Feature Extraction Based on Signal Similarity for Damage Detection with Ultrasonic Guided Waves

Structural health monitoring is crucial for safeguarding critical infrastructure and requires the use of traceable methods. Hence, using explainable machine learning (ML) becomes indispensable for damage detection in such monitored infrastructures. This study presents an explainable ML model based on signal similarity between baseline and measured signals. The goal is to localize damage in composite overwrapped pressure vessels for hydrogen storage. The analyzed signals are ultrasonic guided waves, and their propagation is assessed using the pitch-catch procedure across a sensor network of 15 lightweight transducers. After benchmarking over 2,540 ML algorithms, the best model achieved an average localization deviation of 11.15 mm.