A bayesian probabilistic approach for structural damage detection

Measured structural dynamic response is very helpful for structural health monitoring (SHM). However, the response signals that contain damage information of a structure is difficult to obtain. In other words, traditional classification methods are difficult to be applied to determine structural damage effectively. In this study, a Bayesian-based method that uses only, at model training stage, the information on structural response under healthy conditions is proposed for damage detection, followed by verification of the proposed method by referring to a simulated structure. A frequencydomain health condition index (HCI) is first formulated via a linear transformation. By applying sparse Bayesian learning (SBL) and relevance vector machine (RVM), regression models about the real and imaginary parts of HCI are then established. A quantitative analysis of residuals between the predicted HCI and actual HCI is used as a measure for damage identification. If the predicted HCI deviates considerably from the actual HCI, the damage is identified. By evaluating the simulated structure under different damage conditions, the effectiveness of the proposed method for structural damage detection is verified.