Evaluation of micro-crack states in aluminum plates based on Lamb waves with CNN and multi-head attention mechanisms

This paper investigates the state assessment of aluminum plate micro-cracks by combining nonlinear Lamb waves, CNNs, and multi-head attention mechanisms. The finite element software Abaqus is used to simulate micro-crack damage and analyze the nonlinear effects generated by the interaction of Lamb waves with micro-cracks which have different lengths, directions, and positions. In the signal analysis section, the stacking spectrum is introduced as a means of analyzing the nonlinear characteristics of micro-cracks to study the influence of these three factors on the frequency spectrum of the sensor array receiving signals. In the neural network section, to further reveal the complex relationship between the stacking spectrum and the length, direction, and position of micro-cracks, a network model combining CNN and multi-head attention mechanism is constructed. After training, the network model is tested on the test set, and the mean absolute error (MAE) values for length, direction, and position are 0.11, 2.49, and 0.31, respectively. This means that the trained network model can recognize the complex relationship between the stacking spectrum and the three factors of micro-cracks, thus realizing the effective prediction of the length, direction, position, and implementation of the detection and assessment of aluminum plate micro-cracks.