A signal recovery method for bridge monitoring system using TVFEMD and encoder-decoder aided LSTM. (15th June 2023)

• Record Type: Journal Article
• Authors: Xin, Jingzhou; Zhou, Chaoying; Jiang, Yan; Tang, Qizhi; Yang, Xianyi; Zhou, Jianting
• Is Part Of: Measurement. Volume 214(2023)
• Publication Date: 2023-06-15
• Subjects: Bridge engineering -- Health monitoring -- Signal recovery -- Time varying filtering based empirical mode decomposition -- Encoder-decoder -- Long short-term memory; Weights and measures -- Periodicals; Measurement -- Periodicals; Measurement; Weights and measures; Periodicals; 530.8
• DOI: 10.1016/j.measurement.2023.112797
• Languages: English
• ISSNs: 0263-2241

Abstract

Highlights: TVFEMD is used to deal with the complex bridge monitoring data; ED-LSTM is developed to accurately and efficiently recover the missing data; Case study and robustness analysis prove the superiority of the proposed method. Abstract: Accurate monitoring data in bridge health monitoring systems are critical for grasping the structural operation status. However, because of data missing and distortion induced by the sensor malfunction and the interference of system or measurement noise, it is hard to obtain reliable monitoring data. To settle this problem, this paper proposes a method for the recovery of bridge monitoring data based on the combination of the time varying filtering based empirical mode decomposition (TVFEMD), encoder-decoder (ED), and long short-term memory neural network (LSTM). This hybrid method can transform the problem of data recovery into a series of forecasting tasks. Specifically, TVFEMD is first used to decompose the original signal into several regular intrinsic mode functions (IMFs) with different frequency bands. Then, ED is employed to improve the interpretability and applicability of the original LSTM via constructing a sequence-to-sequence network architecture (i.e., ED-LSTM). Finally, ED-LSTM is established for each IMF to perform the desired prediction, and the final recovery result can be generated by summing each desired prediction in a linear manner. Case study based on the monitoring data in the bridge health monitoring system is used to illustrate the effectiveness of the proposed method. The results show that this method is superior to the other compared methods in terms of the data recovery ability. For example, the proposed method can realize an overall improvement of 33.90% in comparison with the original LSTM. Therefore, this method may have a potential for the recovery of bridge health monitoring data in practice.