Numerical Simulations of Pulverized Coal Combustion †

Coal is an important energy resource for meeting the further demand for electricity, as coal reserves are much more abundant than those of other fossil fuels. In pulverized coal fired power plants, it is very important to improve the technology for the control of environmental pollutants such as NOx, SOx and ash particles including unburned carbon. With the remarkable progress in the performance of computers, it is strongly expected that the computational fluid dynamics (CFD) would be a tool for the development and design of such suitable combustion furnaces and burners for the pulverized coal combustion. The focus of this review is to highlight our recent progress of CFD of the pulverized coal combustion in terms of Reynolds-Averaged Navier-Stokes (RANS) simulation and Large-Eddy Simulation (LES) together with some of future perspectives.


Introduction
Coal is an important energy resource for meeting the further demand for electricity, as coal reserves are much more abundant than those of other fossil fuels.In pulverized coal fired power plants, it is very important to improve the technology for the control of environmental pollutants such as NOx, SOx and ash particles including unburned carbon.In order to achieve these requirements, understanding the pulverized coal combustion mechanism and development of the advanced combustion technology are necessary.However, the combustion process of the pulverized coal is not well clarified so far since pulverized coal combustion is a ver y complicated phenomenon, in which the maximum flame temperature exceeds 1500℃ and some substances which can hardly be measured, for example, radical species and highly reactive solid particles are included.Accordingly, development of new combustion furnaces and burners requires high cost and takes a long period because the empirical process comprises many steps.The computational fluid dynamics (CFD) of the pulverized coal combustion field is being developed with the remarkable progress in the performance of computers.This method, in which the governing equations of the combustion field are solved using a computer, is capable to provide the detailed information on the distributions of temperature and chemical species and the behavior of pulverized coal particles over entire combustion field that cannot be obtained by experiments.In addition, it facilitates the repeated review in arbitrary conditions for the properties of pulverized coal and the flow field at a relatively low cost.It is, therefore, strongly expected that the CFD becomes a tool for the development and design of combustion furnaces and burners.DNS (Direct Numerical Simulation), LES (Large-Eddy Simulation) and RANS (Reynolds-Averaged Navier-Stokes) simulation are typical methods for the CFD of the combustion field in terms of turbulence model.DNS, which directly solves governing equations of fields of the flow, chemical species concentrations and temperature by setting the numerical grid space below the minimum eddies in these fields, has the highest numerical accuracy among the above mentioned methods.Although it is effectively ap-plicable to the basic research, its application to the combustion field at practical levels is ver y difficult as it requires a huge number of grid points and high loads to the computer.On the contrary, RANS simulation is most frequently used in practical applications.This method solves the governing equations by averaging them over the time and replacing resulted Reynolds stresses and turbulent scalar fluxes terms with turbulence models.It can considerably reduce the number of grid points and loads to the computer.However, RANS simulation has several problems, such as difficulty in selection of turbulence models and determination of parameter values contained in them and disadvantage in predicting the unsteady turbulent motions.Accordingly, attention is gathered on LES recently, which directly solves governing equations for relatively large eddies and calculates remaining small eddies using models.This is a sort of space averaging method and has certain advantages that the unsteady turbulent motions are evaluated and the number of parameters contained in models is reduced.Although LES poses high loads to computer compared with RANS simulation, it is likely to be applied to practical fields in near future when the progress of computer performance is taken into consideration.The focus of this review is to highlight our recent progress of CFD of pulverized coal combustion in terms of RANS simulation and LES together with some of future perspectives.In Sec. 2, pulverized coal combustion test furnaces and a burner concerned for the present numerical simulations are described together with a brief concept of pulverized coal combustion boilers.Some of numerical results on LES and RANS are shown in Secs. 3 and 4, respectively.Finally, this review is concluded in Sec. 5.

