A novel method based on CNN-BiGRU and AM model for bearing fault diagnosis

This study presents an innovative fault diagnosis methodology that integrates a convolutional neural network (CNN) with a bidirectional gated recurrent unit (BiGRU) and an attention mechanism (AM). Distinct from traditional techniques that primarily concentrate on enhancing diagnostic efficacy, this method additionally evaluates the impact of employing or omitting fast Fourier transform (FFT) preprocessing, as well as the positioning of the AM in feature extraction, on the accuracy of fault diagnosis. The diagnostic models are constructed using CNN-AM-BiGRU, CNN-BiGRU, and CNN-BiGRU-AM configurations. Subsequently, the AM assigns weights to the spatial and temporal features, enriching their depth and comprehensiveness. The final stage involves a comparative analysis of the probability values generated by CNN-AM-BiGRU, CNN-BiGRU, and CNN-BiGRU-AM across a specific iteration. This comparison is conducted under two distinct scenarios: one involving FFT preprocessing and the other without it, alongside varying positions of the AM in the feature extraction process. Meanwhile, the computational costs of the proposed method are compared to the state-of-the-art methods. The results indicate that the proposed fault diagnosis method significantly decreases computational time and effectively discerns the trend of probability values across diverse iterations.