A Score-Guided Regularization Strategy-Based Unsupervised Structural Damage Detection Method

It is critical to use scientific methods track the performance degradation of in-service buildings over time and avoid accidents. In recent years, both supervised unsupervised learning have yielded positive results in structural health monitoring (SHM). Supervised approaches require data from entire structure various damage scenarios for training. However, it impractical obtain adequate training situations service facilities. addition, most known only take response structure. these situations, contaminated containing undamaged damaged samples, typical real-world applications, prevent models fitting data, resulting loss. This work provides an technique detecting reasons stated above. approach trains on with anomaly score serving as model’s output. First, we devised a score-guided regularization detection expand difference between data. Then, multi-task incorporated make parameter adjustment easier. The experimental phase II SHM benchmark Qatar University grandstand simulator are used validate this strategy. suggested algorithm has excellent mean AUC 0.708 0.998 two datasets compared classical algorithm.