Hydroelectric Generating Unit Fault Diagnosis Using 1-D Convolutional Neural Network and Gated Recurrent Unit in Small Hydro

• Published in: IEEE sensors journal Vol. 19; no. 20; pp. 9352 - 9363
• Main Authors: Liao, Guo-Ping, Gao, Wei, Yang, Geng-Jie, Guo, Mou-Fa
• Format: Journal Article
• Language: English
• Published: New York IEEE 15.10.2019; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: 1-dimension convolutional neural network (1-D CNN); Acceleration; Algorithms; Artificial intelligence; Artificial neural networks; Data structures; Deep learning; Diagnostic systems; Dimensional stability; Fault diagnosis; Feature extraction; gated recurrent unit (GRU); Hydroelectric generating unit (HGU); Hydroelectric power generation; Machine learning; Neural networks; Reconstruction; Segmentation; Sensors; Vibration; Vibrations
• ISSN: 1530-437X; 1558-1748
• DOI: 10.1109/JSEN.2019.2926095

Machine learning algorithm based on hand-crafted features from the raw vibration signal has shown promising results in the hydroelectric generating unit (HGU) fault diagnosis in recent years. Such methodologies, nevertheless, can lead to important information loss in representing the vibration signal, which intrinsically relies on engineering experience of diagnostic experts and prior knowledge about feature extraction techniques. Therefore, in this paper, an effective and stable HGU fault diagnosis system using one-dimensional convolutional neural network (1-D CNN) and gated recurrent unit (GRU) based on the sequence data structure is proposed. First, the raw vibration data is reconstructed by data segmentation, which can improve training efficiency. Second, the reconstruction data under the influence of different running conditions and various fault factors can be effectively and adaptively learned by 1-D CNN-GRU and then determine information fault categories via network inference. Finally, four machine learning methods are applied to diagnosis the reconstruction data based on the experimental dataset. The performance of the proposed method is verified by comparing with the results of other machine learning techniques. Furthermore, the fault diagnostic model, which is trained by the practical vibration signal, has successfully applied in engineering practice.