Acoustic emission for surface quality monitoring in laser shock peening via dual-feature fusion convolution neural network

Real-time monitoring of surface quality during laser shock peening (LSP) manufacturing processes is challenging due to the complex dynamics of laser-matter interactions. This study proposes a method based on acoustic emission (AE) and convolutional neural networks to assess the surface quality of the material after laser shock. Firstly, AE piezoelectric sensing is used to capture the stress wave signal inside the material in real-time during processing, using the surface quality of the LSP-treated 7075 aluminum alloy sample as the classification label. Then, the maximum overlapping discrete wavelet packet transform cepstrum (MODWPTC) is constructed without expert prior knowledge. The adaptive allocation of filter weights based on the frequency band energy of the AE signal allows for information mining at various scales. Finally, a brand-new dual-feature fusion convolutional neural network (DFCNN) is proposed to carry out multi-scale and multi-level deep feature fusion, allowing for real-time precise monitoring of material surface quality. Using the cepstrum time-frequency map as the feature input, the average identification accuracy of the five tests for surface hardness at different laser shock times and surface stress at different laser energies reached 99.20 % and 99.67 %. These experimental results show that the AE features based on cepstrum analysis can overcome the shortcomings of traditional feature extraction methods. Also, the DFCNN with different scales and levels feature fusion modules has the advantages of high recognition accuracy and stable training process.