Broad AnlosaNet: an improved version of AnlosaNet using broad learning for bearing fault diagnosis with limited samples

Bearing fault diagnosis is vital for industrial machinery but faces the challenge of insufficient fault signal samples. To enhance the performance of AnlosaNet with limited samples, we proposed a novel Broad AnlosaNet neural network. This innovative framework integrates a broad learning layer for effective information fusion. Additionally, it incorporates a Transformer block to process reverse signals, enabling the extraction of deep features. The proposed network comprises five onedimensional convolutional neural networks, two Transformer blocks, a long short-term memory network, a gated recurrent unit, and a broad learning layer. The efficacy of the proposed model was validated using our proprietary dataset, achieving an impressive accuracy of 98.61%. In addition, the experiment utilized six other datasets to verify the generalization of the network. The results showed that it has effective generalization.