Title: Anomaly detection of hydro-turbine based on audio feature extraction of deep convolutional neural network

Authors: Shengming He; Zhaocheng Wang; Bo Liao; Jie Zeng; Haorui Liu

Addresses: Yalong River Hydropower Development Company Ltd., Chengdu, Sichuan, China ' Yalong River Hydropower Development Company Ltd., Chengdu, Sichuan, China ' Yalong River Hydropower Development Company Ltd., Chengdu, Sichuan, China ' Yalong River Hydropower Development Company Ltd., Chengdu, Sichuan, China ' Tsinghua AI Plus, Beijing, China; Institute of Computer and Information, Dezhou University, Dezhou, China

Abstract: Anomaly detection of the hydro-turbine operating status is required to achieve safe monitoring of the operating status of hydro-turbines. The detection of hydro-turbine anomalies based on sound signals pertains to acoustic scene recognition. In this study, the features of sound signals were extracted based on the MobileFaceNet neural networks. Using the feature vectors, an improved Gaussian Mixed Model (i-GMM) was built, and the anomaly detection on the test samples was performed. The effectiveness of the i-GMM model anomaly detection method was verified to be capable of achieving 100% based on the bearing data set. The sound data collected from different measurement points in the hydro-turbine served as the training samples to develop the i-GMM model for the operation state. The model results output the anomalous sound events that occurred in the range of the hydro-turbine, which were manually labelled as a wide variety of construction activities.

Keywords: hydro-turbine; MobileFaceNet; improved Gaussian mixed model; anomaly detection; STFT; feature vectors; audio data.

DOI: 10.1504/IJCAT.2023.135584

International Journal of Computer Applications in Technology, 2023 Vol.73 No.3, pp.192 - 202

Received: 02 Nov 2022
Accepted: 14 Apr 2023
Published online: 18 Dec 2023 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article