Large eddy simulation of oxy-coal combustion in an industrial combustion test facility

Large eddy simulation of oxy-coal combustion in an industrial combustion test facility

International Journal of Greenhouse Gas Control 5S (2011) S100–S110 Contents lists available at ScienceDirect International Journal of Greenhouse Ga...

658KB Sizes 0 Downloads 44 Views

International Journal of Greenhouse Gas Control 5S (2011) S100–S110

Contents lists available at ScienceDirect

International Journal of Greenhouse Gas Control journal homepage: www.elsevier.com/locate/ijggc

Large eddy simulation of oxy-coal combustion in an industrial combustion test facility M. Gharebaghi a , R.M.A. Irons b , L. Ma a , M. Pourkashanian a , A. Pranzitelli a,∗ a b

CFD Centre, School of Process, Environmental and Materials Engineering, University of Leeds, Leeds LS2 9JT, United Kingdom E.ON New Build and Technology, Technology Centre, Ratcliffe-on-Soar, Nottingham NG11 0EE, United Kingdom

a r t i c l e

i n f o

Article history: Received 8 February 2011 Received in revised form 6 May 2011 Accepted 16 May 2011 Available online 14 June 2011 Keywords: Pulverised coal combustion LES Oxy-fuel CCS

a b s t r a c t Oxy-fuel combustion is considered as one of the most promising technologies for carbon capture and storage (CCS). In this study, a commercial computational fluid dynamics (CFD) code has been employed for the simulation of an air-fired coal combustion and an oxy-fired coal combustion with recycled flue gas in a 1 MWth combustion test facility. Reynolds–averaged Navier–Stokes (RANS) solutions have been obtained for both cases. Results indicate that the CFD code with existing physical sub-models can provide a reasonable prediction for the air-fired combustion. However, the prediction for the oxy-fired case has not been as satisfactory as expected. In order to assess the impact of the turbulence treatment in CFD on the predictions, large eddy simulation (LES) has been performed for oxy-fired case and compared with the results from the RANS simulation and the available experimental data. Although the results suggest that LES can provide a more realistic prediction of the shape and the physical properties of the flame, there has not been significant improvement in the prediction of the temperature. In addition, the complexity of the problem requires more detailed experimental data for the validation of the LES. In order to improve the validity of numerical simulations for design purposes, further modelling improvements for oxy-coal combustion that are necessary for more accurate predictions are addressed. Based on this study, it is envisaged that the complexity in the oxy-coal combustion process requires more detailed analyses of the available physical sub-models. © 2011 Elsevier Ltd. All rights reserved.

1. Introduction The increase in the emission of the greenhouse gases to the atmosphere and the consequent policies for CO2 reduction force researchers and engineers to look for new technologies that satisfy the new more stringent emission limits introduced by the governments. Although coal-fired power plants are among the largest stationary sources of anthropogenic CO2 emission, coal is considered as a proven stable fossil fuel in both cost and supply for many years to come. Oxy-coal combustion with recycled flue gas for carbon capture and storage (CCS) is one of the candidate technologies that offer the prospect of significantly reducing CO2 emission from power generation industry. In oxy-coal combustion, the recycled flue gas moderates the high flame temperature resulting from the combustion of coal in oxygen-enriched environment. However, elevated concentrations of the combustion products, such as CO2 and the low content of nitrogen have important implications for the heat transfer, combustion chemistry and fluid dynamics in the furnace. Differences in nitrogen and CO2 thermo-chemical properties

∗ Corresponding author. Tel.: +44 113 343 3073; fax: +44 113 246 7310. E-mail address: [email protected] (A. Pranzitelli). 1750-5836/$ – see front matter © 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.ijggc.2011.05.030

lead to a different residence time, heat transfer pattern and flue gas speciation. Furthermore, a higher efficiency of the process is required to compensate, at least partially, the energy loss due to the introduction of additional devices necessary for oxy-fuel combustion, such as an air separation unit. Computational fluid dynamics (CFD) has been acknowledged as a useful design tool for many years in engineering applications (Edge et al., 2011a; Backreedy et al., 2006; Pallares et al., 2007). CFD is capable of providing detailed information on the fluid flow field and the distributions of temperature, pressure and chemical species in the system that cannot normally be obtained by experiments. Recent developments in computer resources and numerical techniques, which contribute to a faster and cost-effective data processing and grid generation, facilitate CFD with more reliability and accuracy. The authors have successfully used CFD in previous studies to predict the performance of air-coal combustion (Backreedy et al., 2006; Gharebaghi et al., 2009). Among various physical sub-models that are typically employed in the CFD modelling of coal combustion, turbulence models and modelling techniques are of considerable importance. Reynolds–averaged Navier–Stokes (RANS) equations are commonly used in engineering practices for fluid flow simulations. This is mainly because of the affordable computational costs and

