|
| Year | Input data and features | DL model | Application | Remarks |
|
| 2016 | Raw vibration | SAE + DNN [57] | Motor fault classification | Integrated “dropout” regularization allowed to avoid overfitting, which in turn improved the classification performance |
| Acoustic emissions/sound signals + STFT | SAE [58] | Rolling bearing fault classification | STFT allowed for fast and effective feature extraction, which in turn increased accuracy of the classifier |
|
| 2017 | Vibration + FFT | AE + ELM [59] | Bearing fault classification | AE-ELM approach increased the speed of the training process. The approach was much faster than SAE. However, the model did not perform well for online sequential learning |
| Raw vibration | SAE [60] | Rolling bearing fault severity level classification | Deep SAE inputted with added noise vibration data has effectively overcome the overfitting problem posed by small training datasets |
| Vibration + time domain features + wavelet energy features + power spectrum features | SAE + DNN [61] | Rolling bearing fault severity level classification | Hybrid features pool was able to tackle the nonlinearity in the vibration data and resulted in more effective classification results |
| Raw vibration | DAE and SAE [62] | Rolling bearing fault classification | The comparative analysis performed between DAE and SAE classifiers showed that the DAE outperformed SAE classifier due to its potentiality of learning complex nonlinear mapping relationships |
| Vibration + compression | SAE + DNN [63] | Rolling bearing fault classification | Data compression technique with SAE enables effective fault classification, with huge datasets through an easy approach |
|
| 2018 | Raw vibration | Ensemble AE [64] | Rolling bearing fault classification | A novel ensemble of 15 AEs, each with different activation function overcome the limitations of individual AE models and removed manual feature extraction |
| Current + time domain and frequency domain features | Deep AE [65] | Unsupervised bearing fault prediction | Results showed effectiveness of the technique in terms bearing fault prediction with clear clustering and high accuracy |
|
| 2019 | Vibration + FFT | ESAE [66] | Rolling bearing fault classification | FFT with ESAE has revealed superior performance of the proposed model in comparison with traditional models |
| Current + FFT | AE [67] | Induction motor fault classification | FFT with the AE yielded superior performance compared to the existing models |
| Raw vibration | SAE + GRU [68] | Rolling bearing fault classification | An optimal hybrid DL mode constructed using SAE and GRU can extract rich features from raw vibration data. Results confirmed the superiority of the proposed model |
|
| 2020 | Raw vibration | Stacked pruning DAE [69] | Rolling bearing fault classification | A novel model called SPADE with pruning operation increased efficiency and precision of the model by decreasing the training amount of the model |
|