Pulverized Coal Combustion Utility Boiler and Test Facilities for Basic Experiments and Numerical Simulations
In this review, numerical simulations of combustion fields in pulverized coal combustion boilers are focused on.The boiler is a system that transforms heat into steam.In general pulverized coal combustion boilers, pulverized coal is supplied from installed burners, which vertically line up at three locations, as shown in Fig. 1.The number of burners for a boiler is 10-40 and the coal feed rate for each burner is 10 tons/h at most.The utility boilers are generally classified into two types in terms of combustion system, namely opposed firing in which burners are placed front and back a boiler (see Fig. 1) and corner firing in which burners are places at four corners of a boiler.For these boilers, it is more important to understand the vertical flame interaction than the horizontal flame interaction, since the flames bended upward by upward main stream and buoyancy vertically affect the other flames each other.
In order to fundamentally investigate the flame interaction, CRIEPI (Central Research Institute of Electric Power Industr y) utilizes a multi-burner pulverized coal combustion test furnace, in which three burners each with a coal combustion capacity of about 100 kg/h are vertically installed, at Yokosuka Research Laborator y, as shown in Fig. 2.
The multi-burner furnace is connected to flue gas treatment equipments such as selective catalytic De-NOx, electrostatic precipitator, gypsum limestone wet type De-SOx, which are similar to those in utility boilers.The height, horizontal width, and depth of this furnace are 11, 0.9, and 1.9 m, respectively.Combustion air is injected into the furnace through the burner and staged combustion air ports located 2.0 m downstream from the upper-stage burner outlet.In addition, a single-burner pulverized coal combustion test furnace (see Fig. 3) is also used at Yokosuka Research Laboratory of CRIEPI to further understand the detailed structure of a singular flame.The coal combustion capacity of it is about 100 kg/h.The single-burner furnace is a cylindrical furnace so that the pure pulverized combustion behavior can be examined without flame bending and interaction.The diameter of this furnace is 0.85 m and the length is 8 m.Combustion air is injected into the furnace through the burner and staged combustion air ports located 3.0 m downstream from the burner.By comparing the combustion characteristics between these two test furnaces, it is expected to obtain the useful information in designing and operating the actual utility boilers.
The burner installed in the above test furnaces is the CI-α (CRIEPI-IHI Advanced Low-Pollution High-Ability) burner with a coal combustion capacity of about 100 kg/h.Pulverized coal is fed into the furnace with the primary air, and the secondary and tertiary air is supplied from the surrounding area of primary air.The CI-α burner is designed to promote mixing near the burner by means of recirculation flow produced by the straight motion of primar y air and the strong swirling motion of the secondary and tertiary air (see Fig. 4).This recirculation flow lengthens the residence time of pulverized coal particles in the high-temperature field near the burner outlet and accelerates the evolution of volatile matter and the progress of char reaction.Therefore, the amount of unburned carbon in fly ash is effectively reduced, but the NOx concentration increases in this region.Then, the NOx concentration is immediately reduced to N2 in the reduction flame existing after the recirculation zone.Furthermore, this NOx reduction effect is promoted by the staged combustion method, in which some of the combustion air separated from the burner is supplied via injection ports mounted on the rear of the furnace.Detailed descriptions of the CI-α burner and its performance can be found in our previous papers [1][2][3] .
In our numerical simulations, the pulverized coal combustion fields in these single-and multi-burner furnaces with the CI-αburner have been simulated, since the experimental data useful for the validation of the numerical simulations are abundant for a variety of coals and combustion conditions.

RANS simulation method
The gas-phase time-averaged continuity equation and conservation equations of momentum, turbulent kinetic energy, dissipation, enthalpy and chemical species coupled with the renormalization group (RNG) k-ε turbulence model 4,5) are solved using the SIMPLE algorithm 6) .
The schematic diagram of the pulverized coal combustion simplified for this calculation is shown in Fig. 5. Coal devolatilization is simulated by a firstorder single reaction model (arrow ( 1)).Gaseous combustion between the volatilized fuel and air and the char burning rate are calculated using the combined model of kinetics and eddy dissipation models  6)-( 9), respectively [9][10][11] .
Schematic of opposed firing boiler for pulverized coal combustion.

