Gearbox fault diagnosis method with multi-domain feature fusion based on attention mechanism

ObjectiveAiming at the limitation that single data domain models are difficult to accurately identify subtle fault features in gearbox fault diagnosis, a fault diagnosis method with multi-domain feature fusion under attention mechanism was proposed to improve diagnostic accuracy, stability and generalization ability.MethodsFirstly, dimensionless features and spectral features were extracted from the time domain and frequency domain of vibration signals. Secondly, deep time-frequency domain features were extracted by combining continuous wavelet transform with convolutional neural network (CNN). Then, an attention mechanism was introduced to dynamically weight and fuse multi-domain features, strengthening key features and weakening redundant information. Finally, a classifier was used to complete fault identification, and the effectiveness of the method was verified based on a secondary gearbox test dataset.ResultsTest results on a secondary gearbox test dataset show that the proposed method achieves a diagnostic accuracy of 99.77%, outperforming single-domain models and verifying its effectiveness and stability in identifying weak faults and adapting to multiple operating conditions.