M. Gharebaghi et al. / International Journal of Greenhouse Gas Control 5S (2011) S100–S110

acceptable accuracy of the predictions from the RANS simulations for engineering applications. However, RANS simulation has several shortcomings in particular in the selection of an appropriate turbulence model and the model parameters for a specific problem to simulate, and also in predicting detailed unsteady turbulent motions. In the past decade, with the continuing increase in speed and memory of computers, large eddy simulation (LES) has begun to play a significant role in understanding and predicting turbulent reacting flows. Riley (2006) has reviewed the performances of LES in typical non-premixed turbulent combustion simulations. By highlighting the importance of the selection of a sub-grid scale model and the effect of the chemistry, he suggested that LES can now be confidently applied to problems with the “mixed is burnt” assumption. Sadiki et al. (2006) have assessed the capabilities of unsteady RANS and LES for simulating gas combustion systems. They have concluded that when appropriate choices of boundary and inflow conditions are made, a good predictability can be achieved using LES. However, few studies on the application of LES for coal combustion simulation, involving flow and heterogeneous reaction of pulverised coal particles, are reported (Edge et al., 2011b; Kurose et al., 2009). Handling heterogeneous combustion including particles requires more efficient LES codes and enhanced computational resources. Recent study by the authors (Edge et al., 2011b) highlighted the improvement that can be gained by employing the LES turbulence treatment on air- and oxy-fired combustion in a smaller scale (0.5 MWth ) test facility. In this study, a further attempt has been made to compare RANS and large eddy simulation techniques in the modelling of oxy-fired pulverised coal combustion with recycled flue gas. The experimental investigation was performed in a 1 MWth combustion test facility (CTF) and the data collected from the experiments were used for CFD model input as well as model validations. A RANS solution has been obtained initially and it has been used later as a starting point for the LES. Predictions using LES have been compared with both the results from the RANS simulation and the available experimental data. Extra attention has been paid to the computational grid generation and optimization.

Fig. 1. Schematic of the CTF.

Table 1 Coal analysis.

Ultimate analysis (%) DAF

2. Experimental setup 2.1. Combustion test facility (CTF) and operating conditions The experimental investigation was carried out in E.ON’s 1 MWth CTF that is located in Ratcliffe-on-Soar, United Kingdom. This CTF has been simulated in previous studies by the authors (Backreedy et al., 2006; Gharebaghi et al., 2009) for investigating air-fired coal combustions. The level of complexity of the CTF is lower than a complete power generation furnace but still allows the simulation of the key processes that may be found in a fullscale industrial boiler. The CTF is equipped with a single wall-fired, swirl-staged low NOx burner with 12 fixed swirl vanes and three registers, namely primary, secondary and tertiary. The swirl can be adjusted by varying the proportion of the flow rates or fluxes to the different registers. The primary, secondary and tertiary swirl angles are 60, 25 and 60 degrees, respectively. A schematic of the CTF is shown in Fig. 1. The furnace wall is refractory-lined with a thermal conductivity of about 0.55 W m−1 K−1 . The walls are cooled by means of water. More detailed information on the CTF can be found in Gharebaghi et al. (2009). The coals used for the air- and oxy-fired tests and their ultimate and proximate analyses are listed in Table 1. For air-fired combustion, the Thoresby coal (UK) has been used. For oxy-fired combustion, a blend of a South-American coal (El Cerrejon) and a UK coal (Cutacre) by a mass ratio of 55/45 has been used. The selection of coals for air- and oxy-coal tests lies in the availability of

S101

Proximate analysis (%) AR

Thoresby (UK)

El Cerrejon (S. America)

Cutacre (UK)

C

80.6

78.9

79.8

H O N S

6.2 9.6 1.9 1.7

5.5 13 1.7 0.9

5 7.6 1.7 7.6

Ash

15.8

7.5

24.7

Moisture VM HCV (MJ/kg)

2.8 30 27.1

3.4 36.9 27.2

6.1 27.7 32.6

the coals at the time of tests. The coals used in this study for both tests are bituminous with a similar C/H ratio and calorific values, see Table 1. The operating conditions of the air-coal and oxy-coal combustions that are investigated in this paper are provided in Tables 2 and 3, respectively. The different feed rates between airand oxy-coal tests are due to the presence of high CO2 concentration in the flue gas, unknown air leakage and operational complications. Furthermore, because of the considerable amount of CO2 in the flue gas and in the boiler during the oxy-coal test, a different oxygen enrichment and flue gas flow rates would be required in order to achieve the same heat transfer pattern of the air-coal condition. Table 2 Operating conditions, air-fired case.

Coal Primary Secondary

Flow rate (kg/h)

Temperature (K)

131.5 ± 0.1 260 ± 4 1120 ± 12

353 353 ± 0.1 563 ± 0.8

S102

M. Gharebaghi et al. / International Journal of Greenhouse Gas Control 5S (2011) S100–S110

Table 3 Operating conditions, oxy-fired case.

Coal Primary Secondary

Flow rate (kg/h)

Temperature (K)

O2 enrichment up to (vol.%)

