Convolutional Neural Network Design Based on Weak Magnetic Signals and Its Application in Aircraft Bearing Fault Diagnosis

Convolutional neural networks (CNNs) have been widely used in bearing fault diagnosis and have achieved promising results. However, due to interference from cage rotation frequency, diagnostic outcomes based on weak magnetic signals and traditional CNNs are often affected. To address this issue, this article proposes a method based on uniform phase intrinsic time-scale decomposition (UPITD). By analyzing the correlation between weak magnetic signals and cage rotation frequency in the time domain, the method effectively separates fault signals from cage rotational signals. In addition, the input size and convolution kernel size in traditional CNNs are typically designed based on empirical values, which may not be optimal. Therefore, this article determines the CNN input size based on the physical characteristics of weak magnetic signals and optimizes the convolution kernel size using the decay rate of exponential function envelopes, further improving diagnostic accuracy. Experimental results show that the proposed method, based on UPITD and CNN, achieves a fault detection accuracy of greater than or equal to 99.34% across various bearing fault types, with a standard deviation of 0.002, which is significantly superior to traditional vibration-based methods. These results demonstrate the superiority and reliability of the proposed approach in bearing fault diagnosis.