Unsupervised Machine Learning Methods for Detecting Anomalies in the Structural Behavior of a University Laboratory

Structural Health Monitoring (SHM) focuses on evaluating structural risk, which requires monitoring the condition, identifying damage or abnormalities, and estimating the structure’s safety. Armonicanalysis is commonly used for anomaly detection considering it provides information about the dynamical properties of the structure, by analyzing variations in the natural frequencies. This article addresses the challenge of identifying the influence of human activities and working machinery on the dynamic behavior of the Applied Mechanics Laboratory at the Department of Industrial Engineering of the University of Salerno using unsupervised machine-learning methods: One-Class Support Vector Machine (OCSVM), Isolation Forest (IF), Long Short-Term Memory- Autoencoders (LSTM-AE). The selection of the most relevant features played an important role in the performance of the machine learning models. The OCSVM and IF methods produced similar results in identifying anomalies in the dataset with unknown behavior. The results of this study provide a promising approach for detecting anomalies by analyzing variations in the behaviour of the structure, specifically by analyzing variation in the harmonic response due to external factors, where the behavior of these anomalies is unknown in the collected data.