Ultrasonic evaluation of wire-to-terminal joints: integrating XGBoost machine learning with finite element feature analysis.

A new scheme for ultrasonic non-destructive evaluation of wire-to-terminal joints was developed in this study. A finite element simulation model of 2D porous media based on X-CT images of an NG sample was proposed to analyse acoustic scattering. The simulation signal features were extracted in the time domain, frequency domain, and time-frequency domain, and entropy analysis was conducted to explore the relationship between different characteristic values and porosity, thereby revealing prominent trends in these features. By controlling parameters, 28 welding samples labelled OK and NG were produced, and their echoes were acquired by an ultrasonic Full Matrix Capture (FMC) system. Features of the full matrix signals were extracted, and the combined XGBoost machine learning was used to classify the quality and order the attribution of features. The result highlighted the significance of the waveform factor, margin factor, and kurtosis which are consistent with simulation results. The accuracy of weld quality identification can reach 84%. The three factors may be performance criteria for ultrasonically welded wire-to-terminal joints. [ABSTRACT FROM AUTHOR]