Evaluation AC2-11

From KBwiki
Jump to navigation Jump to search

Front Page


Test Data

CFD Simulations


Best Practice Advice

Delft-Jet-in-Hot-Coflow (DJHC) burner

Application Challenge AC2-11   © copyright ERCOFTAC 2021


CFD results and Comparison with Experiments

This section presents results for simulations described in the CFD Simulations section, for the DJHC-I-Re4K5 case, i.e. fuel-jet Re = 4500. The results have been published in (Perpignan et al., 2018) but the analysis presented here is more extensive.

Effect of RANS turbulence model

The choice of turbulence model has an impact on the prediction of the spreading of the jet and the mixing rate of fuel, coflow and air. The standard k-ε model is known to overpredict the spreading of round jets. In the case studied here the EDC model in combination with the standard k-ε model tends to overpredict the ignition and leads to a clearly visible peak in the radial temperature profile (measured using CARS) already at x = 30 mm, which becomes larger at x = 60 mm (Figure 11). Choosing the Reynolds Stress Model (RSM) instead of the k-ε model does not eliminate the early temperature peak but predicts its position closer to the centreline, in agreement with the prediction of a smaller spreading rate of the round jet.

AC2-11 fig11a.png AC2-11 fig11b.png
AC2-11 fig11c.png AC2-11 fig11d.png
Figure 11: Comparison of results for the DJHC-I flame using RANS and the standard EDC model (Bao, 2017). Radial profiles at z=30, 60, 90 and 120 mm.

Effect of EDC approach

Proceeding with RSM the question can be addressed how to better predict the mean temperature. A key problem in the prediction of mean radial temperature profiles is the identification of the height at which the ignition is sufficiently strong to significantly affect the mean temperature. In the experiments, the presence of ignition kernels (detected via their chemiluminescence) affected only the high temperature tail of the temperature PDF (measured using CARS) and did not have significant effect on the mean profile. The more elaborate EDC model with local parameters (EDC-LP) does lead to a prediction without a too early peak (Figure 12).

AC2-11 fig12a.png AC2-11 fig12b.png
AC2-11 fig12c.png AC2-11 fig12d.png
Figure 12: Comparison of results for the DJHC-I flame using the standard EDC model and the EDC model with local model constant calculation (Bao, 2017). Radial profiles at z=30, 60, 90 and 120 mm.

Difference between RANS and LES

Although the flow configuration (round jet) allows RANS models to give good predictions, it is of interest to see to which extend LES provides more accurate results. The available results that allow a comparison between RANS and LES are for the CSE (Labahn et al., 2015; Labahn & Devaud, 2016) and the DA-FGM (Huang et al., 2017; Huang, 2018). The temperature peaks seen in the standard EDC model are present in the RANS simulations for both CSE and DA-FGM, although to a lower extent, as shown in the figures below (Figures 13 and 14). These peaks are not present in the LES simulations, which are, therefore, closer to the experimental behaviour. This can be attributed to the better representation of the fluctuations of the mixture fraction. The assumed PDF in the RANS simulations represents all fluctuations, the LES resolves a large part of the fluctuations and the assumed PDF in the LES represents only subgrid scale fluctuations.

However, the agreement between model and experiment closer to the centreline for the axial positions of 60 and 90 mm profiles is better for the RANS simulations than for the LES. This could possibly be resulting from differences in the impact of the inflow boundary conditions. The procedure to define the inlet turbulence is simpler and more robust in RANS. In LES the procedure to define the turbulence at the inlet is more involved and may require further attention (Huang, 2018).

AC2-11 fig13a.png AC2-11 fig13b.png
AC2-11 fig13c.png AC2-11 fig13d.png
Figure 13: Comparison of temperature results for the DJHC-I flame using CSE along with RANS (Labahn et al., 2015) and LES (Labahn & Devaud, 2016) and the DA-FGM model with RANS and LES (Huang, 2018). Radial profiles at 15, 30, 60 and 90 mm.

Prediction of velocity statistics

For all considered modeling approaches the agreement between LDA measurements and predictions of the radial profile of mean axial velocity is generally good. The differences are possibly related to the well-known problem of predicting the spreading rate of a round jet in RANS and the influence of temperature prediction on the density. On the other hand, the agreement between model and experiment in the profile of radial velocity is not good close to the burner (Figure 14). Eventually this could be attributed to unequal density of seeding particles in the fuel jet compared to the coflow, potentially leading to an underestimation of the radial spreading close to the nozzle. However, this suggestion is not supported by the experimental paper of Oldenhof et al. (2011) where it is explained that careful attention was paid to the equal seeding density of fuel and oxidiser.

Prediction of temperature

The predicted temperature results from a combination of boundary conditions and models for turbulence and turbulence-chemistry interaction.

In the outer region the description of surrounding air plays an important role. Temperature plots shown in Figures 12 and 13 show that laboratory air mixes with the coflow leading to reduction of the mean temperature. The rate of air entrainment depends on the adopted boundary conditions in the air region. The approach of giving the air an initial axial velocity of 0.5 m/s used in combination with the DAFGM and CSE approaches performs best.

