An Orthogonal Sparse Weight Matrix Algorithm for Bearing Early Fault Detection and Recognition

As a key rotating component of machinery and equipment, the bearing's condition monitoring and remaining useful life prediction are critical. If we can simultaneously monitor the multiple failure characteristics using one model, the fault detection efficiency and RUL prediction accuracy can be improved. In this paper, we propose a multi-feature health index that can track the degradation trends of different types of faults simultaneously. For data preprocessing, the fast Fourier transform and hyperbolic tangent function transform are used to convert the vibration signal into spectral features as input for constructing the health indicator. The health index, with the help of the hyperbolic tangent transform, can identify early faults earlier than those without the hyperbolic tangent transform. In constructing the health indicators, the orthogonal property can simultaneously track the three types of defects. The addition of sparse terms makes the weight matrix exhibit significant sparsity, which can help distinguish the frequency bands where different types of faults occur and help improve the generalization performance of the project matrix. The XJTU dataset is used to validate the effectiveness of the proposed method for tracking the degradation trend of different bearing fault.