Anomaly Detection Applied to ISHM for Thickness Reduction Analysis in Controlled Environments

Abstract

In this work, three machine learning approaches were evaluated for detecting anomalies in impedance-based structural health monitoring (ISHM – Impedance-based Structural Health Monitoring) of a specimen in a controlled environment. Supervised, unsupervised, and semi-supervised algorithms were chosen to compare them regarding detecting anomalies in an aluminum beam with failure induced by surface machining on one of the faces. After applying the algorithms, it was found that, of the three types of learning, supervised and semi-supervised were the ones that achieved the best accuracy in detecting anomalies. On the other hand, the unsupervised type model did not obtain good results for the conditions investigated. Thus, this can be an important technique comparison achievement for implementing real anomaly detection ISHM systems.