Numerical Investigation on Flashback Mechanisms in Premixed Hydrogen/Air Swirl Combustion

Principal Investigator:
Prof. Christian Hasse

TU Darmstadt

Local Project ID:

HPC Platform used:
SuperMUC-NG at LRZ

Date published:


As a carbon-free fuel, hydrogen (H2) has the potential of emerging as the leading energy carrier for next-generation, zero-carbon power generation, and hence has received considerable attention.  H2 can offer significant benefits over hydrocarbon fuels, such as wide flammability range, low ignition energy, and high diffusivity. However, the use of H2 in gas turbines poses considerable challenges, e.g., the risk of flashback due to its high flame speed, which adversely affects the performance of H2 combustion [1].

Flashback, a problem that occurs in premixed combustors, is the upstream propagation of the flame from the combustor into the premixing tube due to the change in mass flow rate, which could change the combustion process and pollutant emissions as well as cause considerable damage on the combustor [1]. Therefore, a deep understanding of the flashback mechanisms is required before applying pure H2 or H2 enriched fuels in combustion systems.

Flashback can occur because of four main mechanisms [1]: (1) boundary layer (BL) flashback in non-swirling or low swirling flow, (2) combustion induced vortex breakdown (CIVB) flashback in a swirling flow without a bluff-body, (3) combustion instability induced flashback, and (4) combined BL/CIVB flashback in a bluff-body swirl burner. The focus of this work is the latter one, because bluff-body swirling flows are typically employed to enhance mixing and flame stabilization.  The systematic investigation of BL flashback started with the work by Lewis and von Elbe [2], who proposed a classical critical gradient model to evaluate the flashback in channel flows. In the confined BL flashback configurations where the flame is already inside the premix duct before flashback, the formation of reversed flow pockets ahead of laminar/turbulent flames is one of the main characteristics prior to the onset of flashback, which is attributed to a pressure rise induced by the formation of flame bulges and has been reported both experimentally and numerically. On the other hand, flashback in a swirl combustor without a central bluff body has been studied extensively. It is concluded that the flashback is related to vortex breakdown (i.e., CIVB flashback). This is because the baroclinic torque can produce a negative azimuthal vorticity and then induce a negative axial velocity along the vortex axis, leading to the vortex breakdown and facilitating flashback. 

Despite its high relevance for various practical combustor designs, there are only a few studies on flashback in a bluff-body swirl burner. The Clemens group [3] experimentally identified two modes of flame propagation during flashback, i.e., (1) small-scale bulges propagating in the negative streamwise direction; and (2) large-scale flame tongues swirling with the bulk flow while leading flashback. More recently, a new flashback mode was discovered by Ebi et al. [4], and they found that the upstream propagation of flame can be led by flame bulges. Due to the inherent three-dimensional and transient nature of flame flashback, 3D parameter measurements and high-speed detecting devices are required, which leads to a significant challenge for experimental investigations of flashback.

It is noted that the above investigations of flashback in bluff-body swirl burners only provided 2D or local 3D flow field (without flame) measurements due to the limitation of laser-based techniques, which cannot comprehensively capture flow-flame interaction during the swirling flashback. The overarching objective of this project is to provide for the first time a detailed understanding of complex flow-flame interaction during flashback in a bluff-body swirl burner using Direct Numerical Simulation (DNS), which would help to reveal the mechanism(s) allowing the flame to propagate upstream during flashback and to improve practical swirl burner designs.

Results and Methods

Within this project all simulations and analyses are based on direct numerical simulations (DNS) which are conducted with the DNS code DINO [5]. The solver is designed for the simulation of low-Mach number reactive flows, where spatial derivatives in the governing equations are discretized with 6th order finite differences. The temporal integration is done by a 3rd order semi-implicit Runge-Kutta scheme. Based on the distributed memory architecture of SuperMUC parallelization of the solver is achieved by the message passing interface (MPI), where an excellent scalability up to 65,536 cores is achieved.

The DINO code is parallelized in two dimensions using the 2DECOMP&FFT library that acts on top of standard MPI and FFTW. The Poisson equation for pressure is solved by means of FFT for both periodic and non­periodic boundary conditions, but with dedicated pre­ and post­processing FFT techniques in the latter case. An implicit time integration of the stiff chemical source terms has been implemented, relying on a semi-implicit Runge­Kutta 3rd­order. In the DINO code, the chemical source terms are computed using the open­source Cantera library. The transport properties are computed either with the Cantera library or with the EGlib­ 3.4 library.

