Mechanical fault time series prediction by using EFMSAE-LSTM neural network. (March 2021)

• Record Type: Journal Article
• Authors: Guo, Jianwen; Lao, Zhenpeng; Hou, Ming; Li, Chuan; Zhang, Shaohui
• Is Part Of: Measurement. Volume 173(2021)
• Publication Date: 2021-03
• Subjects: Mechanical equipment -- Error fusion of multiple SAEs -- Long short-term memory -- Multi-threshold -- Prediction; Weights and measures -- Periodicals; Measurement -- Periodicals; Measurement; Weights and measures; Periodicals; 530.8
• DOI: 10.1016/j.measurement.2020.108566
• Languages: English
• ISSNs: 0263-2241

Abstract

Highlights: A new fault time series prediction is carried out for nonlinear time series curve. The data with large noise and inconspicuous feature were extracted and processed. A trend curve based on square prediction error is used for condition monitoring. The proposed method achieves efficiency and effectiveness. Abstract: The working state of mechanical parts plays an important role in the safe and reliable operation of equipment. Therefore, in order to enhance the dependability and security of mechanical equipment, it is very important to accurately predict the changing trend of mechanical components in advance. This paper proposes a method combining error fusion of multiple sparse auto-encoders with long short-term memory for predicting mechanical fault time series. First, error fusion of multiple sparse auto-encoders layer can extract multi-feature time series and fuse them into a square prediction error trend curve representing each channel and a threshold control line of bearing system health judgment. Then, long short-term memory layer predicts the irregular trend in the square prediction error trend curve, and sets a number of threshold control lines according to different machine fault variation trends for more accurate prediction. The experimental results prove the availability and superiority of this method in the prediction of mechanical fault time series.