A deep learning based method for classifying multi-class faults in a gearbox is presented. A set of 900 vibration signals representing the normal condition and nine faults comprises the dataset used in this research. The recorded vibration signals are pre-processed for extracting the first and second derivatives as well as the first five Intrinsic Mode Functions (IMFs) by applying the Empirical Mode Decomposition (EMD) method. A 2D representation of these signals is the feature space used for classifying ten conditions of a gearbox using a Long Short Term Memory (LSTM) neural network. The 2D feature space is subdivided along the temporal axis in segments of the same size as the LSTM network. These segments are classified and a voting systems is proposed for attaining the signal classification. A 10-fold cross-validation is used for evaluating the proposed deep learning model. An average accuracy up to 99.4 % for classifying the faults is attained during the cross-validation.
|Title of host publication||Proceedings - 2019 International Conference on Sensing, Diagnostics, Prognostics, and Control, SDPC 2019|
|Editors||Chuan Li, Shaohui Zhang, Jianyu Long, Diego Cabrera, Ping Ding|
|Publisher||Institute of Electrical and Electronics Engineers Inc.|
|Number of pages||7|
|State||Published - Aug 2019|
|Event||2019 International Conference on Sensing, Diagnostics, Prognostics, and Control, SDPC 2019 - Beijing, China|
Duration: 15 Aug 2019 → 17 Aug 2019
|Name||Proceedings - 2019 International Conference on Sensing, Diagnostics, Prognostics, and Control, SDPC 2019|
|Conference||2019 International Conference on Sensing, Diagnostics, Prognostics, and Control, SDPC 2019|
|Period||15/08/19 → 17/08/19|
Bibliographical noteFunding Information:
ACKNOWLEDGMENT This work was supported in part by the MOST Science and Technology Partnership Program (KY201802006), and Universidad Politécnica Salesiana through the research group GIDTEC.
© 2019 IEEE.
Copyright 2020 Elsevier B.V., All rights reserved.
- Deep learning
- Faults classification
- Long short term memory networks
- Vibration signals