8.1 Experimental Case Setup
This chapter focuses on the simulation results of the semi-industrial scale coal fur-nance No 1, studied experimentally by the International Flame Research Founda-tion (IFRF). In the experimental studies of Weber et al. (1992) the furnace was equipped with the Aerodynamically Air Staged Burner (AASB), in the following referred to as IFRF Burner. The IFRF burner, as shown in Fig. 8.1, consists of two annular injectors, where the inner coal injector (width 13mm) provides the coal feed and transport air and the outer injector (width 47mm) issues the pre-heated swirling combustion air with a swirl number of S≈1. Combustion air and coal are injected into a quarl, which has a length to diameter ratio of unity (i.e.
Lquarl =Dquarl=234mm). The furnace has a total length of 6.25m and a cross-section of 2m×2m. The operating conditions of the burner are summarised in Tab. 8.1. The fired coal was a Göttelborn high volatile bituminous B (HVBB), the PA and UA of which are given in Tab. 8.2.
Measurements are performed using LDV for particle velocities with the coal par-ticles as tracers and ceramic shielded thermocouples for gas phase temperatures.
For the measurement of species concentrations gas samples are extracted from the furnace using a sampling probe. Concentrations of O2are evaluated by a paramag-netic analyser, CO and CO2are evaluated by an infrared analyser and NOx, HCN, NH3, N2O by a chemiluminescent analyser.
125
8.1.EXPERIMENTALCASESETUP126
z=0 x
z=ze A
B
scaling=2x
Quarl cross-section along the A-B plane, scaling=2x
234Comb.AirOuter
140Comb.AirInner
134CoalInletOuter 108CoalInletInner 468 1000
LFurnace LQuarl
πLQuarl πLQuarl
PCC-LES Domain Swirled DES
Domain Transition
DES Domain
Map Map+Swirl
Figure 8.1: Computational domains for the IFRF furnace simulation, dimensions given in [mm]
The IFRF furnace has been subject of several numerical studies, e.g. RANS by Weber et al. (1992), Peters and Weber (1997) and LES by Olenik et al. (2015), Rieth et al. (2016), and Rieth et al. (2017b). Using the particle properties given in Tab. 7.1 and with the help of preliminary simulations the left hand side of the diluteness criterion Eq. (3.11) has been evaluated as ≈0.03<<1. Thus, the IFRF case can be seen as a dilute particle jet and particle-particle interactions can be neglected, which makes the IFRF case suitable for simulations employing the EUL-LAG approach described in chapter 5.
Table 8.1: Experimental conditions, Weber et al. (1992)
Coal Type Göttelborn hvBb
Pulverized-coal Feed Rate 263kgh−1 Transport Air Flow Rate 421kgh−1(@343K) Combustion Air Flow Rate 2670kgh−1(@573K) Combustion Air Reynolds Number 0.8×105 Combustion Air Swirl Number 0.93
Table 8.2: Properties of Göttelborn hvBb Coal, UA (daf) left, PA (as received) right Weber et al. (1992)
Content [wt %]
Carbon 79.3
Oxygen 13.7
Hydrogen 4.7
Nitrogen 1.3
Sulphur 1.0
Low Heating Value 32.32 MJ/kg
Content [wt %]
Ash 8.3
Volatile Matter 37.0 Fixed Carbon 52.5
Moisture 2.0
8.2 Computational Case Setup
As for the CRIEPI case, the following strategy was applied to obtain accurate and stable simulation results. First, a base case setup was developed, here referred to as BC. Details of the base case setup, corresponding results, and their discussion were previously published by Olenik et al. (2015) and are given here mainly as an additional reference for the most recent simulation results. In a second step, based on the BC a revised setup, referred to as RC, for parameter studies was developed.
8.2. COMPUTATIONAL CASE SETUP 128 Then a number of parameter studies were conducted to find an optimal computa-tional setup for the final simulations. The setup of the BC and the RC case are given in Tab. 8.3.
Table 8.3: Summary of model parameters for the BC and RC cases.
Case BC RC
Computational grid
Number of cells 1M 1.3M
Domain length 2m 4m
Furnace wall temperature 1200 [K] 1400 [K]
Computational setup
Turbulence model Smagorinsky Type
(cf. Sec. 5.1.2)
σ-Model, Nicoud et al.
