Abstract
Fault diagnosis method for gearbox by using extensive empirical wavelet transform-based entropy local tangent space alignment and imporved deep extreme learning machine (EEWTEWLTSA-IDELM) is presented in this paper. Firstly, extensive empirical wavelet transform (EEWT) is used to decompose the vibration signals of gearbox because the novelty of EEWT lies in the pre-estimation of frequency threshold based on signal amplitude modulation.Secondly, a novel entropy-weighted local tangent space alignment (EWLTSA) is presented to reduce the feature dimensionality of gearbox data, entropy weighting quantifies the clarity of local neighborhood structure and re-weights or filters neighborhoods accordingly, thereby improving both graph construction and tangent-space estimation.Then, enhanced feature extraction method for gearbox using EEWT and EWLTSA (EEWTEWLTSA) is obtained.Thirdly, a novel improved deep extreme learning machine (IDELM) is presented for fault diagnosis of gearbox.In the improved deep ELM algorithm, we incorporate regularization term into the loss function to control model complexity and enhance generalization, and a differential evolution strategy-based grey wolf optimization is presented to optimize the training parameter of the deep ELM algorithm. In case 1,the proposed EEWTEWLTSA-IDELM method achieves a remarkable gearbox fault diagnosis accuracy of 99.5%, demonstrating superior performance compared with alternative approaches. The comparative methods yield the following accuracy rates: EEWTLTSA-IDELM (98%), EEWTLTSA-DELM (97%), EWTLTSA-DELM (96%), EWT-ELM (93%), MDCNN-LSTM (94%),LLSTM (94.5%) ,VMD-DNN (92.5%), and WT-ANN (90%). In case 2,the proposed EEWTEWLTSA-IDELM method achieves a remarkable gearbox fault diagnosis accuracy of 98.5%,and the comparative methods yield the following accuracy rates: EEWTLTSA-IDELM (96.5%), EEWTLTSA-DELM (95.5%), EWTLTSA-DELM (94.5%), EWT-ELM (91%), MDCNN-LSTM (92.5%),LLSTM (93.5%) ,VMD-DNN (91.5%), and WT-ANN (88%).To evaluate the anti-interference capability of EEWTEWLTSA-IDELM, the methods including EEWTLTSA-IDELM, EEWTLTSA-DELM, EWTLTSA-DELM, EWT-ELM, MDCNN-LSTM, LLSTM, VMD-DNN, WT-ANN are compared with EEWTEWLTSA-IDELM under noise interference with different signal-to-noise ratios. The experimental findings confirm that RVMD-CCGTELM is well-suited for fault diagnosis of gearbox, even in the presence of noise interference with different signal-to-noise ratios.