Weak bond detection in single-lap shear bonds by evaluating vibroacoustic modulations with artificial neural networks

Abstract

Adhesive bonding is an essential method for joining composite materials. However, the occurrence of contaminations, resulting in a not detectable weakened adhesion, persists. This study aims to uncover weak bonds with the vibroacoustic modulation method, a nonlinear ultrasonic testing method, where ultrasonic guided waves are modulated by a simultaneously applied, high amplitude pump wave. Afterwards, the measurements are evaluated by a deep learning approach. A previous dataset of 40 single-lap shear specimens (ASTM D5868-01), in which artificial interfaces in the form of circular PTFE films or release agent contaminations were introduced, was extended by a second dataset with 14 specimens of a different laminate to evaluate the robustness and transferability of the method. The proposed neural network approach can reliably recognize the bonding flaws in the training dataset and even has high accuracies on the transfer dataset, demonstrating the tremendous potential for the nondestructive evaluation of adhesive joints.