Fault Diagnosis Method of Rolling Bearing Based on MMFF-3DCNN

To address the challenge of comprehensively characterizing bearing fault features using a single sensor in complex industrial environments, a bearing fault diagnosis method integrating adaptive-pooling-based weighted multi-modal feature fusion (MMFF) with a three-dimensional convolutional neural network (3DCNN) is proposed. Vibration and acoustic signals are employed to provide complementary fault information. Multi-scale and multi-modal two-dimensional (2D) features are extracted using multiscale short-time Fourier transform (STFT) and Mel-spectrum (Mels). These features are subsequently unified and stacked into 3D data through the MMFF strategy. The resulting 3D data are then fed into the 3DCNN for fault classification. Experimental results show that the proposed method achieves a best diagnostic accuracy of 99.25% on the bearing fault dataset and maintains robustness under high-noise conditions. The proposed method provides an effective solution for rolling bearing fault diagnosis under complex operating conditions and has significant value for engineering applications.