Simulating Bluff-body Flameholders: On the Use of Proper Orthogonal Decomposition for Combustion Dynamics Validation

Contemporary tools for experimentation and computational modeling of unsteady reacting flow open new opportunities for engineering insight into dynamic phenomena. In the work presented here, a novel use of proper orthogonal decomposition (POD) is described to validate the structure of dominant heat release and flow features in the flame, shear-layer, and wake of a bluff-body-stabilized flame. A general validation process is presented which involves a comparison of experimental and computational results, beginning with single-point mean statistics and then extending to the dynamic modes of the data using POD to reduce the ensemble of instantaneous flow field snapshots. The results demonstrate the use of this technique by applying it to large eddy simulations of the bluff body stabilized premixed combustion experiment.

Large-eddy simulations (LES) using both Fluent and OpenFOAM were conducted to reproduce experiments conducted in an experimental test rig which was built as part of this work to study the behavior of turbulent premixed flames stabilized by bluff bodies. Planar Particle-Image Velocimetry (PIV) and filtered chemiluminescence were used to characterize the flow in the experiment's reacting and non-reacting regimes respectively. While PIV measurements could be compared directly to the velocity field in the simulations, the chemiluminescence measurements represented a line-of sight signal which was not directly comparable to the LES model. To account for this, the heat release in the LES models was integrated along simulated lines of sight by solving an additional discretized differential equation with heat release as the source term.

The results show generally good agreement between the dominant modes of the experiment with those of the numerical simulations. By isolating the dynamic modes from each other via the proper orthogonal decomposition, it was shown the models were able to accurately reproduce the size, shape, amplitude, and timescale of various dynamic modes which exist the experiment, some of which are dwarfed by the other flow features and are not apparent using time-averaging approaches or by inspection of instantaneous snapshots of the flow.