Dynamic data-driven stability map prediction in combustion systems

Prediction of thermoacoustic instabilities is a critical issue for both design and operation of combustion systems. Sustained high-amplitude pressure oscillations cause mechanical stresses in the structural components of the combustor, leading to thermomechanical damage. This paper proposes a dynamic data-driven method to construct stability maps of combustion systems as a function of pertinent process parameters. Given the knowledge of a combustion system's behavior at certain operating conditions, a Bayesian nonparametric method has been adopted to predict the system stability for operating conditions at which experiments have not been conducted. The proposed method also quantifies the uncertainty in prediction, resulting from measurement noise, insufficient training data, inaccurate parameter estimates etc. The proposed method has been validated in a laboratory environment with experimental data of pressure time-series from a lean-premixed swirl-stabilized combustor apparatus.