Fault Diagnosis of T0 Chopper Bearings under Strong Noise by Fusing Neural Blind Deconvolution and Sparse-Domain Kalman Filtering

The fault diagnosis of T0 chopper bearings presents a significant challenge due to strong background noise and complex operating conditions. To address this, this paper proposes a novel diagnostic method by fusing Neural Blind Deconvolution (NNBD) and Sparse-Domain Kalman Filtering. The framework first employs NNBD to enhance incipient fault impulses from signals heavily corrupted by strong noise. Subsequently, features are extracted via sparse representation and then temporally optimized using a Kalman Filter in the sparse domain (SR-KF) to improve robustness and effectively track dynamic changes. Finally, a Bidirectional Long Short-Term Memory (Bi-LSTM) network classifies the optimized feature sequences for a precise diagnosis. Validated on the T0 chopper bearing experimental dataset, the proposed method achieves $99.96 \%$ accuracy, significantly outperforming standard deep learning models and providing an effective solution for high-precision rotating machinery.