Damaged Structural Joint Integrity Prediction Using Dynamic Deflection Data and Convolution Neural Network Technique

The research predicts the dynamic responses of adhesively bonded single-lap joints (SLJs) using a numerical finite element analysis (FEA) model. The dynamic responses of adhesively bonded SLJs are numerically computed to assess the influence of damage (delamination and/or a crack) on structural integrity. The dynamic responses of various SLJ components are predicted numerically and compared with published data and in-house experimental values, including the mesh independence test. The experimental findings deviate from the numerical results, with differences ranging from a minimum of 4.47% to a maximum of 22.4%. Additionally, the bonding characteristics are checked through field emission scanning electron microscope images to understand the inherent morphological interaction of the polymer adherend. The optical microscope is utilized to check the uniformity of adhesive thickness throughout the bonded region. A convolution neural network architecture is developed to assess the integrity of damaged SLJ structural components using their dynamic deflection responses. In this regard, the dynamic response dataset has been generated numerically through a novel technique with the help of a customized Python script in association with the commercial finite element tool (ABAQUS). The proposed convolution neural network achieves 99.28% prediction accuracy in classifying damage. Finally, the numerical model is extended to analyze the influential parameters, including damage (delamination/crack) and their positions and orientations on the dynamic responses of adhesively bonded SLJ structural components.