146.3 ± 0.8 183 ± 6 893 ± 15

353 353 ± 0.1 509 ± 0.8

– 20 ± 0.5 35 ± 0.5

For oxy-coal combustion, the primary and secondary flows are a mixture of recycled flue gas enriched with oxygen. The primary flow has a lower oxygen enrichment than the secondary flow. In addition, approximately 15% of the secondary flow (mass basis) is split from the secondary flow prior to the furnace and is reintroduced at a later stage to produce three overfiring streams. In this way, while a higher burnout is achieved, the possibility of NOx formation from the thermal and fuel paths would decrease. Due to air-ingress and variability of pressure in the CTF during the oxyfiring tests, a considerable amount of N2 is present in the recycled flue gas. Therefore, fuel-N is not the only source of NOx formation in the furnace. This additional N2 has been included in the calculation of the flow rate and the composition of the secondary flow in the CFD simulation of the oxy-coal combustion. Available measured data include the temperature and the flow rate of the fuel and oxidiser at the burner inlets, a few temperature measurements within the furnace, and the gas temperature and major species concentrations at the exhaust of the furnace. These will be discussed later in Sections 2.4 and 2.5. In addition, a video clip of the flame in the region close to the burner has been recorded during the oxy-firing tests and still images from it are used for comparison. 2.2. Flue gas recycling system For recycling of the flue gas, two streams of flow are taken from the main flue gas stream after leaving the boiler. The first stream is passed through two high-performance cyclones for particulate removal, and then it is slightly cooled and enriched with oxygen before re-entering the boiler as the secondary flow. The second split of the flue gas is blended with the primary flow as the carrier of the coal, and it has to be dried or contain very small amount of water. Therefore, it is passed through a chiller until the water content of the stream reduces to about 1% before mixing with the coal. All the flue gas is recycled back to the boiler and the actual recycle ratio is around 89%. 2.3. Stoichiometry of the test conditions For the air-coal combustion condition, the stoichiometry is set to achieve ≈3 vol.% (dry basis) of oxygen at the outlet of the boiler. The measurement is performed at the exit of the convection duct which is about 10 m after the actual exit of the boiler. This amount of oxygen indicates ≈10% of excess air required. In oxy-coal combustion with recycled flue gas, due to the presence of CO2 and H2 O in the flue gas in considerable amounts, definition of stoichiometry is slightly different from that of air-coal combustion. In oxy-coal conditions, efforts have been made to adjust the oxygen at the exit of the boiler to about 4 vol.% (dry basis). Reaching this criterion requires about 21 and 35 vol.% of oxygen content in the primary and secondary flows, respectively. 2.4. Flue gas measurements (O2 , CO and NO) For flue gas species measurement, the flue gas is extracted from the allocated ports using a suction probe and then it is quenched by water. This quenching step is done to ensure that the kinetically controlled reactions will be frozen and the measured values in the

analyzer would mirror the species concentrations in the collection point. Then the collected flue gas is dried by means of silica gel before entering the analyzer, where filters are also provided to stop the particulates. In the analyzer the concentration of O2 , CO and NO are monitored and stored in files. 2.5. Temperature measurements Temperature measurements within the furnace have been carried out using a two-colour optical pyrometer. Pyrometers of such kind have been proven to be one of the few analytical methods applicable to pulverized coal combustion. However, the presence of soot and particulates and the fluctuation of the flame, add to the errors associated with these pyrometers. It is envisaged that using this method in the cloudy environment close to the flame, would lead to measurement errors of ≈100 K. Temperatures of the cooling water around the CTF have been measured in order to monitor the level of heat flux through the furnace wall. Gas temperatures at the inlet and at the exaust are measured using thermocouples and the values are saved in electronic files. 3. Numerical modelling Accurate modelling of pulverised fuel combustion is very challenging. For air-fired combustion, with advanced turbulence models and improved fuel combustion models, reasonable predictions have been achieved (Backreedy et al., 2006; Pallares et al., 2007; Stopford, 2002) although slight discrepancies do exist in terms of the prediction of local temperatures and radiative heat transfer near to the burner. With the recirculation of flue gas to replace the air in the oxy-fired combustion conditions, significant changes take place to the furnace environment where N2 is replaced with the heavier CO2 . Different heat capacity and radiative property lead to changes in the gas velocity, temperature, heat transfer as well as char combustion processes. A significant difference in the characteristics of the flame can be clearly observed in the video clips captured during air- and oxy-coal combustion experiments. Previous simulations on air-fired combustions have also shown the importance of the intermittence in the combustion and heat transfer (Edge et al., 2011b). Experience shows that although RANS simulation can model unsteady flow to some extent, it tends to converge to a steady state solution if overall state of the process is steady. In this paper, attempts have been made to use the same physical sub-models for air- and oxy-fired combustions except for LES to simulate the oxy-fired combustion in the CTF, in particular when the flame shape, temperature and major species concentrations are of interest. The CTF has been modelled in full three dimensions employing the commercial CFD software ANSYS® FLUENT version 12.1. FLUENT has been employed for the numerical study of pulverized coal combustion previously (Backreedy et al., 2006; Edge et al., 2011b; Toporov et al., 2008) and proven to be efficient in prediction. Particular care has been given to the generation and optimisation of the numerical grid for the LES. The numerical grid is crucial for both RANS and in particular LES in order to achieve reliable CFD predictions. A considerably finer mesh has been realised for the LES of this CTF, and the same mesh has been used for the RANS simulations. Most part of the mesh employed is made of hexahedral cells. Even though tetrahedral meshes are easy and time-efficient to construct for complicated geometries, such as those involving a swirling burner with vanes, an unstructured grid made mostly of hexahedral cells is preferred in order to reduce numerical diffusivity. Due to the complexity in the geometry of the CTF, the application of a small amount of polyhedral cells has been necessary inside the burner. In terms of mesh size, a well-resolved LES

M. Gharebaghi et al. / International Journal of Greenhouse Gas Control 5S (2011) S100–S110

S103

Fig. 2. Particulars of the mesh: (a) burner and (b) mid plane section of the combustion chamber.

