In-situ inference of the thermoacoustic properties of an industrial combustion system

A Bayesian framework is proposed for building and calibrating physics-based models of industrial thermoacoustic systems, using the Rolls–Royce SCARLET test rig as a case study. Several candidate models are constructed, and their uncertain parameters are inferred directly from experimental data. Bayesian model comparison is then used to identify the most probable candidate model, balancing \chng{data fit} and model complexity. The selected physics-based model reproduces both non-reacting and reacting measurements with high \chng{precision}. This model is then used to confirm the findings of recent work, which demonstrates an inconsistency in a commonly used method for measuring the flame response in complex combustion chambers. This paper goes further to provide an improved method for identifying the flame response from data. Moreover, because this process learns the parameters of a flame model rather than just processing the experimental data, the model can interpolate and extrapolate, and provide deeper insight into the underlying physics.