(2011) Interpolation schemes,
convection term, momentum
equation TVD CDS
Interpolation schemes,
species and enthalpy TVD TVD
Axial inflow bulk velocity 22 ms−1 44 ms−1
General coal particle models
Particle heat capacity OpenFOAM default Merrick (1983) (cf. App. D.3.6) Char conversion
COxidationKineticDif-fusionLimitedRate Baum & Street Pyrolysis model Badzioch & Hawksley
(SKR, cf. Sec. 6.2.2) Richards & Fletcher (RF, cf. Sec. 6.2.4)
Particle dispersion none Bini & Jones *
(cf. Sec. 5.3.2)1
Parcel injection rate 2×106 0.5×106
Pyrolysis modelling
Volatile species CaHbOcNd CaHbOcNd
Homogenous
reactions R36, R37 R33
Molar mass CaHbOcNdi 50 30
JANAF polynomial C3H8 adapted C2H6
However, due to the large computational costs, fewer parameters compared to the CRIEPI case, presented in chapter 7, have been investigated. Investigated parameters include the effect of the domain length and grid resolution, statisti-cal convergence, inflow bulk velocity, and the influence of the particle dispersion model. These parameter studies are discussed in detail appendices D.4.1, D.4.1, D.4.2, D.4.3, and D.4.4, cf. Tab. 8.4. The central conclusions of theses parame-ter studies helped to define the computational setup of the final simulations which
are discussed later in this section. The grid density study demonstrated that a computational grid of approximately 13.4M cells, corresponding to a cell size of ap-proximately 2 mm within the quarl is a good compromise between the constraining computational costs and accuracy of the LES results. As shown by the domain length study a computational domain length of 4 m suffices to avoid interferences from the outlet boundary on the sampled data. Furthermore, the statistical con-vergence study shows that a sufficient concon-vergence of the slowly converging scalars such as species mass fractions can be expected after 10 s of simulation time. Be-yond that a sampling window of 0.75 s gives reasonably smooth statistical data.
The inflow bulk velocity study demonstrates the impact of uncertainties within the combustion air mass flow rate on the global combustion process. The particle dis-persion model study shows a limited effect of particle disdis-persion within the tested parameter range.
Table 8.4: IFRF parameter studies.
Study Appendix
Domain length D.4.1
Computational grid resolution D.4.1 Statistical convergence D.4.2 Inflow bulk velocity D.4.3 Particle dispersion model D.4.4
Simulations were conducted in two stages. First, a simulation on a very coarse grid containing approximately 100k grid cells was conducted for 10 s of simulation time. This initial simulation is needed to approximate slowly converging quantities such as species concentrations in the recirculation zones near the furnace walls.
The results of the coarse simulation are then mapped to an intermediate grid (re-finement level 7, 4.5M grid cells, cf. appendix D.3.1 for more details on the mesh refinement). The simulations were then restarted and run for another 2 s. Finally, the intermediate simulation results are mapped to a fine grid (refinement level 10, 13.4M grid cells) and run for another 1.0 s with a sampling period of 0.7 s. Further details of the employed grids are given in appendix D.4.1. Compute times on the Cray XC40 are 1 day, 3 days, and 5 days on 48, 240, and 480 Intel Xeon E5-2680 CPU cores for the coarse, intermediate, and fine simulations,2, respectively. The approach of splitting the simulation in a preliminary simulation on a coarse grid and mapping to finer grids is similar to the approach of Rieth et al. (2016).
How-2Excluding the time spent for simulation on the previous coarser grid.
8.2. COMPUTATIONAL CASE SETUP 130 ever, due to the inferior serial and parallel efficiency of OpenFOAM compared to simulation code PsiPhiused by Rieth et al. (2016), simulations were restricted to fewer core hours. This limited the simulations to fewer grid cells. Additionally, it motivated a cell stretching ratio of 5 between smallest and largest cells to obtain a sufficiently fine mesh in the quarl region. All simulations employed CFL numbers between 0.4 and 0.2, which corresponds to an approximate time step of 1×10−4s, 2×10−5s, and 1×10−5s for the coarse, intermediate, and fine simulation. Table 8.5 summarises the simulation times and computational costs of the different stages.
Table 8.5: Number of grid cells and various simulation properties for different mesh levels for a domain length of 4m.
