Data-driven crack behavior anomaly identification method for concrete dams in long-term service using offline and online change point detection

Cracking is a common threat to dam structural safety. It is desirable to establish models that can accurately assess the influence of cracks on the structural safety of concrete dams in time. The structural condition assessment for dams can be categorized into offline structural state review based on historical monitoring data and online real-time detection based on updated data. Moreover, the offline review can be further divided into two scenarios, depending on whether the number of change points is known in advance. To solve the above practical problems, three different offline and online changepoint detection (CPD) methods, including dynamic programming segmentation, bottom-up segmentation, and online Bayesian CPD methods are introduced. A concrete gravity-arch dam with 300 m length and 5 m depth horizontal cracks stretched across the downstream of various blocks in long-term service is used as the case study. Crack opening displacement collected by resistance joint meters is used to demonstrate the feasibility of the proposed identification methods. The experimental results show that the underlying change points of crack behavior can be accurately and timely detected by the proposed model, and the exact dates of change points can also be obtained. The calculated results are roughly consistent with the observations of visual inspections and are consistent with the recorded historical engineering management report. The proposed model does not require prior physical knowledge about concrete cracks. It is practical and flexible to be embedded in dam automated structural health monitoring systems to deal with large-scale monitoring data related to structural changes.