should aim to resolve at least 80% of the turbulence energy of the flow (Pope, 2000). Therefore the cell size distribution employed has been determined to resolve this level of turbulent kinetic energy in the energy-containing range. For this purpose a preliminary RANS simulation has been carried out on a coarser grid to estimate the distribution of the turbulence kinetic energy k and the integral length scale, k1.5 /, where  is the turbulent dissipation rate. Then the necessary cell size has been considered equal to 1/12 of the turbulence integral length scale. Unfortunately this requirement could not be satisfied inside the burner since the wall-bounded flow field in the burner is characterised by a very small turbulence integral length scale, unless a prohibitive huge number of cells were employed. We take the view that this will not have a substantial influence on the overall combustion simulations. According to the available computational resources, the criterion of resolving at least 80% of the turbulence energy has been satisfied everywhere except in the burner, giving a total cell count of approximately three million. Fig. 2 shows the mesh employed around the burner and the combustion chamber of the CTF. 3.1. Heat transfer model Heat transfer in the furnace is mainly dictated by the radiative heat transfer among gases, coal particles and the wall of the furnace. The discrete ordinate (DO) model (Chandrasekahr, 1960; Murthy and Mathur, 1998) is employed together with the weighted sum of gray gases model (WSGGM) to estimate the absorption coefficient of the gas mixture. The emissivity of the mixture of gas is dominated by the emissivity of the tri-atomic gas molecules, such as CO2 , H2 O and SO2 . The DO model is used with equal quadrature divisions in three directions (3 × 3). The number of ordinate is S8 for modelling radiation in this study. It has been found, e.g. Porter et al. (2010), that solutions of coal combustion show a low sensitivity to the increase in the pixellation to a value greater than 3. Further, particle emissivity and scattering factor of 0.8 and 0.9, respectively, have been used (Backreedy et al., 2006). 3.2. Combustion model Combustion of a blend of two coals usually does not show the same characteristics as combusting a single coal of averaged properties, in particular when the properties of the two coals are very different. Since the two coals used for oxy-fired test in this study are from the same rank, therefore, in the CFD simulation, the blend has been considered as a single coal with averaged properties. The combustion modelling takes a similar approach as Backreedy et al. (2006). A single kinetic rate devolatilisation model has been

employed based on the calculated rate parameters for the coals. Kinetic rate parameters are obtained using the FG-DVC (Fletcher et al., 1992) code. The variation of the particle mass mp is calculated as follows: −

dmp = k[mp − (1 − fv,0 )(1 − fw,0 mp,0 )] dt

(1)

where fv,0 is the mass fraction of volatiles initially present in the particle, mp,0 is the initial particle mass and k is the kinetic rate. Wet combustion is taken into account by means of fw,0 , which is the mass fraction of evaporating/boiling material (water). Enhanced devolatilisation at higher temperatures can cause the volatile yield to exceed the proximate analysis fraction. To model this, a ratio of 2 of the high temperature and the low temperature yields has been assumed. High concentration of CO2 encourages gasification reactions during char combustion in the oxy-fired combustion with recycled flue gas and ideally this effect should be taken into consideration. However, it is assumed that the gasification process is a secondary stage reaction and has been ignored in the simulation. Therefore, it is assumed that the char is first oxidised to CO and then to CO2 in the bulk of the gas mixture, and the intrinsic model of Smith (1982) has been employed. Similar values for the activation energy and the pre-exponential factor of intrinsic reactivity have been used based on Backreedy et al. (2006). A specific surface area of typically ≈25, 000 m2 /kg is assumed for the reactions on the coal particles. Turbulent volatile combustion has been calculated with a turbulence–chemistry interaction model, the eddy dissipation model (EDM) of Magnussen and Hjertager (1976), where a mixing-limited combustion is assumed. A two-step global reaction mechanism has been assumed as follows: volatiles + O2 ⇒ CO + H2 O 1 CO + O2 ⇒ CO2 2 For the RANS simulation, the chemical reaction rate is governed by the large-eddy mixing time scale, k/. Combustion proceeds whenever turbulence is present (k/ > 0). For the LES, large turbulence eddies are resolved and the reaction rate is controlled by the subgrid-scale mixing rate that is calculated as follows: −1 sgs =



where 1 Sij ≡ 2

2Sij Sij



∂uj ∂ui + ∂xj ∂xi

(2)



is the strain rate tensor (ANSYS, 2009).

(3)

S104

M. Gharebaghi et al. / International Journal of Greenhouse Gas Control 5S (2011) S100–S110

3.3. Pollutant formation

3.6. Large eddy simulation

NOx is formed both from the fuel nitrogen and the nitrogen from air-ingress. In this simulation, both thermal and fuel NOx have been calculated at the post-processing stage. Partitioning of fuel-N between char and volatiles has been derived using FG-DVC. It is assumed that the released fuel-N is converted to HCN and subsequently reacts to form NO or N2 . Regarding thermal NOx , partial-equilibrium assumption has been made for [O] and [OH] radicals calculations.

In the LES, Favre-filtered Navier–Stokes equations have been solved over the unstructured mesh of the furnace. A generic filtered ¯ variable (x) is defined by

Transport equations for the gas-phase are solved using an Eulerian approach and the coal particles are tracked in a Lagrangian frame of reference across the calculated flow field. Although in modelling pulverized coal combustion, particle–particle interaction may influence the momentum and energy transfer between gas- and solid-phases, in this study the particles interaction is assumed as negligible. In RANS calculations, for a single particle, the inertia is balanced by the forces generated through drag, gravity and other possible sources. The drag force (per mass) of the particle is calculated as follows:

(4)

where FD , , p , dp , CD and Re represent the drag force, gasphase viscosity, particle density and diameter, drag coefficient and Reynolds number, respectively. Reynolds number is calculated as follows:

(6)

where D is the fluid domain and G is the filter function that determines the scale of the resolved eddies. The finite-volume discretisation itself implicitly provides the filtering operation:



(x ) dx ,

x ∈ v

(7)

v

where V is the volume of a computational cell. The filter function G(x, x ) applied is