Stage coarse medium fine
Level [−] 2 7 10
Grid Cells [−] 100k 4.5M 13.4M
∆x[mm] 10 2.8 2.0
Simulation time [s] 10 2.0 1.0
Sampling start [s] 2.0 1.0 0.3
Number of CPU cores [−] 48 240 480
Compute time [d] 1 3 5
Cells per CPU core [−] 2000 18750 27916 Time step size∆t[s] 1×10−4 2×10−5 1×10−5
For the simulations the furnace geometry has been simplified by placing the burner in the centre of the furnace. Furthermore, the furnace cross section has been assumed to be rectangular with a cross section of 2 m×2 m. Additionally, the inlet zone, consisting of the combustion air annulus, upstream of the quarl has been separated to an extra domain, as shown in Fig. 8.1. On the separate domains turbulent swirled LES inflow data is generated using detached eddy simulations (DES). Details of this approach are given in the subsequent section 8.2.1 and in appendix A.8.2.
The boundary conditions for velocity, pressure and temperature are summarised in Tab. 8.6 and the ones for species mass fractions in Tab. 8.7. A typical combination of Dirichlet and Neumann boundary conditions is chosen. For the velocity and tem-perature fields Dirichlet boundaries for all except the outlet boundary are taken.
For the pressure field Neumann boundary conditions for all except the outlet are chosen.
As shown in Tab. 8.7, fresh air is injected through the coal and combustion air inlet, in the following referred to as coal inlet. To ignite the mixture of volatile gases
Table 8.6: Boundary conditions, where N indicates Neumann and D Dirichlet boundaries
Boundary Velocity
[ms−1]
Pressure [kgm−1s−2]
Temperature [K]
Coal inlet D, 23 N D, 343
Combustion air inlet D, mapped N D, 573
Furnace walls D, No-slip N D, 1400
Bluff body wall D, No-slip N D, 800
Quarl wall D, No-slip N D, 800
Outlet N D, 1e5 N
and to stabilise the flame a small amount of products is injected via the combustion air inlet and coal inlet, as needed for the EBU TCI model. A Neumann boundary condition is used for the walls and the outlet. The domain is initialised as stagnant hot combustion products with YCO2=0.215, YH2O=0.040, YN2=0.698, and YO2= 0.047. The species mass fractions are obtained assuming a chemical equilibrium and a complete conversion of the coal volatiles and char.
Table 8.7: Boundary conditions of species mass fractions, where N indicates Neu-mann andDDirichlet boundaries
Boundary O2 N2 CO2 H2O
Coal inlet D, 0.23 D, 0.75 D, 0.01 D, 0.01 Combustion air inlet D, 0.23 D, 0.75 D, 0.01 D, 0.01
Furnace walls N N N N
Bluff body wall N N N N
Quarl wall N N N N
Outlet N N N N
Figure 8.2 shows the employed computational grid and indicates the locations of the various boundary conditions. The mesh shown is the coarsest mesh with approximately 100k cells and a moderate mesh stretching of 5 in axial direction.
Additionally, a close-up of the quarl region of the fine mesh is given.
8.2. COMPUTATIONAL CASE SETUP 132
outlet
furnace walls furnace walls
bluff body quarl wall comb. air
comb. air coal coal
Figure 8.2: Schematic of the computational grid. Shown is a close-up of the first 2 m of the coarse mesh, level 2 (top), a close-up of the quarl of the fine mesh, level 10 (bottom, left), and a 3D view of the coarse mesh (bottom, right).
For the simulations a Rosin-Rammler distribution with a corresponding spread parameter of n=3.5 was selected. As can be seen from Fig. 8.3 the particle size dis-tribution shows a reasonable agreement with the reported measurements of Weber et al. (1992).
0.00 0.33 0.67
0 42 83 124
CDF[-]
Particle Diameter [µm]
Weber92LES
Figure 8.3: LES and experimental profiles of particle size distribution (CDF)
In Tab. 8.8 a summary of the selected models and numerical parameters of the refined model in comparison to the setup of the BC case is given. A number of general enhancements, e.g. refinement of the computational grid, extension of the domain length, selection of the σ turbulence model, adaption of the furnace wall temperature, application of the CDS interpolation scheme for convection terms, and correction of the combustion air bulk velocity according to the nominal value have been implemented. Furthermore, several Lagrangian particle models have been improved, this includes the implementation of the Merrick (1983), cf. the ap-pendix Eq. (D.5), model for the particle heat capacity, and the implementation of the Richards and Fletcher (2016), cf. Sec. 6.2.4 pyrolysis model. Additionally, the parcel injection rate was reduced in order to reduce the computational costs. The selected parcel injection rate of 0.5×106s−1, in combination with a Rosin-Rammler shape parameter of n=3.5 leads to approximately 2500 particles per parcel, which results in stable simulations even for the finest tested mesh.
