A Bayesian probabilistic approach for damage detection of a population of nominally identical structures: Application to railway wheel condition assessment

This paper proposes a Bayesian probabilistic approach to deal with the damage detection of a population of nominally identical structures. In this approach, a probabilistic reference model is first established with sparse Bayesian learning to describe structural dynamic characteristics of all nominally identical healthy structures using structural health monitoring data. Then, the conditions of the rest of structures can be identified through the examination of discrepancies between the new monitoring data and model predictions. To formulate the damage detection in a more scientific way, the discrepancies are examined by means of Bayesian hypothesis testing that allows to qualitatively and quantitatively evaluate structural conditions. To validate the feasibility and effectiveness of the proposed approach, its application to railway wheel condition assessment is presented with the use of online monitoring data collected by an optical fiber sensing track-side monitoring system.