Numerical Investigation on Reactive Flows of Several Practical Applications
Prof. Andreas Kempf
Local Project ID:
HPC Platform used:
SuperMUC-NG at LRZ
This project collates individual applications from the Fluid Dynamics group at Duisburg-Essen University. The subprojects include the investigation of phenomena from the fields of nanoparticle synthesis, supersonic flows and stratified burners using LES, DNS and direct chemistry.
This first subproject targeted the assessment of cyclical variations in internal combustion engines using LES in a Lagrangian-Particle framework. HPC facilities for the simulations were required due to the resolution necessary to achieve good results in the complex geometry as well as the run time over several cycles.
The inhouse solver PsiPhi was employed to simulate 30 consecutive cycles of the optical research engine at the TU Darmstadt. A density based framework was used to solve the momentum and scalar transport equations using eight-order CDS supported by a tenth order filter for momentum and TVD for the scalar equations. Walls and moving parts were modeled using an efficient Lagrangian-Particle and immersed boundary based method.
Flow fields were compared to assess the validity of the proposed method and simulation. Experimental and computational flow fields agreed well. An instantaneous snapshot during intake can be seen in Figure 1.
Cyclical variations were assessed by seeding randomly generated Lagrangian tracers in the intake manifold and tracing them down to ignition. Temporal and spatial information of these particles is saved to files and thermodynamical trajectories of the different states experienced by the particles were reconstructed.
The proposed method was able to predict the underlying physics well
The aim of this subproject was the improvement of simulation tools for solid fuel combustion (biomass, coal, waste) utilizing Lagrangian-particles.
Here, LES of the Central Research Institute of Electric Power Industry (CRIEPI) lab pulverized coal flame have been conducted using a four-dimensional flamelet model based on two mixture fractions for volatiles and a hydrogen pilot, a normalized progress variable and the enthalpy . The effect of the suction probe on the scalar field measurements was tested by simulating this probing, observing relative changes up to 50% in various quantities and locations. By consideration of these probe effects, the agreement between the experiment and simulation could be improved significantly; at the same time, the simulation also provided the unperturbed scalar fields, without probing effects. The flamelet model gives a robust and cost-effective prediction of the investigated laboratory flame, provided that the probing effects are considered. The HPC facilities for the simulations were necessary due to the high resolution necessary to get good agreement with experimental results as well as the many test runs to investigate different model parameters and models.
Ongoing research in flamelet modeling is made in the context of co-firing of coal and ammonia to reduce pollutants by introducing a hydrogen carrier as a fuel. The introduction of a new fuel stream complicates the flamelet description and is part of the ongoing work.
This third sub-project aimed at improving the understanding of detonation cellular structures which in turn is not only important for safe operation of combustion devices that make use of detonations waves but also to prevent unwanted detonations in nuclear reactors or process plants.
The resources in this sub project were primarily used for three-dimensional simulations of detonation wave propagation in confined channels and tubes. In particular, the results were used to investigate the influence of the geometry on the propagation dynamics and the structure of the wave. In the past, detonation simulations have been mostly restricted to 2D computational domains, due to immense resolution requirements. While detonation kernels in 2D stem from the collision of two transverse shocks, two classes of detonation kernels exist in 3D, introduced as line kernel blasts and multi kernel blasts, which contain substantially more extreme states. Rectangular channels lead to a highly regular matrix of transverse waves, thus reflecting the confining geometry, while tubes lead to a more complex and chaotic detonation structure (see Figure 2). The results have been presented on ICDERS and are currently under review for publication .
A grid resolution of 10 micrometers was necessary to capture important length scales present in the detonation structure, resulting in a total number of 1.7 billion numerical cells. Since a low dissipation Riemann solver with high order reconstruction and, even more important, detailed reaction kinetics were used, i.e. the solution of a stiff system of ODE's was calculated at every point for each time step, the cost of these simulations is very high, making the use of a super computer essential.
The fourth subproject investigates the effect of subgrid stresses on the coagulation kernel with simulations of the SpraySyn burner.
The resources were used for three-dimensional LES simulations of the SpraySyn burner. The flame chemistry was resolved with a flamelet generated manifold approach, and each spray droplet was resolved with an individual lagrangian particle. A particular focus was on nanoparticle modeling. Here, a sectional model was used with and without a subgrid model to investigate the effect on the coagulation rate for the first time. Further, detailed measurements of the particle size distribution in the flame became available for the first time and allowed a detailed validation of the simulation results. The simulation required several validation simulations with the largest one having 200 million cells and 0.5 million time steps.
As a result, the experimental data could be reproduced sufficiently and confirm the simulations while no significant effect of the subgrid model was observed .
For future work on this project, the subgrid effects could be resolved via an FDF-method.
The fifth subproject investigated the formation of nano-particles in reactive flows using direct numerical simulations. These simulations resolved not only the smallest scales of the turbulent flow, but also the smallest relevant scales of the nanoparticle field - the Batchelor scales. The main objective here is to understand the physics of diffusion, coagulation, and nucleation and to use the DNS database for future modeling efforts.
All simulations were performed with the in-house code PsiPhi, using direct chemistry with 8 species and 3 reactions and a sectional model for nanoparticles (20 sections) on a computational domain with 500 x 500 x 100 numerical cells at a grid resolution of ∆ =40 μm (see Figure 3). Each simulation required 4M core-h on 25,000 CPU cores. The reasons for the high computational cost were a) the application of direct chemistry and b) the number of sections (and thus transport equations) required for the nanoparticle phase.
Figure 2 presents three-dimensional blue iso-surface Q1=1.0×1019 [#/m3] (representing monomers of titanium dioxide nanoparticles) with vortical structures colored by temperature.
The simulation of this subproject represents the state-of-the-art in the DNS of flames producing nanoparticles, and will provide meaningful, transferable and sustainable insight into particle formation, to guide the modelling efforts.
 D. Meller et al., Energy & Fuels 35 (9) (2021) 7133-7143.
 J. Crane et al., Proc. Combust. Inst. (2022), Under review
 J. Sellmann et al. Powder Techn. Under revision
 S.J. Baik et al. ICLASS (1) (2021)
Prof. Dr. Andreas Kempf
Carl-Benz Straße 199
Raum: NETZ 1.09a