Note that the BC case does not include any particle dispersion model. The im-plementation of the dispersion model for cases RF and RF2S was incorrect and overestimated the dispersion given by Bini & Jones by a factor of 1/p
∆t. Thus the dispersion model is denoted as Bini & Jones * in Tab. 8.3. However, the parameter analysis of the CRIEPI case and additional tests carried out for the RC setup, cf. ap-pendix D.4.4, demonstrate that the effect of the dispersion model is very small and can be neglected. The significant computational costs associated with the highly resolved IFRF computations do not justify new simulations, and results with an overestimated (but still negligibly small) particle dispersion are shown in the re-mainder of this thesis.
The final set of simulations consists of three cases, the BC case, the postulate substance Richards & Fletcher (RF) case, and the postulate substance Richards &
Fletcher 2 Separate Species (RF2S) case with two separate postulate substances.
8.2. COMPUTATIONAL CASE SETUP 134
Table 8.8: Summary of model parameters for the IFRF case
Case BC RF RF2S
Computational grid
Number of cells 1M 13.4M 13.4M
Domain length 2m 4m 4m
Furnace wall
temperature 1200 [K] 1400 [K] 1400 [K]
Computational setup Turbulence model Smagorinsky
Type (cf. Sec. 5.1.2)
σ-Model, Nicoud et al. (2011)
σ-Model, Nicoud et al. (2011) Interpolation schemes,
convection term,
momentum equation TVD CDS CDS
Interpolation schemes,
species and enthalpy TVD TVD TVD
Axial inflow bulk
velocity 22 ms−1 44 ms−1 44 ms−1
General coal particle models Particle heat capacity OpenFOAM
default
Merrick (1983) (cf. App. D.3.6)
Merrick (1983) (cf. App. D.3.6) Pyrolysis model Badzioch &
Hawksley (SKR, cf. Sec. 6.2.2)
Richards &
Fletcher (RF, cf. Sec. 6.2.4)
Richards &
Fletcher (RF, cf. Sec. 6.2.4) Particle dispersion none Bini & Jones *
(cf. Sec. 5.3.2) Bini & Jones * (cf. Sec. 5.3.2)
Parcel injection rate 2×106 0.5×106 0.5×106
Char conversion COxidationKinet-
icDiffusionLimit-edRate Baum & Street Baum & Street Pyrolysis modelling
Volatile species CaHbOcNd CaHbOcNd CaHbOcNd1, CaHbOcNd2 Homogenous
reactions R36, R37 R33 R34, R35
Molar mass
CaHbOcNdi 50 30 30, 44
JANAF polynomial C3H8 adapted C2H6 adapted C2H6, C3H8
Input Q-factor 1.5722 1.3883
-The following set of homogeneous reactions is considered for the simulations.
The volatile species reacts to H2O and CO2according to reaction (R33) for the single species Richards & Fletcher model and according to reactions (R34-R35) in case of the two separate postulate substances approach. In both cases the stoichiometric
coefficients are determined according to Eq. (6.46).
CaHbOcNd+1.9561 O2→1.5134 CO2+1.399 H2O+0.027849 N2 (R33) CaHbOcNd1+1.2446 O2→0.22049 CO2+3.2369 H2O+0.064432 N2 (R34) CaHbOcNd2+3.2622 O2→3.2622 CO2+1.0362 H2O+0.020625 N2. (R35) The BC simulations employed a two step reaction mechanism, reactions (R36-R37), where the postulate substance reacts to CO and subsequently to CO2 in a second reaction.
CaHbOcNd+1.9825 O2→2.97 CO+1.125 H2O+0.015 N2 (R36)
CO+0.5 O2→CO2. (R37)
Comparison of the single postulate substance reaction (R33) to the reaction used in the BC dataset, reaction (R36), shows that for the present RF simulation a smaller CO2to H2O ratio is used. This is a result of a lower input3Q-factor evalu-ated with PKP compared to the Q-factor reported by Knill et al. (1989). The effect of the pyrolysis model and the different Q-factor is discussed further in section 8.4.
The compostions and corresponding lower heating values of the considered postu-late substances are given Tab. 8.9.
Table 8.9: Elemental compositions and lower heating values of the postulate sub-stances CaHbOcNd, CaHbOcNd1, and CaHbOcNd2.