Fig. 1
Multi-burner furnace.Fig. 2 Single-burner furnace.Fig. 3 Concept of CI-α burner.Fig. 4 Concept of pulverized coal combustion.Fig. 5 3.2 Effect of coal properties Bituminous coal, whose fuel ratio is about 1.0 -2.5, is largely used in utility boilers in Japan because of its high ignitability, high combustibility, and high calorific value.However, since the global demand for coal is considered to increase, it is desired to utilize not only the bituminous coal but also low-rank coals with high moisture and high ash contents.Therefore, it is very important to develop the numerical simulation code which can precisely capture the effect of coal properties on combustion characteristics. (

1)Numerical conditions
The simulated test furnace is the single-burner furnace with the CI-α burner.The computational domain and simplified burner geometry are shown in Fig. 6.A part of cylinder (-π/6 θ π/6) is taken as the region of calculation.The feed rates of the primary, secondary and tertiary air are given to match with the actual experiments [18][19][20] . Te heat-value of coal is assumed to be equal to 2.74×10 6 kJ/h and the total flow rate of the primary, secondary and tertiary air are determined so that the excess concentration of O2 at the outlet becomes 4 % (air ratio = 1.24).The rate of staged combustion (= volume flow rate of the staged combustion air/volume flow rate of the air for combustion) is 30%.The number of grid points is taken as 61(x) ×58(r) ×11(θ) (the grid division in the directions of x and r are finely set in the vicinity of burner).It is assumed that the pulverized coal particles consist of particles with diameters of 5, 20, 40, 60, 80 and 100μm (the mean diameter of the pulverized coal is about 40 μm).
(2)Bituminous coals Table 1 lists the properties of five types of bituminous coal subjected to the test 12) .Fig. 7 compares the axial distributions of time-averaged gas temperature and O2 and NO concentrations for Newlands coal with the experiments.Fig. 8 shows the distributions of streamwise gas flow velocity, gas temperature and O2 and NO concentrations (values of characteristic quantities increase as the color changes from blue to red).Qualitatively good agreement is shown between calculations and experiments for all distributions, which indicates that real pulverized coal combustion fields are well reproduced.The quantitative comparison of calculations and experiments, however, shows that the peak value of gas temperature of the former is higher than the latter and the temperature rapidly become lower in the downstream region than the latter.They differ in that the O2 concentration is higher and the peak value of NO concentration is lower in the region of x < 3 m.Possible reasons for these differences are incomplete agreement of shape settings of furnace and burner with those in the experiments and insufficient accuracy of turbulence model in the region near the burner with strong fluid shear.Figs. 9 and 10 compares the relationship of the unburned carbon fraction, Uc* (propor tion of unburned part for the combustible part in the coal), the combustion efficiency, Ef (= 1-Uc*) and the fuel ratio, FR (proportion of fixed carbon in the coal/proportion of volatile matters in the coal) and the relationship of the conversion of fuel N to NO, CR (proportion of N part converted to NO for N part in the coal) and FR/FN, respectively, with the experiments.It is experimentally known that Uc* and CR increase with increasing FR and FR/FN, respectively.It is confirmed from these figures that similar increasing tendency are obtained in the calculations without any contradiction, although they show quantitative differences.
Schematic of computational domain and burner.Fig. 6 Bituminous coal properties.Table1 Computational domain.

Simplified burner geometry
(3)Effects of ash and moisture contents Table 2 shows the coal properties of high ash coal, namely Ikeshima coal (Japanese bituminous coal), which have almost the same fuel ratio and carbon, hydrogen, nitrogen and oxygen contents on a dry ash free basis, but different ash content of 36, 44 and 53 wt% 13) .These samples are prepared by the floatation separation.The ash contents of the tested coals are higher than those of usual coals, which are fired in the thermal power stations in Japan.Fig. 11 shows the comparisons of NOx concentration and unburned carbon fraction, Uc*, at the furnace exit with the experiments.The numerical trends are found to be in general agreement with the experiments.As the ash content increases, both the NOx concentration and Uc* at the furnace exit increase.The results suggest that the present numerical simulation for the pulverized coal combustion is applicable for the pulverized coal combustion at least to know the qualitative effects of the ash.It is considered that the discrepancies between calculations and experiments are probably due to that the shapes of the furnace and the burner modeled in the calculations are not completely the same as those of the experiments and that turbulence model used are not sufficient, as described earlier.Also, compared to the experiments, the calculated Uc* at the furnace exit is low especially for high ash content coals [Fig.11 (b)].It is speculated that this is attributed to the fact that the interaction between ash and combustible matter is not taken into account for the numerical model of char oxidization.To estimate these combustion char-Axial distributions of gas temperature and O 2 and NO concentrations (Newlands coal).

Fig. 7
Distributions of streamwise air velocity, gas temperature and O 2 and NO concentrations (Newlands coal).

