Review Article

The Importance of Feature Processing in Deep-Learning-Based Condition Monitoring of Motors

Table 5

Application of RNNs and its variants in condition monitoring of motors.

YearInput data and featuresDL modelApplicationRemarks

2016Raw vibrationLSTM [102]Motor fault detectionThe model outperformed the existing models owing to its deep architecture

2017Raw vibration from multiple sensorsLFGRU [103]Motor fault classificationThe model was able to effectively classify motor faults owing to deep structure without any extensive feature extraction
Vibration + time domain features + frequency domain featuresRNN [104]Rolling bearing fault predictionModel showed better prediction results with frequency domain features

2018Raw vibrationLSTM [105]Motor fault classificationThe results confirmed effectiveness of the technique compared to existing methods owing to gated functions of the model
Vibration + WTD + time domain featuresSAE + LSTM [106]Prediction of bearing performance degradationResults confirmed effectiveness of the method in achieving the task compared to the conventional methods

2019Vibration + hybrid feature + RQALSTM [107]Induction motor fault classificationMethod effectively classified the motor faults due to the antinoise feature extraction techniques
Raw vibrationLSTM [108]Rolling bearing fault classificationThe results demonstrated efficiency in the bearing fault classification owing to inherent mechanism and deep architecture
Vibration + temperature + WELSTM + PSO [109]Bearing degradation assessment and RUL predictionResults confirmed effectiveness of the method in estimating the bearing life states and degree of degradation
Raw vibrationGRU + AFSA + ELM [110]Adaptive rolling bearing fault classificationResults revealed robust performance of the model owing to optimized parameter selection by AFSA and accurate classification by ELM
Raw vibrationBiD-LSTM [111]Bearing fault classificationResults demonstrated effectiveness of the method owing to its long-term dependency in time series data
Vibration + time, frequency, hybrid featuresHRGUN + KPCA + EWMA [112]Future HI prediction and RUL prediction of the bearingKPCA and EWMA allowed to effectively track the degradation process and promisingly predicted the HI and RUL of the bearing

2020Raw vibration + complex wavelet packet energy moment entropyEnhanced deep GRU [113]Early bearing fault classificationThe monitoring health index technique and the modified training algorithm allowed to detect fault at an early stage