Bayesian Calibration and Bias Correction for Forward Model-driven SHM

Abstract

Forward model-driven structural health monitoring (SHM) is an alternative approach to the two main categories of SHM research: model-driven and data-driven processes. It is a methodology whereby a validated numerical model is used, in a forward manner, in order to produce simulated damage state data for machine learning applications. This paper explores the use of Bayesian calibration and bias correction (BCBC) in order to simulate representative predictions of observational data. The technique allows calibration of a numerical model to be performed in a Bayesian scheme whilst accounting for bias that may occur due to simplifications of the underlying physics in the model. This paper demonstrates the application of BCBC, in a forward model-driven framework, for producing representative damage features for different damage extents and shows a comparison with both experimental and non-bias corrected damage features