Fig. 8
Relation between Uc* and FR.Fig. 9 Relation between CR and FR/FN.Fig. 10 acteristics ash content coal more precisely, the char oxidization model, which takes the presence of ash into account, should be proposed.
To investigate the effects of the moisture in the coal, Kimoto et al. 20) imitatively increases the moisture content by injecting the steam generated by a steam-generation apparatus into the primar y air.Therefore, numerical simulations in the same conditions are performed.It should be noted here that contrar y to the introduction of coal moisture, the consumption of the latent heat of vaporization of the moisture is not necessary in this case.The test fuel is Newlands bituminous coal.The equivalent moisture content, MC ( = mass of the moisture content in the coal/mass of the coal (dry basis)×100), are varied as 0, 17.6 and 42.8.Newlands coal originally contains 2.20 moisture, but the original 2.20 moisture is vapor-ized through a pulverizer and the steam is removed.The present computations are carried out for MC = 0, 20 and 40, and all the moisture in the coal is assumed to evaporate in transit to the furnace 14) .In Fig. 12, the variations of unburned carbon fraction, Uc*, and NOx conversion, CR, at the outlet of the furnace are shown against MC, together with the experiments.It is found that Uc* and CR for MC = 0 well correspond to the experiments, and that the increase of MC leads an increase in Uc* and decrease in CR.These trends are similar to those given by the experiments although the effect of MC on Uc* is weaker in these predictions than in the experiments.Although the difference in Uc* is considered to be caused by the deficiency of the application of the char burning model used here, the model with higher accuracy has not been proposed yet.
High ash coal properties.Table2 Effects of ash content on NOx concentration and Uc*; ○, experiment; ─, calculation.

Fig. 11
Effects of moisture on Uc* and CR.Fig. 12 3.3 Application to complex geometries Although numerical simulations of the pulverized coal combustion fields have been conducted by many researchers, most of the burner geometries are strongly simplified as shown earlier, in spite of the fact that subtle differences in the burner shape significantly affect the coal combustion characteristics.Hence, the geometries of single-and multi-burner furnaces which install the CI-α burners are faithfully duplicated.
(1)Simgle-burner furnace The test furnace considered is the single-burner furnace with CI-α burner.The numerical procedure and conditions are the same as those described earlier.The geometries of the computational domain and the CI-α burner are shown in Fig. 13.These are designed to faithfully match the actual geometries 15) .The computational domain is half of the furnace, and a periodic condition is applied in the azimuthal direction.The swirl vane angles for secondary and tertiary air are set at 81 and 72 , respectively, which are optimum values for bituminous coal (these values are zero when the swirl force is zero).Fig. 14 shows the predicted gas velocity vectors in the CI-α burner and near the burner.It is observed that the secondar y and tertiar y combustion air introduced from the outside in the normal direction are shifted in the azimuthal direction by swirl vanes and the swirling velocity reaches up to 30 m/s.Also, as expected from the previous experiment for coal flow using NO tracer gas 2) , recirculation flow can be seen in the region near the burner.The swirling air, which is ejected along the burner throat with the maximum velocity of about 18 m/s, is divided into upstream and downstream flows near the side wall, and small recirculation flow around the burner (arrow A) and large recirculation flow in the central region along the z-axis (arrow B) are formed.Fig. 15 show the comparison of axial distributions of gas temperature and chemical species concentrations.Overall, the calculated axial distributions of gas temperature and O2 concentration are in better agreement with the experiments than those in the simplified-shape case (see Fig. 7), although there are some quantitative discrepancies, for example, the predicted gas temperature rises earlier and higher, the O2 concentration shows larger values around the region just after the position where staged combustion air is injected.Although the figure is omitted here, it was found that the prediction of the NOx concentration is improved compared to the simplified-shape case 15) and that present numerical method can capture the general combustion characteristics of high-fuel-ratio coals 16) .
(2) Multi-burner furnace RANS simulation is also applied to a pulverized coal combustion field in the multi-burner furnace with three CI-α burners.The geometries of the computational domain is shown in Fig. 16.The swirl vane angles for secondary and tertiary air are set at 72 and 63 , respectively, which are optimum values for bituminous coal (these values are zero when the swirl force is zero).The thermal input of the coal combustion test furnace is 8.62×10 6 kJ/h (the coal feed rate for each burner is approximately 100 kg/h).The air ratio is 1.24, and the excess O2 concentration at the furnace outlet is 4 %.The staged combustion air ratio is set at 30 %.The mass ratio of the pulverized coal (dry base) to the primary air is 1:2.2, and the mass ratio of secondary air to tertiary air is 1:6.Bituminous coal (i.e., Newlands coal) is chosen for the test fuels.Fig. 17 shows the comparisons of distributions of gas temperature and chemical species concentrations between calculations and experiments 21) .On the whole, simulated and experimental contour plots are consistent qualitatively, although there are some quantitative discrepancies, which might be caused by the deficiency of RANS simulation.Although the details are omitted here, this study also suggested the possibility that overall unburned carbon fraction of coal-fired plants can be reduced by supplying coal with a low combustibility to the middle-or lower-stage burner and supplying coal with a high combustibility to the upper-stage burner.
Schematic of computational domain of single-burner furnace and CI-α burner.

