Cross-speed spindle motor bearings fault diagnosis combined with multi-space variable scale adaptive filter and feedforward hybrid strategy

Abstract

The vibration signal of the spindle motor contains complicated mixed modulation harmonics and background noise when computer numerical control (CNC) machine tools perform machining tasks. Additionally, frequent changes in the running speed of the spindle motor cause significant variations in the signal feature distribution, making fault diagnosis challenging. The adaptive sinusoidal fusion convolutional neural networks (ASFCNN) is proposed to achieve cross-speed spindle motor bearings fault diagnosis. The ASFCNN extracts multi-spatial and variable-scale fault features through the multi-spatial variable-scale adaptive sinusoidal filter (MVASF) for noise reduction. And a multi-level feedforward hybrid strategy (MFHS) is designed to fuse multi-layer features of the convolutional neural network (CNN) and time sequence information for fault feature enhancement. The proposed method is evaluated on a multi-source spindle motor dataset under real working conditions. Experimental results show that the ASFCNN model significantly outperforms the compared classical models in terms of diagnosis accuracy, the effectiveness and interpretability are validated through the visualization methods.