An intelligent fault diagnosis method for CRDM based on linkage fusion of time-frequency features

As a key device for controlling the reactor’s normal operation, the control rod drive mechanism (CRDM) will face multiple threats of failures during long-term operation. For such device with multi-source excitation and clearance coordination, its fault information is often submerged by impact interference signals. The key to solving this problem is studying a deep-level feature extraction and fusion method to overcome the interference. In this way, a fault diagnosis method for CRDM based on linkage fusion of time-frequency features is proposed in this paper. Firstly, the multi-scale time-domain features of the vibration signal of CRDM are extracted based on the time-shift multi-scale diversity entropy (TSMDE). Secondly, the frequency-domain vibration signal of CRDM is obtained by fast Fourier transform, and its energy features are extracted and compressed using convolution and pooling. Thirdly, a linkage fusion network module based on time-frequency features is proposed to perform preliminary feature screening and learning on multi-scale time-domain features as well as frequency-domain energy features, and then linkage fusion of the time-frequency features is carried out based on an independent weight distribution strategy. After that, an improved convolutional neural network (ICNN) is proposed to dynamically constrain the correct fusion of CRDM time domain and frequency domain features. Finally, using the experimental data of the drive mechanism to carry out fault diagnosis experimental verification of the method proposed in this paper, and compare and analyse the diagnostic results with a variety of current methods, proving the effectiveness and superiority of the method proposed in this paper.