G(x, x ) =

1/V, 0,

x ∈ v x otherwise

(8)

Filtering the Navier–Stokes equations leads to the definition of the subgrid-scale stress: ij ≡ ui uj − u¯ i u¯ j

(9)

The subgrid-scale stresses are unknown and modelled according to the Businnesq hypothesis, computing the subgrid-scale turbulent stresses from ij −

¯ dp |up − u| Re = 

(x )G(x, x ) dx D

1 ¯ (x) = V

3.4. Particulate phase modelling

18 CD Re FD = p dp2 24



¯ (x) =

1  ı = −2t S¯ ij 3 kk ij

(10)

where t is the subgrid-scale turbulent viscosity and S¯ ij is the rateof-strain tensor for the resolved scale defined by 1 S¯ ij ≡ 2



∂u¯ j ∂u¯ i + ∂xj ∂xi



(11)

(5)

where , , dp , up and u¯ are the gas density and viscosity, particle diameter and velocity and average velocity of the gas field. In LES, the contribution of sub-grid scale fluctuations to the turbulent ¯ kinetic energy is considered through modification of u.

3.5. RANS simulation A steady-state RANS simulation has been carried out using the standard k– model (Launder and Spalding, 1972) for both air- and oxy-fired coal combustions. Standard wall functions have been employed at the walls to model the boundary layer. The velocity–pressure coupling has been handled through the SIMPLE (Patankar, 1980) algorithm. Second order upwind scheme has been used to solve the momentum, energy and species transport equations. The RANS simulation of the oxy-coal combustion has been used as initial condition for the LES. Although a fully converged solution is not necessary for this scope, both the air- and the oxycoal RANS simulations have been carried out until full convergence to compare their results with the respective LES. The convergence criteria have been set with residuals for the momentum equations less than 10−4 and scalar equations less than 10−6 . The RANS simulations have been carried out on a Linux workstation equipped with one quad-core Intel® Q9550, 2.83 GHz processor and 8GB RAM, requiring about 25 days of computational time to achieve full convergence.

The isotropic part of the subgrid-scale stresses  kk is not modelled, but added to the filtered static pressure term. To model t , the subgrid-scale model considered here is the wall-adapting local eddy-viscosity (WALE) model proposed by Nicoud and Ducros (1999). In addition, since the grid resolution in the burner is not fine enough to resolve the boundary layer on the burner walls, a wall modelling approach has been necessary, therefore the Werner and Wengle (1991) wall functions available in FLUENT for LES have been considered. The second order bounded central differencing scheme has been employed for the momentum equation, while second order upwind scheme has been used for scalars except for radiation equations. All the calculations have been advanced in time using the second order implicit scheme. The time step chosen to advance the solution is 2 × 10−4 . Such a value leads to a Courant number of less than 1 in every region of the combustion chamber. A maximum number of 30 iterations have been required in order to achieve the necessary convergence at every time step due to strong coupling of the equations. Statistics for value averaging have been collected from 0.4 to 1.0 s of the simulation. The sampling interval considered corresponds to the time step used. The LES has been performed on the Advanced Research Computing (ARC1) Linux cluster of the University of Leeds, United Kingdom, using 50 Intel® Nehalem X5560, 2.8 GHz processor cores and 2 GB RAM/core. One second of the combustion time requires about 15 days of computational time. The computational efficiency when running on 50 processor cores is around 0.7. The CPU time is dependent on the number of particles injected and on the frequency of injection.

M. Gharebaghi et al. / International Journal of Greenhouse Gas Control 5S (2011) S100–S110

S105

Table 4 Mean values at the exhaust of the modelled facility.

Air-fired Temperature (K) O2 (%, dry) CO (ppm) NO (mg/MJ) Loss of ignition (%) Oxy-fired Temperature (K) O2 (%, dry) CO (ppm) NO (mg/MJ) Loss of ignition (%)

exp.

RANS

1180 ± 50 3.1 ± 0.5 11 ± 5 310 ± 10 3.9

1232 3.7 25 340 3.0

1383 ± 50 4.5 ± 0.5 35 ± 5 151 ± 10 NA

1533 6.0 68 406 4.6

LES – – – – – 1523 6.0 50 380 6.1

3.7. Boundary conditions The primary and the secondary flows, including the overfiring streams, have been specified with mass flow inlet conditions at the furnace burner inlets according to the measurement data shown in Table 2 and Table 3. The exit of the furnace has been specified with a pressure outlet boundary condition. Average values of temperature, measured during the experimental investigations, have been used as thermal boundary conditions at the external surface of the furnace wall. Coal particles are assumed spherical with a uniform density of 1400 kg/m3 and they are injected into the furnace from the primary inlet of the burner. The size distribution of the coal particles is fitted to a Rosin-Rammler distribution ranging from 75 to 300 ␮m with a mean particle size of 125 ␮m. 4. Results 4.1. Air-fired combustion As discussed in Section 2, the experimental data available for the air-fired case consist of measured gas temperature and concentrations of O2 , CO and N2 at the exhaust and gas temperatures at a few points in the vertical part of the furnace. These measurements are compared with the RANS results in Table 4 and Fig. 3. It is observed that the predictions for the air-coal combustion using the existing models and RANS approach for turbulence are in a reasonable agreement with experimental data. Temperature values are predicted within an error margin of ≈100 K and their trend in the furnace is correctly simulated. Because of the lack of experimental data, no further comparison has been possible. Fig. 4 show the con-

Fig. 4. Air-fired combustion: (a) temperature distribution and concentrations of (b) O2 and (c) CO2 from RANS. Temperature in K. Concentrations in mass fraction.

tour plot of gas temperature, O2 and CO2 mass fractions in the CTF obtained from the RANS simulation. 4.2. Oxy-fired combustion Fig. 3. Air-fired combustion; comparison of gas temperature at the vertical mid plane for measurements and RANS predictions. Distance is referring to vertical direction from the burner centre line.

