|
| Year | Input data and features | DL model | Application | Remarks |
|
| 2016 | Raw vibration | CNN + DFT [84] | Motor fault classification | Owing to the automatic feature extraction, the method produced promising results compared to classical feature engineering methods |
| Raw vibration | DTS-CNN [85] | Motor fault classification | Dislocation layer yielded better performance than standard CNN owing to rich feature extraction |
| Raw current | 1D-CNN [86] | Bearing fault classification | The simple architecture of the model allowed real-time detection of the faults |
| Raw vibration | ADCNN [87] | Bearing fault classification and their severity level classification | Adaptive architecture of the CNN allowed to learn rich features from the data, which in turn increased its performance |
|
| 2017 | Vibration + WPI | Multiscale deep CNN [88] | Spindle bearing fault classification | Novel feature learning techniques with multiscale deep CNN allowed the model to outperform other traditional models through learning local and global features simultaneously |
| Raw vibration data from two sensors | IDSCNN [89] | Bearing fault classification | Improved data fusion technique produced better classification results by addressing the problems in the conventional feature extraction techniques |
|
| 2018 | Raw vibration | TICCN [90] | Bearing fault classification | Dropout layers with the wide and small convolutional layers supressed the noise of data. Meanwhile, ensemble model allowed to attain high classification accuracy |
| Raw vibration | CNN [91] | Motor health indication | The method was able to overcome outlier regions and effectively learned features from the data |
| Raw vibration | DNCCN [92] | Bearing fault classification | The model produced effective classification results by overcoming data imbalance problem |
|
| 2019 | Raw vibration | VI-CNN [93] | Rolling bearing fault classification | 2D vibration image allowed the method to effectively classify the faults without any feature learning and denoising technique |
| Raw vibration | ST-CNN [94] | Bearing fault classification | The method produced higher performance than existing methods owing to automatic feature extraction by S-layer |
| Vibration + STFT | ICN [95] | Bearing fault classification | Inception block resulted in better generalization than CNN |
| Raw current signatures | CNN + IF [96] | Bearing fault classification | Promising results were obtained owing to information fusion |
|
| 2020 | Raw vibration + WPT | CNN [97] | Rolling bearing fault classification | Classification performance improved owing to the gray-scale vibration images |
| Vibration + STFT | CNN + SELU function [98] | Rolling bearing fault classification | The method effectively classified the fault owing to the 2D-images of the data and regularization |
| Raw current signatures | CNN [99] | Stator winding fault detection | The method can effectively detect stator winding faults from raw current data without any preprocessing |
|
