Colloquium: Mihnea Floris (FPT)

This thesis investigates the detection and suppression of autoignition flashback events in a reheat hydrogen combustor model inspired by Anslado Energia's GT36. The dynamics of this system are investigated through the means of LES simulations, which indicate a repeating pattern where the autoignition takes places in the mixing duct, aided by the propagation of pressure waves and the subsequent heat release. Upon acquiring the data from this simulation, the number of variables is pruned through the means of a  dimensionality reduction technique called co-kurtosis PCA. The emergent features are then used in a modularity-based clustering algorithm which manages to successfully predict the flashback events. In addition, this algorithm is put through several robustness tests, which validate its potential for online prediction. Lastly, in new LES simulations, water is injected based on the prediction time and a preliminary design for the spray is found, which manages to successfully suppress the flashback.

Supervisor: Dr. A. K. Doan, Dr. I. Langella