Similar to the air-fired case, measurement data for the temperature and species concentrations at the exhaust of the furnace, and a few points inside the furnace, are available for the oxy-fired case.

S106

M. Gharebaghi et al. / International Journal of Greenhouse Gas Control 5S (2011) S100–S110

Fig. 5. Oxy-fired combustion; comparison of the flame shape: (a) picture of the actual flame, (b) RANS flame, (c) LES flame (instantaneous values) and (d) LES flame (mean values). The flame is represented by the gas phase temperature contours.

Flame images that are extracted from the video clips captured during combustion tests are also available for comparison. Fig. 5 shows a comparison of an image of the actual flame (Fig. 5(a)) with the flames predicted by the RANS simulation (Fig. 5(b)) and the LES (Fig. 5(c) and (d)) of the oxy-coal combustion. The picture in Fig. 5 has been extracted from a video clip taken during the tests. It is evident, as expected, that the RANS simulation is not able to predict the turbulence structures generated downstream the burner. This leads to a longer and narrower flame than the actual one. On the contrary, the LES flame is wider in the region close to the burner and the strong turbulence in the flame region appears more evident. The shape of the flame immediately after the burner exit and its non-symmetrical nature is not captured in RANS simulation while LES predicts this clearly. In addition, the recirculation zone in the burner exit (darker area in Fig. 5(a)) is significantly more lengthened in the RANS predictions and again LES results indicate a better comparison to the still image taken from the flame. Furthermore, the mixing of the secondary flow with the coal particles and the effect of the swirls is more evident in the LES than in the RANS simulation. Fig. 6 shows a three-dimensional representation of the flame predicted by the LES in the furnace. Fig. 7 shows the contour plot of gas temperature, O2 and CO2 mass fractions from the RANS simulation in the central plane of the furnace, while Figs. 8 and 9 show the contour plot of the same variables from the LES, instantaneous and mean values, respectively. Numerical predictions of the temperature and species concentrations at the exhaust, and the temperature in the furnace, are compared with the experimental data in Table 4 and Fig. 10. Results for oxy-coal combustion from both RANS and LES show a more pronounced discrepancy with the measurements than for the airfired case. In particular, the temperature is generally overestimated, even if its trend is correctly predicted. It can be seen that although the O2 mole fraction is predicted within a ±1% difference to the experimental data, the NO concentration at the exhaust is much higher than the measurements, most likely as a consequence of the

overestimation in the temperature. The LES shows a slight improvement of the temperature predictions, but still close to the RANS results. 5. Discussion Taken into consideration the uncertainties in the experimental measurements and modelling approaches, the results presented in Table 4 and Fig. 3 show that RANS simulation of the air-fired coal combustion with existing sub-models for combustion, turbulence

Fig. 6. Oxy-fired combustion; three-dimensional representation of the flame in the furnace. Temperature in K.

M. Gharebaghi et al. / International Journal of Greenhouse Gas Control 5S (2011) S100–S110

Fig. 7. Oxy-fired combustion: (a) temperature distribution and concentrations of (b) O2 and (c) CO2 from RANS. Temperature in K. Concentrations in mass fraction.

and radiative heat transfer can provide predictions that are within acceptable error margins. Values of the species concentrations and temperature are predicted within a relatively narrow error margin in comparison to the experimental data. In addition, the trend of the gas-temperature in the chamber is correctly simulated. On the contrary, the RANS simulation of the oxy-fired coal combustion has led to a larger discrepancy in the species concentrations and the temperature compared with the available measured data, see Table 4 and Fig. 3. This is observed in particular for the temperature,

S107

Fig. 8. Oxy-fired combustion: (a) temperature distribution and concentrations of (b) O2 and (c) CO2 from LES (instantaneous values). Temperature in K. Concentrations in mass fraction.

although the trend of the temperature profile in the combustion chamber is correctly predicted. Because of the discrepancy in the predicted major species and in particular the temperature, a high NOx prediction is subsequently obtained. The O2 concentration in the air-fired condition showed reasonable agreement with the experimental data and a lower carbon in ash value is obtained under the higher temperature. Under the oxy-coal combustion, the overpredictions of O2 concentration and gas temperature are attributed

S108

M. Gharebaghi et al. / International Journal of Greenhouse Gas Control 5S (2011) S100–S110

Fig. 10. Oxy-fired combustion; comparison of gas temperature at the vertical mid plane for measurements, RANS and mean LES predictions. Distance is referring to vertical direction from the burner centre line.

