Sensor fault detection, localization, and reconstruction for online structural identification

In this study, a novel sensor fault detection, localization, and reconstruction approach is proposed for online structural identification. The proposed method avoids the requirement of massive training data from the normal operating sensor network and presents a computationally efficient approach to diagnose and estimate the typical sensor faults in a dense sensor network for time-varying structural systems. First, a two-level Bayesian model class selection strategy is introduced for sensor fault detection and localization. By evaluating the plausibilities of the model classes in the two-level strategy, detection and localization of possible faulty sensors can be realized with low computational cost. After detecting and locating the faulty sensors, an online updating algorithm based on a Kalman filter and an extended Kalman filter is then utilized to simultaneously estimate the sensor faults and identify the structural system. Two illustrative examples are presented to validate the efficacy of the proposed method. The results show that the proposed approach offers a reliable and efficient sensor validation methodology for online structural identification.