It is noted that a cylindrical DINO version has been developed with the help of the project to achieve DNS calculations of flashback efficiently.

DNSs of burner stabilized flame and flashback

In this project, two DNSs are performed: (1) DNS of a stable, lean premixed swirling hydrogen/air flame in the bluff-body swirl burner; and (2) DNS of lean premixed swirling hydrogen/air flame flashback in the bluff-body swirl burner. The two DNS results are shown in Fig. 1. Flashback is triggered by disturbing the stable burning flame through impulsively increasing the equivalence ratio (ϕ) uniformly at the inlet plane from ϕ=0.5 to ϕ=0.8.

The computational domains comprise a region of 291x72x72 or 350x88x88 flame thicknesses for the first DNS or second DNS, respectively. They require 3,200x900x900 and 3,500x1,000x1000 grid points for the first DNS and second DNS, respectively. Concerning the overall simulation time, the DNS is estimated to run for about 150 ms and 200 ms for stable flame and flashback DNS calculations, respectively. The stable burning case is found to have reached a statistically steady state after 150 ms and the flashback process lasts over a time span of 200 ms which was assessed qualitatively by a visual inspection of the flame front evolution during flashback.

It is seen from Fig. 1 that at ϕ=0.8 the flame can propagate upstream and hence flashback occurs. The flame front is swirling around the center body, which is indicated by the red arrow. It is noted that owing to a higher axial velocity near the outside wall, the flame mainly propagates near the bluff-body wall.

Flashback modes

Figure 2 illustrates the instantaneous flame fronts during flashback at different themral boundary conditions. Two flame structures can be observed along the flame front: (1) a large-scale flame tongue, and (2) multiple small-scale flame bulges, which are consistent with experimental observations [3]. Similar to [3], we refer to the rather large, leading part of the flame front as a flame tongue and refer to smaller structures along the flame front as flame bulges which tend to form on the trailing side of the flame tongue.

Figure 2 shows that the leading flame tongue tends to swirl upstream (green arrow), while the flame bulges tend to propagate upstream in the negative streamwise direction (yellow arrow). It is found that the flashback process can exhibit different modes when changing the thermal boundary conditions. Specifically, for Twall = 350 K case, the flashback is led by a large-scale flame tongue, and the flame tongue swirls in the direction of the flow upstream and propagates upstream in the mixing tube.

At the trailing side of flame tongue, there exist multiple small-scale bulges that are convex to the coming flow. The flame bulges cannot achieve a sustained flame propagation against the swirl flow direction and hence cannot lead to the net upstream flame propagation. In this context, the lowest tip of flame front always locates at the flame tongue rather than these bulges, indicating that flashback in this case is led by the large-scale swirling flame tongue, denoted as Mode I. However, it is seen that for the adiabatic case, the flashback is led by small-scale flame bulges, denoted as Mode II. When the temperature of the central bluff-body is fixed at 500 K, there also exist a flame tongue and multiple flame bulges, like Twall = 350 K case.

However, compared to Twall=350 K case, the flame tongue and flame bulges are separated from each other at Twall=500 K. More importantly, the lowest flame tip no longer locates at the flame tongue only. Instead, the location of the lowest flame tip switches between the flame tongue and the flame bulges, which indicates that flashback in this case is led by different structures, switching between Mode I (swirling flame tongue) to Mode II (non-swirling flame bulges).

On-going Research / Outlook

The preliminary results shown in this report imply that the thermal condition of bluff body has a significant on flashback characteristics. To understand the underlying mechanisms, we are investigating near-wall flame-flow-wall interactions.


[1] A. Kalantari, V. McDonell, Prog. Energy Combust. Sci. 61 (2017) 249–292.

[2] B. Lewis, G. von Elbe, J. Chem. Phys. 11 (1943) 75-97.

[3] D. Ebi, N.T. Clemens, Combust. Flame 168 (2016)39-52

[4] D. Ebi, R. Bombach, P. Jansohn, Proc. Combust.Inst. 38 (2021) 6345–6353.

[5] A. Abdelsamie, G. Fru, T. Oster, F. Dietzsch, G. Janiga, D. Thévenin, Comput. Fluids131 (2016), 123-141.


Scientific Contact

Dr. Wang Han
L1|01 281
Otto-Berndt-Str. 2
64287 Darmstadt
+49 6151 16-24142