Improved Feature Extraction of Guided Wave Signals for Defect Detection in Welded Thermoplastic Composite Joints

Ultrasonic guided waves have been widely used to investigate defects in mechanical, naval, civil, and aerospace structures. In this context, machine learning has shown to be a powerful diagnostic tool since interpreting guided wave signals towards modeling damage indexes is not straightforward.This paper investigates different feature extraction methods and machine learning modeling paradigms for defect detection in ultrasonically welded thermoplastic composite joints. The main contribution is the development of a purely data-driven approach, based on supervised machine learning, capable of detecting weld defects successfully. We show that feature engineering enhances the effectiveness of the model. The approach combining a support vector machine with autoregressive features resulted in the best overall performance, with the accuracy being improved by 91.50\% when compared to feature extraction methods found in the literature