he need for processing large amounts of data recorded by structural health monitoring (SHM) systems has been fostering interdisciplinary SHM strategies employing artificial intelligence (AI) algorithms for detecting damage. However, the opacity of several AI algorithms hinders their widespread adoption in SHM practice. To enhance the trust of practitioners in AI algorithms, this paper proposes an explainable artificial intelligence (XAI) approach for SHM. The approach builds upon the capabilities of unsupervised learning algorithms for detecting outliers indicative of structural damage in structural response data. Moreover, features in the data governing outlier detection are “explained” to the user, thus ensuring transparency in decision making. The XAI-SHM approach is validated via simulations of a pedestrian bridge that may or may not include damage. The results show that the XAI-SHM approach is capable of distinguishing between damage and random fluctuations of structural properties, while decisions made by the XAI-SHM approach are clearly explained.
