Bayesian inference of airfoil icing condition from simulated ice shapes

Abstract

Motivated by the difficulty of accurately determining the inflow parameters in icing wind tunnels and flight tests, the Markov Chain Monte Carlo (MCMC) method, a commonly used Bayesian inference method, is explored to solve the inverse problem with the help of an icing software. The icing software, SJTUICE, is used to produce ice shapes for inversion and serves as the prediction tool in the iterations of the inversion problem. The influence of prior estimation of different icing parameters on the convergence and accuracy of the inversion problem is discussed. The feasibility of the MCMC method in inferring the inflow condition in terms of rime ice and glaze ice is assessed. Generally, fast convergence and good accuracy in terms of single inflow parameter inversion can be easily achieved. However, the number of iterations required increases rapidly with the number of inflow parameters in the MCMC method.