Abstract: To solve the problem of difficult recognition of rolling bearing failure modes due to the complex and variable operating conditions of power plant equipment,a rolling bearing fault acoustic pattern recognition method based on IPNCC-MCSKNet was proposed to realize the efficient recognition of rolling bearing faults under the operating conditions of variable rotational speed. Firstly,the acquired bearing acoustic signals were preprocessed,noise reduced,and feature difference integrated to form Improved Power-Normalized Cepstral Coefficients (IPNCC). Then the various voiceprint features including IPNCC were extracted to construct multi-channel input features,and the mechanism that the Selective Kernel (SK) convolution module can adaptively adjust the size of the convolution kernel was utilized to establish a multi-channel selective kernel network model (MCSKNet). Finally,voiceprint pattern modeling and fault recognition were carried out on different fault forms of rolling bearing samples. The experimental results showed that the proposed model achieved an average diagnostic accuracy of 95.99 % in the diagnostic task under multiple variable speed conditions,which was 13.98 % to 26.55 % higher than other deep learning models,and the model was more robust. The results can provide new ideas for rolling bearing acoustic feature extraction and fault diagnosis.

Key words: rolling bearings, voiceprint modeling, fault recognition, IPNCC, MCSKNet, selective kernel convolution