A Bayesian method for calibrating computer models to test data
Y.K. Lee, Dimitri N. Mavris, Vitali V. Volovoi, Ming Yuan · Inverse Problems in Science and Engineering · 2011
Model calibration is an inverse problem in which unknown model parameters are to be inferred from test data. Due to the ill-posed characteristics of inverse problems, model calibration is a challenging problem. This article presents an inverse method based on the statistical inverse theory. In this method multiple formulations for calibration are used simultaneously, and their results are averaged according to how likely each formulation is. The inverse method is applied to calibration of a computer model of a turbojet engine. It is demonstrated that the method accurately estimates the model parameters with limited amount of noisy data. It is also demonstrated that the method is capable of identifying possible alternative solutions and that using multiple competing formulations gives rise to more accurate and conclusive results than using single general formulation.