Title: NASA space station rolling bearings anomaly detection based on PARA-LSTM model

Authors: Yingqian Zhang; Jiaye Wu; Hui Xie; Rongru Hua; Qiang Li

Addresses: School of Civil Engineering, Sichuan University of Science and Engineering, Zigong 643000, China ' School of Mechanical and Electrical Engineering, Southwest Petroleum University, Chengdu 610500, China ' Technology Center, Sichuan Shengtuo Testing Technology Co., Ltd., Chengdu 610045, China ' Technology Center, Sichuan Shengtuo Testing Technology Co., Ltd., Chengdu 610045, China ' Technology Center, Sichuan Shengtuo Testing Technology Co., Ltd., Chengdu 610045, China

Abstract: Anomaly detection in time series data identifies abnormal events or behaviours. Traditional methods include principal component analysis (PCA) combined with Mahalanobis distance and long short-term memory (LSTM). Autoencoders and neural network techniques have been applied to the problem of anomaly detection. Still, challenges remain, such as large training data volume, network parameter initialisation, low training efficiency, and poor anomaly detection performance. This paper proposes an anomaly detection method based on parallel-long short-term memory (PARA-LSTM), which constructs two parallel processing structures. The method was tested on the rolling bearing vibration dataset collected by the NASA space station. It could detect anomalies five days ahead of the actual system destruction time, outperforming the PCA method by detecting anomalies one day earlier. PARA-LSTM has good performance, stability, and generalisation ability.

Keywords: autoencoder; bearing vibration; anomaly detection; Mahalanobis distance; autoencoder network; parallel-long short-term memory; PARA-LSTM.

DOI: 10.1504/IJSNET.2024.136334

International Journal of Sensor Networks, 2024 Vol.44 No.1, pp.49 - 61

Received: 01 Sep 2023
Accepted: 07 Sep 2023
Published online: 30 Jan 2024 *

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