Method for hybrid materials diagnosis based on ultrasonic testing signal analysis through Dynamic Time Warping and machine learning combination

Ultrasonic testing is one of the most commonly used non-destructive testing (NDT) techniques due to its low cost and wide applicability. Automation and artificial intelligence (AI) are utilised to enhance performance and efficiency, often in high-tech solutions or specifically for monolithic materials. Consequently, a new method for testing fibre-metal laminates (FML) using A-scans is proposed. The approach employs AI-supported signal analysis to compare measurements with those from undamaged areas. The XGBoost library was used to develop the model, and Dynamic Time Warping (DTW) was employed to assess signal similarity, including shape-based analysis (DTW_z). The method was tested on undamaged samples, increased gain scenarios, delamination, and bottom recess. Chosen threshold values were not exceeded in healthy cases. In the increased gain scenario, despite DTW exceeding the threshold fourfold, the signal shape confirmed structural integrity. For delamination and holes, DTW thresholds were exceeded by up to 21%, while for DTW_z, were exceeded by 3% and 7%, respectively. Additional distance matrices can also visualise the changes reflected in the shape of optimal alignment paths. When focusing on the most variable signal regions, DTW reached 140% of the threshold value, while DTW_z attained 136% and 175% of their thresholds for delamination and cutouts. Furthermore, applying constraints improved detection accuracy and reduced processing time, increasing average DTW values from 28% to 36% for delamination and from 60% to 88% for recess, while the average DTW_z increased from 19.3% to 20.8% and from 26.1% to 29.6%, respectively.