Bayesian Experimental Design for Data Assimilation in Thermoacoustics

Bayesian data assimilation is a powerful tool for generating quantitatively-accurate physics-based models of thermoacoustic systems. Depending on the model, this method can require assimilation of thousands of experimental data points to achieve sufficient confidence. Conducting so many experiments on aircraft or rocket engines would be prohibitively expensive. To address this problem, we apply methods from information theory and Bayesian experimental design to identify the experimental configurations that provide the maximum information when assimilating into a given model. We demonstrate the power of this approach by applying an optimal experimental design strategy to select a subset of data points from a large experimental data set. The full data set contains 7000 experiments collected from 175 stable operating conditions of a hot wire Rijke tube. We show that by assimilating just 20% of the data points with the highest information content, we can achieve 90% of the confidence that is gained by assimilating the full data set.