Name
Abstract | ||
|---|---|---|
Unlike traditional machine learning techniques, convolutional neural networks (CNNs), one of deep learning methods, automate the feature extraction process required for an effective classification. In general, CNN based bearing fault diagnosis analyzes raw signals to classify the localized faults. However, bearings are subjected to non-stationary speeds due to various operating conditions, and thus, CNN cannot determine optimal features of the various conditions while analyzing raw signals, reducing classification accuracy. In this paper, we propose a pre-processing step to improve the performance of the CNN based fault diagnosis by extracting envelope spectrums (ES) on the raw signals. As ES demodulates the signals to provide the information inherent in defect frequency of faults and its variations to non-stationary speeds, CNN can learn to extract distinctive features to diagnose bearing defects effectively. The proposed method is evaluated on acoustic emission based low speed bearing data. The trained CNN model is tested on data with different revolutions per minute (RPM), and it achieves the classification accuracy greater than 94.8%. |
Year | Venue | Field |
|---|---|---|
2017 | MIWAI | Pattern recognition,Convolutional neural network,Computer science,Bearing (mechanical),Feature extraction,Invariant (mathematics),Artificial intelligence,Deep learning,Acoustic emission,Machine learning |
DocType | Citations | PageRank |
Conference | 1 | 0.36 |
References | Authors | |
3 | 3 | |
Name | Order | Citations | PageRank |
|---|---|---|---|
| Dileep Kumar Appana | 1 | 8 | 1.27 |
| Ahmad Wasim | 2 | 14 | 3.88 |
| Jong Myon Kim | 3 | 144 | 32.36 |