Cavitation fault diagnosis of hydropower units based on multi-channel acoustic emission signal fusion

To address the difficulty in identifying cavitation faults in hydropower units due to insufficient signal diversity and much noise interference， this paper proposes a cavitation fault diagnosis method for hydropower units based on multi-channel acoustic emission signal fusion. Firstly， the multi-channel acoustic emission signals were collected from a cavitation test on the hydropower unit cavitation simulation test bench， and these signals were processed via data compression to form a cavitation fault dataset. Then the acoustic emission signals were transformed into Mel spectrograms， and frequency weighting was applied to remove noise from high-frequency signals and enhance features in low-frequency signals. Finally， a multi-channel deep convolutional neural network model based on decision-level fusion was constructed by combining the convolutional block-attention modules and the D-S evidence theory. This model was used for training and testing on the hydropower unit cavitation fault samples to obtain diagnosis results. The results show that this method can effectively distinguish cavitation faults under different operating conditions. Compared with other modeling methods， it exhibits higher diagnostic accuracy and better noise resistance， providing significant reference value for the practical application of cavitation fault diagnosis of hydropower units.