Damage detection of aircraft composites using Laplace wavelet neural network

This paper presents a Laplace Wavelet Neural Network (LWNN) for aircraft composite material damage detection, addressing issues with generalization and interpretability in current neural network methods. The LWNN incorporates the one-sided attenuation characteristics of the Laplace wavelet function into the first layer of a convolutional neural network (CNN), solving problems like gradient vanishing and difficulty in fitting waveform data. The wavelet’s scale and translation parameters are treated as learnable, allowing the network to create a multi-scale wavelet dictionary for tapping signals and extract meaningful wavelet features. These features are then processed by the CNN, with an attention mechanism used for classification. Experimental results show that the LWNN model performs best with fewer parameters, proving its practical effectiveness in damage detection.