Fault Diagnosis Method for Rotating Machinery Based on FFT-CNN-Transformer-CrossAttention

Aiming at the challenges such as difficult feature capture and insufficient feature extraction faced in the fault diagnosis of rotating machinery, a fault diagnosis method based on FFT-CNN-Transformer-CrossAttention that integrates time–frequency domain features and the cross-attention mechanism is proposed. This method utilizes the Fast Fourier Transform (FFT) to convert one-dimensional temporal fault signals into frequency-domain features. The time–frequency domain signals are processed, respectively, through the CNN to extract local impact patterns and frequency energy distributions, and combined with the multi-head self-attention mechanism of the Transformer encoder, the long-term dependencies are captured. Meanwhile, the complementarity of fault-related features is enhanced by means of the cross-attention mechanism, and the feature propagation is optimized through layer normalization and residual connection to improve the feature capture ability of the model. Finally, the Hyperband hyperparameter optimization algorithm is used to select the best hyperparameter combination. The experiment adopts the wind turbine gearbox dataset of Beijing Jiaotong University and the rolling bearing dataset of Case Western Reserve University in the United States. The ablation experiments are carried out to verify the importance of each module for the performance improvement of the model. The best prediction accuracy rates of this model on the two datasets are 98.39% and 100%, respectively. The experimental results show that the proposed method is superior to comparison models such as CNN, Convolutional LSTM Network (CNN-LSTM), and Convolutional Neural Network Bidirectional Gated Recurrent Unit (CNN-BIGRU) in evaluation indicators such as prediction accuracy rate, recall rate, and [Formula: see text]1 score, and has better capabilities in fault diagnosis. The proposed method can be applied to the fault diagnosis of equipment such as wind turbine units and high-speed trains.