Carbon Hydrogen Oxygen Nitrogen lhv [MJkg−1]
CaHbOcNd 0.60 0.09 0.28 0.03 25.47
CaHbOcNd1 0.09 0.22 0.63 0.06 16.12 CaHbOcNd2 0.80 0.05 0.14 0.01 29.01
Reaction rates of the heterogeneous chemistry were modelled by the model of Baum and Street (1971) in contrast to the previously em-ployed OpenFOAM specific implemention of the char conversion model called COxidationKineticDiffusionLimitedRate. The differences in the implementa-tion and effects on the LES results are discussed in the appendix D.3.12. In both cases, however, a global reaction step mechanism is used,
C+O2→CO2, (R38)
3Averaged over the evaluated heating rate range.
8.2. COMPUTATIONAL CASE SETUP 136 where CO2is directly emitted to the gas-phase.
8.2.1 Inflow Data Generation
In order to reduce computational costs, the computational domain is split into three separate sub-domains, namely the transition DES domain, the swirl DES domain, and the PCC-LES domain as shown in Fig. 8.1. The DES domains were employed to generate transient LES inflow data for the combustion air inlet. In a first step turbulent unswirled inflow data was generated by continuously mapping the veloc-ity field from the transition domain outlet to the transition inlet via cyclic boundary conditions while keeping a specific pressure gradient between both boundaries in order to enforce the correct mass flow rate of 2670kgh−1. After approximately 15 flow through times (FTT) the velocity data was written out over a period of 0.75 s, corresponding to 45 FTT using thelibInflowlibrary from thepipeFoamrepository.
Instantaneous velocity fields of the unswirled DES case are shown in Fig. 8.4 (left).
Unswirled Swirled
Figure 8.4: Snapshots of the instantaneous axial velocity component (left) and the swirling velocity component (right) of the unswirled and the superimposed swirled inflow within the annulus as calculated by means of DES. The axis of symmetry of the inflow annulus is indicated by the dashed lines.
In a second step a swirling velocity component was superimposed to the unswirled DES velocity data at the inlet of the swirling DES domain. With the swirling inflow velocity a second DES was performed for another 1 s. After 0.25 s the velocity data at the annular outlet was written out. The instantaneous velocity fields of axial and swirling velocity component along the annulus for the swirled DES are displayed in Fig. 8.4 (right). Further details of the DES generating the inflow data are summarised in Tab. 8.10. The swirled inflow data is then mapped to
the PCC-LES domain for the final simulations. This procedure allows one to reduce the computational costs since the LES data needs to be generated only once and can subsequently be reused for several parameter studies of the PCC-LES case. Addi-tionally, this procedure makes the PCC-LES independent of the mesh constraints of the transitional DES domain, i.e. small cell sizes in the transitional layer to allow initial disturbances to grow. Furthermore, the complexity of the block-structured PCC-LES grid is reduced, which results in homogeneous cell sizes improving the numerical stability of the simulations.
Table 8.10: Computational setup of the inflow generation DES
Number of Cells 2.8M
Simulation domain length πLQuarl Average y+inner wall 2.79 Average y+outer wall 3.21 Turbulence model Spalart (2009)
8.3 General Flow and Flame Structure
In this section an overview of the structure of the flame is given with a discussion of the main features of the velocity, temperature and species mass fraction fields.
This discussion is based on the results of the RF2S case. By inspection of the axial velocity fields, shown in Fig. 8.5, two recirculation zones can be identified. Firstly, an internal recirculation zone (IRZ) within the quarl diameter and secondly an ex-ternal recirculation zone (ERZ) close to the corners and along the lateral furnace walls. The negative axial velocity caused by the vortex breakdown in the IRZ re-circulates hot combustion products into the quarl. The recirculated hot combustion gases are an important factor for the flame stabilisation by heating the injected coal particles. The IRZ extends up to z≈0.5m downstream. The ERZ which has lower negative peak velocities compared to the IRZ expands further downstream.
8.3. GENERAL FLOW AND FLAME STRUCTURE 138
0 0.5
1 1.5
2
z
−0.5 −0.25 0r 0.25 0.5
−0.5 −0.25 0r 0.25 0.5
Figure 8.5: Snapshots of the instantaneous axial velocity component (left) and the swirling velocity component (right). For better visibility a close-up of the quarl region is shown. Dimensions given in [m].
Figure 8.6 shows the instantaneous gas-phase temperature field from the LES.
