|
| Year | Input data and features | DL model | Application | Remarks |
|
| 2016 | Vibration + hybrid features | DBM [75] | Rolling bearing fault classification | Among three features, hybrid features allowed the model to perform much better than other models |
| Multisensor vibration + time domain features | DBN [76] | Bearing fault classification | Multivibrations fusion technique with DBN outperformed the model with single sensor |
| Vibration + time domain features | DBN [77] | Bearing degradation states classification | Weibull distribution dealt with fluctuations in statistical features, which in turn enhanced the ability to classify the bearing degradation states |
|
| 2017 | Vibration + FFT | DBN [21] | Induction motor fault classification | The developed automatic fault classifier effectively detected faults of motors in manufacturing |
|
| 2018 | Raw vibration | DBM [78] | Rolling bearing fault classification | Binary units of DBM were replaced with Gaussian units so that DBM can process real-value data. Results confirmed the usefulness of the method |
|
| 2019 | Acoustic emission signals | DBN + PCA + ls-SVM [79] | Rolling bearing fault classification | Combination of DBM and PCA enabled for learning better features, which in turn increased the accuracy of the model |
| Vibration + VMD + HT | VHDBN [80] | Rolling bearing fault classification | Employed VMD and HT for better feature extraction. The results of proposed method confirmed great advantages in classification accuracy |
|
| 2020 | Raw vibration + sliding window + coarse-grained method | MSDBN [81] | Various mechanical fault classification | Multiscale feature extractor enabled for learning better features, which in turn increased the accuracy of the method than standard DBN |
| Vibration + wavelet package decomposition (WPD) | DBN + D-S theory + GA + PSO [82] | Bearing fault and severity level classification | GA-PSO algorithm improved learning capability and computation efficiency of the DBM classifier. In addition, the D-S theory-based vibration data fusion increased the accuracy of the model |
|