Fig. 13
Gas velocity vectors in and near the CI-α burner (Newlands coal).

Fig. 14
Axial distributions of gas temperature and O 2 , H 2 O and CO 2 concentrations (Newlands coal).

Fig. 15
Schematic of computational domain of multi-burner furnace and CI-α burner.Fig. 16 4. Large-Eddy Simulation (LES) 4.1 LES method One of the LES solvers used here is the Front-Flow/red for unstructured grids, which has been originally developed at the Institute of Industrial Science of the University of Tokyo in the Frontier Simulation Software for Industrial Science project as a part of the IT program sponsored by the Ministry of Education, Culture, Sports, Science and Technology (MEXT), extended by CRIEPI, Kyoto University, and NuFD (Numerical Flow Designing, LTD).This solver is based on the SMAC/SIMPLE algorithms and the Finite Volume Method (FVM) with an unstructured grid system.The gas-phase continuity equation and conservation equations of momentum, enthalpy and species coupled with the dynamic Smagorinsky model 22,23) are solved.The mathematical models for the pulverized coal combustion are basically the same as those used in RANS simulation, as described earlier.
The cold and combusting flow fields in the singleburner furnace with the CI-α burner are considered.The geometries of the computational domain and the CI-α burner are designed to faithfully match the actual geometries, as shown in Fig. 18.The swirl vane angles for secondary and tertiary air are set at 81 and 72 , respectively.The computational domain is divided in about 450, 000 tetrahedral volumes with cone-shaped clustering on burner downstream for both LES and RANS simulation.The details of the governing equations, mathematical models and numerical procedures for LES of pulverized coal combustion are described in our previ-ous papers [24][25][26] .

Comparison between LES and RANS simulation
In order to verify the superiority of LES in capturing turbulent swirling flow behavior, LES results for the swirling cold (non-combusting) flows in the pulverized coal combustion furnace are compared with RANS simulation results using the same computational grid and the experiments 25) .In the experi-Distributions of gas temperature and O 2 and NO concentrations (Newlands coal).Fig. 17 Schematic of computational domain of single-burner furnace and CI-α burner.Fig. 18 ments, instantaneous streamwise and azimuthal air velocities are measured using the Ar Laser twodimensional multi-scan Laser-Doppler velocimeter.Fig. 19 shows the comparisons of distributions of axial velocity, Uz, on the flow direction plane and azimuthal velocity, Uθ, on the cross section plane at z = 0.235 m between LES and RANS simulation.Here, the distributions are shown by instantaneous value for LES and time-averaged value for RANS simulation, respectively.The deviation from the axisymmetric flow formed in the time-averaged flow field is observed in the instantaneous Uz distribution for LES.This deviation originates from the vortices created in the region near the burner outlet.It is also found that in the recirculation flow formed near the burner outlet, the backward flow velocity (negative Uz) for LES is smaller than that for RANS simulation.In addition, the outer positive Uz region does not expand to downstream near the side wall for LES, while it is remarkable for RANS simulation.Also, the region showing the ring-shaped large Uθ is less for LES than that for RANS simulation, which means that the swirling flow and the related recirculation flow for LES are weaker than those for RANS simulation.Fig. 20 shows the comparisons of radial distributions of time-averaged axial velocity, Uz, and azimuthal velocity, Uθ at z = 0.235 m among LES, RANS simulation and experiments.Here, for the experiments, measured data in the entire range of the diameter is displayed.It should be noted here that since the secondary and tertiary air cannot be injected uniformly along the circumferential direction due to the limitation of the experimental equipment, the complete axisymmetric flow is not observed in the furnace.Therefore, although the detailed quantitative discussion cannot be done here, the behavior of the recirculation flow formed at r = 0.015−0.12m for LES is similar to that for the experiments.On the other hand, the backward flow velocity of the recircu-lation flow for RANS simulation is much larger than that for the experiments.Also, in the region where the peak value of Uθ appears around r = 0.14 m, Uθ for LES is slightly higher than that for the experiments, while Uθ for RANS simulation is about three times higher.From these comparisons, it is revealed that LES is much superior to RANS simulation to predict the axial velocity in the recirculation flow formed in the swirling flow and the azimuthal velocity around this recirculation flow.The superiority of LES to capture the behavior of the swirling and recirculation flows is considered due to the accuracy in predicting the momentum exchange in the strong shear flow, namely the mixing effect by the turbulence.It was also confirmed that the r.m.s. of axial velocity for LES is in general agreement with the experiments and that these superiority of LES is similarly observed in the cases where the swirl vane angles for secondary and tertiary air are changed.
Distributions of Uz on the flow direction plane (upper) and U θ on the cross section plane at z = 0.235 m (lower).