Fig. 9. Oxy-fired combustion: (a) temperature distribution and concentrations of (b) O2 and (c) CO2 from LES (mean values). Temperature in K. Concentrations in mass fraction.

to the possible error associated with the char combustion and radiation models and a higher carbon content in ash has been observed compared to the air-fired case, see Table 4. Due to practical problems, the fly ash samples collected from the oxy-coal combustion were not suitable for loss of ignition measurements and therefore in Table 4 only the calculated values for loss of ignition of the oxyfired condition are provided. In oxy-coal conditions, the effect of C-CO2 and C-H2 O reactions are not considered, which would affect

the prediction of loss of ignition rate and overall heat transfer between the particle and the gas phase. These gasification reactions are endothermic in nature and therefore it is envisaged that considering these reactions would contribute to lowering the gas temperature along with the improvements from updating the radiation model. In oxy-combustion simulation, similar results to the RANS simulation have been obtained from the LES. The temperature is still overestimated, but its trend is again correctly predicted. While the RANS simulation of the air-coal combustion is able to predict both the trend and the values of the temperature within a relatively narrow error margin, the RANS and the LES of the oxycoal combustion show a general overestimation of the temperature values. More accurate treatment of the turbulence using the LES provides a clear improvement in the simulation of the flow field in the combustion chamber, giving a more accurate reproduction of the flame. Nevertheless, it seems to be evident that the LES by itself is not able to lead to a substantial improvement in the accuracy of the predictions of species concentrations and temperature for an oxy-coal combustion. The large discrepancy observed for oxy-coal combustion may be partially due to errors in the measurement data with the application of the optical pyrometer in the presence of the burning particles. It is envisaged that the errors corresponding to the measurements might be higher than the values shown in Fig. 10. However, inaccuracy in model assumptions, in particular for the radiative heat transfer calculations, is considered to be the main contributor to the discrepancy. One of the important factors for the accurate prediction of the gas temperature in the furnace is the correct calculation of the heat transfer using the radiation model and thermal properties of the gaseous mixture. It has been found that the gas temperature prediction in both the RANS and LES cases are highly sensitive to the radiation model. The reason for the over-prediction of the temperature is believed to be the fact that the effect of high CO2 concentration in the combustion chamber on the radiative heat transfer has not been fully considered. It has been demonstrated (Porter et al., 2010) that the use of the standard gray-WSGGM correlated with coefficients for fuel-air combustion can lead to substantial errors in the calculation of the radiative source term and hence greatly alters the predicted temperature distributions in the system in fully coupled calculations. The use of a non-gray spectral model is therefore essential in order to improve the accuracy of the temperature and heat transfer prediction in the simulation of oxy-fuel combustion. The intrinsic coal combustion model and the model constants employed in this paper have been obtained for air-fired coal combustion simulations. Similar to what has been shown in previous studies (Backreedy et al., 2006; Pallares et al., 2007; Stopford, 2002), the models employed here are reliable for

M. Gharebaghi et al. / International Journal of Greenhouse Gas Control 5S (2011) S100–S110

air-coal combustion and reasonable results have been obtained for species and temperature predictions, although an error margin of ≈50–100 K has been observed for temperature predictions. In this study, the same physical sub-models as used for air-coal combustion have been used for modelling the oxy-coal combustion. Therefore, some deviations in modelling oxy-coal combustion using air-coal models are foreseen, considering the higher temperature, enriched oxygen level and the presence of high concentrations of CO2 in the combustion chamber for the oxy-fired case. With the use of a substantially finer and optimised computational mesh, as well as the use of the DO model in this study, it is noted that an improved prediction for the trend of the temperature distribution in the region shown in Fig. 10 has been achieved compared to the air-fired case. However, it is evident that improving the computational mesh only, but still incorporating the sub-models developed for air-firing, would be insufficient to accurately simulate oxy-firing scenarios. The numerical simulations presented here for the oxy-fired pulverised coal combustion have pointed out the need for improvement and development of specific physical sub-models, and have raised several challenges for oxy-fuel combustion simulations: 1. Chemistry: The presence of CO2 attributes to a different combustion pattern by strengthening the char gasification reactions. These reactions are slower by three orders of magnitude in the flame temperature, however, they are endothermic and would affect the overall heat of reaction. 2. Heat transfer: Due to the high concentration of CO2 (>60%) in the flue gas, it is essential to consider a non-gray gas model in order to correctly predict the absorption and the emission and subsequently the temperature in the furnace (Coelho, 2007). Also, CO2 has a different heat capacity in comparison with N2 and therefore, a different convective heat transfer pattern from an air-fired combustion is expected in the furnace. In this study, measured cooling water temperatures have been used as boundary conditions for calculating the heat flux through the furnace wall. However, larger temperature discrepancies, both inside the furnace and at the furnace exit, have been observed for the oxy-coal case than the air-coal case, indicating a larger error in the heat transfer calculations for the oxy-coal combustion. 3. Turbulence-radiation interaction: It has been reported that radiation smoothes the temperature fluctuations and its effect is strengthened by the increase of the temperature. On the other hand, the effect of the turbulence on the radiation in the flame, and the interaction of radiation, density and temperature fields with the turbulence, should ideally be taken into consideration. However, this is a very complex task and involves high computational requirements for coupling of turbulence and radiation. Currently, there is not a generally accepted turbulence-radiation interaction model available for direct numerical simulation (DNS) or LES of reacting flow. The use of time-averaged solutions for the radiative transfer equations is efficient, but requires detailed probability density functions for relevant variables and a number of assumptions. Such an approach is still a promising candidate for the future. 6. Conclusions In this study, a combustion test facility has been simulated using RANS simulation and LES. Considering the uncertainties in the experimental measurements and the physical sub-models, it has been found that RANS simulation provides reasonable predictions for air-fired condition, using the existing physical sub-models. However, calculations for the oxy-fired combustion using RANS, showed a significant quantitative deviation from the experimen-

