Acoustic emission waveform filtering method based on autoencoder and BP neural network

To address the issue of acoustic emission (AE) signal superposition, a waveform filtering method based on an autoencoder and backpropagation (BP) neural network is proposed to effectively separate AE signals from noise. The performance of the model in classifying AE and noise signals was evaluated through pencil lead break experiments and numerical simulations. The results show that the autoencoder-BP neural network model achieved excellent classification performance, with a recognition rate of 96% for AE signals and 98% for noise signals. After filtering using the proposed model, the processed data significantly improved the localization accuracy of AE sources. This study provides an effective AE signal processing method for structural health monitoring systems and holds important significance for enhancing the accuracy of safety monitoring in concrete structures.