Fig. 19
Radial distributions of Uz and U θ .

Challenges to pulverized coal combustion field
Very recently, LES has been applied to an actual pulverized coal combustion field in the single-burner furnace with the CI-α burner 26) .Fig. 21 shows the predicted distributions of instantaneous axial velocity, gas temperature and O2 concentration on the flow direction plane.It is observed that a swirling recirculation flow is formed in the central region close to the burner and its size and strength dynamically change with time.Also, this unsteady flow behavior strongly affects the ignition characteristics.Fig. 22 shows the comparisons of axial distributions of time-averaged gas temperature and O2 concentration between calculations and experiments 18,27) .The predicted distributions qualitatively agree with the experiment.Thus, unlike RANS simulation, LES can precisely capture this unsteady combusting flow motion, whose nature has a potential to greatly improve the numerical accuracy.In addition, LES has an advantage that the number of model parameters is much less than that for RANS simulation.However, for the LES of the pulverized coal combustion, there remain difficult challenges in terms of not only mathematical models on turbulent combustion and pollutant emissions but also computational costs.

Conclusions
In this review, our recent progress of Computational Fluid Dynamics (CFD) of pulverized coal combustion is discussed in terms of RANS (Reynolds-Averaged Navier-Stokes) simulation and LES (Large-Eddy Simulation).It can be said that regarding the prediction of flow field, LES is much superior to RANS simulation and therefore expected to become a useful tool for the development and design of new combustion furnaces and burners in near future.However, the combustion of pulverized coal is a complicated phenomenon in which a wide variety of processes such as the devolatilization of pulverized coal, combustion of volatile matters and combustion at the surface of particles, in addition to the dispersion behavior of pulverized coal in the gas stream, simultaneously act to each other.Therefore, the pulverized coal combustion behavior is not fully understood and modeled yet.In particular, coal devolatilization model is found to strongly affect the numerical accuracy recently 28) .The improvement of turbulent combustion model is also necessary.The more sophisticated turbulent combustion model, in which the detailed reaction mechanism can be taken into consideration such as flamelet models often used for gaseous and spray combustion [29][30][31][32] , is desirable. Frthermore, the application of LES to much larger and more complicated fields such in utility boilers is challenges in terms of numerical stability, accuracy and costs.
Distributions of instantaneous axial velocity, gas temperature and O 2 concentration on the flow direction plane.

Fig. 21
Comparisons of axial distributions of time-averaged gas temperature and O 2 concentration between calculations and experiments.
Fig. 22 7) (arrows (3)-(5)) and Field et al.ʼs model 8) (arrow (2)), respectively.The NOx formation model is employed in a post-processing fashion, where a converged combustion flow field solution is first obtained before performing the NO prediction.For the formation of NOx, three different mechanisms are employed, namely Zeldovich NOx, prompt NOx and fuel NOx formation mechanisms indicated by arrows (