S109

tal data. An attempt has been made to improve the predictions by applying large eddy simulation for an advanced turbulence treatment. Due to its more accurate representation of the turbulence, the LES has shown a more accurate reproduction of the flame behaviour and characteristics compared to the RANS simulation. However, there was no significant improvement observed in the temperature predictions. The reason lies in the fact that although the temperature prediction is influenced by the turbulence simulation, the extent of the improvement will be dependent on many other physical sub-models such as heat transfer and char combustion. In oxy-fuel combustion of pulverised coal with recycled flue gas, it is critical to employ an advanced radiative heat transfer model taking into considerations the non-gray effect in the presence of high concentrations of CO2 and H2 O. The gray gases assumption is not able to correctly predict the radiative heat transfer in an environment with a high concentration of CO2 and H2 O. In addition, an upgrade of the existing coal combustion sub-models, a more detailed reaction mechanism and a more accurate turbulent combustion model are necessary. Authors believe that in modelling oxy-coal combustion, once LES approach is joined with accurate radiation and combustion models, the improvements in the predictions would be more significant. Acknowledgements Financial support from UK EPSRC (OXY-CAP UK/RCUK-China), Dorothy Hodgkin Postgraduate Awards/EON is gratefully acknowledged. The authors would like to thank Dr Domenico Caridi, ANSYS UK Ltd. for his useful suggestions. References ANSYS Inc., 2009. ANSYS FLUENT Theory Guide. ANSYS Inc., USA. Backreedy, R., Fletcher, L., Ma, L., Pourkashanian, M., Williams, A., 2006. Modelling pulverised coal combustion using a detailed coal combustion model. Combustion Science and Technology 178, 763–787. Chandrasekahr, S., 1960. Radiative Transfer. Dover Publications. Coelho, P., 2007. Numerical simulation of the interaction between turbulence and radiation in reactive flows. Progress in Energy and Combustion Science 33, 311–383. Edge, P., Gubba, S., Ma, L., Porter, R., Pourkashanian, M., Williams, A., 2011. LES modelling of air and oxy-fuel pulverised coal combustion—impact on flame properties. Proceedings of the Combustion Institute 33, 2709–2716. Edge, P., Gharebaghi, M., Irons, R., Porter, R., Porter, R., Pourkashanian, M., Smith, D., Stephenson, P., Williams, A., in press. Combustion modelling opportunities and challenges for oxy-coal carbon capture technology. Chemical Engineering Research and Design, available online. Fletcher, T., Kerstein, A., Pugmire, R., Solum, M.S., Grant, D., 1992. Chemical percolation model for devolatilisation. 3. Direct use of carbon-13 NMR data to predict effects of coal type. Energy and Fuels 6, 414–431. Gharebaghi, M., Goh, B., Jones, J., Ma, L., Pourkashanian, M., Williams, A., 2009. Numerical Investigation of Char Reactivity in Oxy-coal Combustion in Carbon Capture and Sequestration Technologies. ASME-Turbo Expo, Orlando, FL, USA. Kurose, R., Watanabe, H., Makino, H., 2009. Numerical simulations of pulverized coal combustion. KONA Powder and Particle Journal, 144–156. Launder, B., Spalding, D., 1972. Lectures in Mathematical Models of Turbulence. Academic Press, London, England. Magnussen, B., Hjertager, B., 1976. On mathematical models of turbulent combustion with special emphasis on soot formation and combustion. In: 16th International Symposium on Combustion ,. The Combustion Institute, pp. 719–729. Murthy, J., Mathur, S., 1998. Finite volume method for radiative heat transfer using unstructured meshes. Journal of Thermophysics and Heat Transfer 12, 313–321. Nicoud, F., Ducros, F., 1999. Subgrid-scale stress modelling based on the square of the velocity gradient tensor. Flow, Turbulence and Combustion 62, 183–200. Pallares, J., Arauzo, I., Williams, A., 2007. Integration of CFD codes and advanced combustion models for quantitative burnout determination. Fuel 86, 2283–2290. Patankar, S., 1980. Numerical Heat Transfer and Fluid Flow. Taylor & Francis. Pope, S.B., 2000. Turbulent Flows. Cambridge University Press. Porter, R., Liu, F., Pourkashanian, M., Williams, A., Smith, D., 2010. Evaluation of solution methods for radiative heat transfer in gaseous oxy-fuel combustion environments. Journal of Quantitative Spectroscopy and Radiative Transfer 111, 2084–2094. Riley, J., 2006. Review of large eddy simulation of non-premixed turbulent combustion. Journal of Fluid Engineering 128, 209–215.

S110

M. Gharebaghi et al. / International Journal of Greenhouse Gas Control 5S (2011) S100–S110

Sadiki, A., Maltsev, A., Wegner, B., Flemming, B., Kempf, A., Janicka, J., 2006. Unsteady methods (URANS and LES) for simulation of combustion systems. International Journal of Thermal Sciences 45, 760–773. Smith, I., 1982. The combustion rates of coal chars: a review. In: 19th International Symposium on Combustion ,. The Combustion Institute, pp. 1045–1065. Stopford, P., 2002. Recent applications of CFD modelling in the power generation and combustion industries. Applied Mathematical Modelling 26, 351–374.

Toporov, D., Bocian, P., Heil, P., Kellermann, A., Stadler, H., Tschunko, S., Förster, M., Kneer, R., 2008. Detailed investigation of a pulverized fuel swirl flame in CO2 /O2 atmosphere. Combustion and Flame 155, 605–618. Werner, H., Wengle, H., 1991. Large-eddy simulation of turbulent flow over and around a cube in a plate channel. In: 8th Symposium on Turbulent Shear Flows, Munich, Germany.