Flow parameter estimation using laser absorption spectroscopy and approximate Bayesian computation

Abstract

Given spatially sparse or lower dimensional experimental measurements, approximate Bayesian computation (ABC) and numerical simulations can be used to estimate unknown characteristics of complex multi-physics engineering systems. Here we describe the ABC approach and use it to estimate the speed of high-temperature gases exiting an industrially-relevant catalytic burner, as well as to estimate the completeness of combustion within the burner. Using vertical profiles of absorption-weighted average temperature from laser absorption spectroscopy (LAS) at three different burner operating conditions, we combine ABC and large eddy simulations (LES) to generate posterior distributions of inflow speeds and heat addition characteristics above the burner. We show that the ABC method correctly estimates trends in the inflow speed for different conditions, and we find that there is a strong likelihood of incomplete combustion for higher equivalence ratios. We evaluate the predictive capability of the approach using an observing system experiment, indicating that the ABC method, when combined with LES, is able to accurately predict LAS measurements. We thus demonstrate that ABC is an effective tool for obtaining additional insights from available experimental measurements, thereby improving understanding of real-world engineering systems.

Olga Doronina

Postdoctoral Reseacher

My research interests include data-driven turbulence modeling, statistical analysis and machine learning