It can be seen that the ignition starts inside the quarl in the shear layer between the combustion air stream and the coal particle inlet. This is likely a consequence of the EBU TCI model, since here the necessary prerequisites, shear and non-zero mass fractions for fuel, oxidiser, and products are present. The thin reaction layer widens atz≈0.25m where the flame sheet propagates towards the stagnation point at the centreline. Parts of the hot combustion products exit the flame in downstream direction, while a part is recirculated towards the quarl. Instantaneous peak tem-peratures within the shear layer reach approximately 2600K. Furthermore, it can be seen that the flame is surrounded by a zone of low temperature fresh gases from the combustion air inlet. This surrounded layer of relatively cold gases persists up toz≈0.4m before the mixing with the hot furnace gases increases the temperature considerably.
0 0.5
1 1.5
2
z
−0.5 −0.25 0r 0.25 0.5
Figure 8.6: Snapshot of the instantaneous temperature field. Dimensions given in [m].
Figure 8.7 shows snapshots of the instantaneous species mass fraction fields of CO2, O2 and the volatile species CaHbOcNd1 and CaHbOcNd2. Within the IRZ the lowest mass fraction of O2and highest mass fractions of CO2can be observed. Ad-ditionally, a considerable amount of excess O2 exits the flame. The O2 is in part consumed by the subsequent char conversion and to some extent exits the furnace at the outlet. The volatile species are almost exclusively present inside and in the close proximity of the quarl. The V-shape of the volatiles indicates that the pyrol-ysis process is completed and the majority of the volatiles burned before the coal particles are recirculated. Finally, it can be seen that the high temperature volatile yield CaHbOcNd2is dominant especially in the hot shear layers. However, as shown in the subsequent discussion, a non-negligible amount of the low temperature yield CaHbOcNd1can also be found outside of the quarl region.
8.4. SENSITIVITY TO THE PYROLYSIS MODEL 140
YO2 YCO2
YCaHbOcNd1
YCO2
YCaHbOcNd2 0
0.5 1 1.5
2
z
0 0.5
1 1.5
2
z
−0.5−0.25 0 0.25 0.5r
−0.5−0.25 0 0.25 0.5r
Figure 8.7: Snapshots of the instantaneous species mass fraction fields. Di-mensions given in [m].
8.4 Sensitivity to the Pyrolysis Model
This section discusses the influence of different pyrolysis models and model coeffi-cients on the pyrolysis process and the LES results. For this, three parameter sets are selected. For the BC dataset the SKR model coefficients were manually opti-mised based on the experimental data by Knill et al. (1989). Both the PKP-SKR
and the PKP-RF coefficient sets are generated by using the PKP coal preprocessor (cf. Sec. 6.3) and based on the properties of the employed coal. Tables 8.11 and 8.12 summarise the pyrolysis model coefficients for the Knill SKR dataset, the PKP-SKR and the RF models, respectively. All simulations are based on the RF setup from Tab. 8.8 where the number of grid cells was limited to 1M for the present study.
Table 8.11: Optimised coefficients for the SKR based pyrolysis models Coefficient Knill PKP-SKR
YYi, DAF[−] 0.65 0.49 A[s−1] 2e4 2e8 TAc[K] 6103 13097
Table 8.12: Richards & Fletcher pyrolysis model parameters for Eqs. (6.15) to (6.17)
Parameter CaHbOcNd1 CaHbOcNd2 λi [−] 0.21 1.0 Ai [s−1] 448169 2e8
Ti,Ac[K] 8225 11082
σi1[-] -327 303
σi2[-] -2788 -2700
σi3[-] 26077 5431
σi4[-] -24860 14457
The volatile yield over residence time for the three extracted temperature pro-files, using the procedure described in chapter 6.4, is shown in Fig. 8.8. It can be seen that the simulation based on the Knill pyrolysis data has the highest volatile yield for all temperature profiles. Furthermore, the Knill dataset shows the ear-liest onset of pyrolysis of the three tested models for the low temperature profile.
At higher peak temperatures and heating rates, however, the PKP-optimised mod-els show a comparable pyrolysis onset. For the tested PKP-based modmod-els the final volatile yield is below the Knill dataset within the tested temperature range. Addi-tionally, the final volatile yield of the RF model is lowest at low temperatures and heating rates and approaches the final volatile yield of the Knill-SKR dataset at high temperatures. Furthermore, since the Knill-SKR dataset has both a lower pre-exponential factor and activation temperature compared to the PKP-based datasets, it can be seen that with increasing temperature profiles the order of the pyrolysis progress changes. Thus, for the lower temperature profile the Knill-SKR dataset