Identification Method for Cage Rubbing Faults of Flywheel Bearings Based on Characteristic Frequency Ratio and Convolutional Neural Network

Bearings are the core components of space inertia actuators such as flywheels. Their running status is directly related to the performance of the machine. Currently, cage rubbing faults are commonly seen during the operation of flywheel bearings and their vibration signals are particularly complex, making it difficult to identify the rubbing status. In response to this issue, this article proposes a process of identification method for cage rubbing faults by characteristic frequency ratio (CFR) and one-dimensional convolutional neural network (1DCNN). This method can fully utilize the expert experience and classification advantages of the neural network model to accurately identify the operating status of flywheel bearings. Firstly, the raw data is obtained through bearing vibration experiments. Then, these data are processed for preliminary state identification based on demodulated resonance technique and CFR to separate the abnormal bearings and the normal bearings. On this basis, the 1DCNN model is constructed to automatically extract fault features and classify states, realizing fault localization. The results indicate that this method is feasible and effective in identifying the rubbing faults of bearing cages.