Energy Fuels 1999, 13, 1−22.(22) Donskoi, E.; McElwain, D. L. S. Approximate modelling of coalpyrolysis. Fuel 1999, 78, 825−835.(23) Miura, K.; Maki, T. A simple method for estimating f(E) andk0(E) in the distributed activation energy model. Energy Fuels 1998,12, 864−869.(24) Zhu, X.; Zhu, Z.; Zhang, C. Study of the coal pyrolysis kineticsby thermogrametry. J. Chem. Eng. Chin. Univ. 1999, 13, 223−228.(25) Fu, W.; Zhang, Y.; Han, H.; Wang, D. A genearl model ofpulverized coal devolatilization. Fuel 1989, 68, 505−510.(26) Anthony, D. B.; Howard, J. B. Coal devolatilization andhydrogasification. AIChE J. 1976, 22, 625−656.(27) Chen, J. C. Distributed activation energy model ofheterogeneous coal ignition. Combust. Flame 1996, 107, 291−298.

Figure 9. Locations in the center of the heat-recovery coke oven.

Figure 10. Pressure distribution at different coking times.

Energy & Fuels Article

dx.doi.org/10.1021/ef400218y | Energy Fuels XXXX, XXX, XXX−XXXG

(28) Tang, W.; Wang, C.; Chen, D. Kinetic studies on the pyrolysisof chitin and chitosan. Polym. Degrad. Stab. 2005, 87, 389−394.(29) Hashimoto, K.; Miura, K.; Watanabe, T. Kinetics of thermalregeneration reaction of activated carbons used in waste watertreatment. AIChE J. 1982, 28, 737−746.(30) Versteeg, H. K.; Malalasekera, W. An Introduction to Computa-tional Fluid Dynamics: The Finite Volume Method; Pearson EducationLimited: London, U.K., 2007.(31) Launder, B. E.; Spalding, D. B. Lectures in Mathematical Modelsof Turbulence; Academic Press: London, U.K., 1972.(32) Si, J.; Wen, Z.; Liu, X. Heat transfer model of coking chamberbased on moisture nonlinear evaporation. J. Zhejiang Univ., Sci. 2007,41, 1746−1749.(33) Magnussen, B. F. On the structure of turbulence and ageneralized eddy dissipation concept for chemical reaction in turbulentflow. Proceedings of the 19th AIAA Aerospace Science Meeting; St. Louis,MO, Jan 12−15, 1981.(34) Alvarez, R.; Barriocanal, C.; Canga, C. S.; Canga, J. S.; Diez, M.A.; Gayol, O. M.; Miyar, E. A. Coke oven gas control by one-line gaschromatography. Chromatographia 1989, 27, 611−616.(35) Chui, E. H.; Raithby, G. D. Computation of radiant heat transferon a non-orthogonal mesh using the finite-volume method. Numer.Heat Transfer, Part B 1993, 23, 269−288.(36) Raithby, G. D.; Chui, E. H. A finite-volume method forpredicting a radiant heat transfer in enclosures with participatingmedia. J. Heat Trans. 1990, 112, 415−423.(37) Backreedy, R. I.; Jones, J. M.; Ma, L.; Pourkashanian, M.;Williams, A.; Arenillas, A.; Arias, B.; Pis, J. J.; Rubiera, F. Prediction ofunburned carbon and NOx in a tangentially fired power station usingsingle coals and blends. Fuel 2005, 84, 2196−2203.(38) Ma, L.; Gharebaghi, M.; Porter, R.; Pourkashanian, M.; Jones, J.M.; Williams, A. Modelling methods for co-fired pulverised fuelfurnace. Fuel 2009, 88, 2448−2454.(39) Patankar, S. V. Numerical Heat Transfer and Fluid Flow;Hemisphere Publishing Corporation: New York, 1980.(40) Miura, K. A new and simple method to estimate f(E) and k0(E)in the distributed activation energy model from three sets ofexperimental data. Energy Fuels 1995, 9, 302−307.(41) Criado, J. M.; Perez-Maqueda, L. A.; Sanchez-Jimenez, P. E.Dependence of the preexponential factor on temperature: Errors in theactivation energies calculated by assuming that A is constant. J. Therm.Anal. Calorim. 2005, 82, 671−675.(42) Guo, Z.; Tang, H. Numerical simulation for a process analysis ofa coke oven. China Particuol. 2005, 3, 373−378.

Energy & Fuels Article

dx.doi.org/10.1021/ef400218y | Energy Fuels XXXX, XXX, XXX−XXXH