In the part of the coflow at radial position larger than the mixing layer with fuel and smaller than the mixing layer with air an unexpected trend is observed between the experimental profiles at z = 30 mm and z = 60 mm. The experimental profiles show an increase in mean temperature level between x = 15 mm and 30 mm. The hypothesis can be made that this is due to a heat release effect, either directly (progress of reaction in the coflow stream) or indirectly (progress of reaction after a progress of radial mixing of enthalpy and oxygen). In absence of mixing the model assumption that the flow is in thermochemical equilibrium prevents prediction of a temperature increase. As a result, the agreement between model predictions and experiments for mean temperature in the core of the coflow is not as good at x = 60 mm as it is at x = 30 mm.

Including unmixedness (mixture fraction and temperature fluctuations) at the coflow inlet or alternatively extending the computational domain to inside the coflow annulus might be effective to reach better understanding and better agreement between modeling and experiment.

AC2-11 fig14a.png
AC2-11 fig14b.png AC2-11 fig14c.png
AC2-11 fig14d.png AC2-11 fig14e.png
AC2-11 fig14f.png AC2-11 fig14g.png
AC2-11 fig14h.png AC2-11 fig14i.png
AC2-11 fig14j.png AC2-11 fig14k.png
Figure 14: Comparison of results of axial and radial velocities for the DJHC-I flame using CSE along with RANS (Labahn et al., 2015) and LES (Labahn & Devaud, 2016), DA-FGM model with RANS and LES (Huang, 2018), and different EDC approaches (Bao, 2017). Radial profiles at 15, 30, 60, 90 and 120 mm.

LES simulations directly provide values of (resolved) temperature rms. The experimental radial profile of temperature standard deviation at x = 15 mm (Figure 15) is flat and more than 100 K, similar to what is present in the measured profile at x = 3 mm. This is because none of the simulations presented here has taken this initial level of temperature fluctuations into account or is able to predict the initial development of temperature rms correctly (See Sarras et al., 2014 for a simulation including the temperature rms profile as a boundary condition at the inlet in RANS-trasnported PDF simulations). Nevertheless, the increase in temperature rms due to the presence of the mixing layers between fuel and coflow and between coflow and air is predicted in agreement with the experiments by both CSE and DA-FGM.

AC2-11 fig15a.png AC2-11 fig15b.png
AC2-11 fig15c.png AC2-11 fig15d.png
Figure 15: Comparison of results of temperature standard deviation for the DJHC-I flame using CSE along with LES (Labahn & Devaud, 2016) and the DA-FGM model LES (Huang, 2018). Radial profiles at 30, 60 90 and 120 mm.


  • Bao H. Development and validation of a new Eddy Dissipation Concept (EDC) model for MILD combustion, MSc Thesis, Delft University of Technology, 2017.
  • Huang X, Tummers MJ, Roekaerts DJEM. Experimental and numerical study of MILD combustion in a lab-scale furnace. Energy Procedia 2017;120:395-402.
  • Huang X. PhD Thesis, Delft University of Technology, 2018
  • Labahn JW, Devaud CB. Large Eddy Simulations (LES) including Conditional Source-term Estimation (CSE) applied to two Delft-Jet-in-Hot-Coflow (DJHC) flames. Combustion and Flame 2016;164:68-84.
  • Labahn JW, Dovizio D, Devaud CB. Numerical simulation of the Delft-Jet-in-Hot-Coflow (DJHC) flame using conditional source-term estimation. Proceedings of the Combustion Institute 2015;35:3547-3555.
  • Magnussen B, 1981, January. On the structure of turbulence and a generalized eddy dissipation concept for chemical reaction in turbulent flow. In 19th Aerospace Sciences Meeting (p. 42).
  • Oldenhof E, Tummers MJ, van Veen EH, Roekaerts DJEM. Role of entrainment in the stabilisation of jet-in-hot-coflow flames. Combustion and Flame 2011;158:1553-1563.
  • Perpignan A.A.V., Gangoli Rao A., Roekaerts, D.J.E.M., Flameless Combustion and its Potential Towards Gas Turbines, Progress in Energy and Combustion Science 2018; 69: 28-62
  • Sarras G, Mahmoudi Y, Arteaga Mendez LD, van Veen EH, Tummers MJ, Roekaerts, DJEM. Modeling of turbulent natural gas and biogas flames of the Delft Jet-in-Hot-Coflow burner: effects of coflow temperature, fuel temperature and fuel composition on the flame lift-off height. Flow, Turbulence and Combustion 2014;93:607-635.
  • van Oijen JA, de Goey PH. Modelling of premixed laminar flames using flamelet-generated manifolds. Combustion Science and Technology 2000;161:113-137.

Contributed by: André Perpignan, Dirk Roekaerts, E. Oldenhof, E.H. van Veen, M.J. Tummers, Hesheng Bao, Xu Huang — TU Delft

Contributed by: Jeffrey W. Labahn — Stanford University

Front Page


Test Data

CFD Simulations


Best Practice Advice

© copyright ERCOFTAC 2018