Rotating machinery is a kind of significant equipment that widely used in nuclear power plants (NPPs). The harsh operating environment and long-term continuous operation of the rotating machinery can cause various faults due to wear, vibration, et al, that threatens the safety of the NPPs. Intelligent fault diagnosis techniques can timely discover the abnormality of the rotating machinery, that received extensively attention in recent years. A fault diagnosis method for NPPs rotating machinery based on MoCo siamese neural network is proposed to address the issues of high noise, small sample, and low accuracy in fault diagnosis under actual operating conditions. The wavelet transform is used to denoise the sensor signals of rotating machinery and extract time-frequency features. The training samples are encoded by the siamese neural network method. The momentum contrast (MoCo) method is used to update the encoder of the siamese neural network. The cosine similarity is used to measure the similarity of sample coding features. The dataset of rotating machinery from Machinery Failure Prevention Technology (MFPT) is adopted to validate the effectiveness and accuracy of the MoCo siamese neural network method. The results shows that the proposed fault diagnosis method has strong noise resistance capability and can accurately diagnose rotating machinery in small sample conditions, demonstrating the potential application value in the fault diagnosis of NPPs rotating machinery.

Fault Diagnosis;Rotating Machinery;Wavelet Transform;Momentum Contrast;Siamese Neural Network