Robust correlation mapping of train-induced stresses for high-speed railway bridge using convolutional denoising autoencoder. (March 2023)

• Record Type: Journal Article
• Authors: Niu, Jin; Li, Zhonglong; Zhuo, Yi; Di, Hao; Wei, Jianfeng; Wang, Xin; Guo, Yapeng; Li, Shunlong
• Is Part Of: Structural health monitoring. Volume 22:Number 2(2023)
• Page Start: 832
• Page End: 845
• Publication Date: 2023-03
• Subjects: Train-induced stress -- high-speed railway bridge -- convolutional denoising autoencoder -- spatiotemporal correlation -- robust data mining; Structural health monitoring -- Periodicals; Structural stability -- Periodicals; Strength of materials -- Periodicals; Nondestructive testing -- Periodicals; Constructions -- Stabilité -- Périodiques; Résistance des matériaux -- Périodiques; Contrôle non destructif -- Périodiques; Electronic journals; 624.17
• DOI: 10.1177/14759217221095191
• Languages: English
• ISSNs: 1475-9217

Abstract

Train-induced stresses in different monitoring points not only reflects local mechanical characteristics of the structural components but also has inherent spatiotemporal correlations in the high-speed railway bridges. Mapping correlations among the stress responses can assist recognizing train-induced stress pattern and lay foundation for structural health diagnosis. However, correlation mapping and feature extraction of structural responses rely heavily on the data integrity, the pre-constructed model may be out of action due to data acquisition/transmission error, sensor faults, etc., commonly exists in the structural health monitoring system. Considering the characteristics of various incomplete train-induced stresses, this study presents a robust correlation mapping of incomplete data to complete data using one-dimensional convolutional denoising autoencoder. Stacked convolutional layers are employed as encoder to extract robust spatiotemporal feature of incomplete stresses, and transposed convolutional layers served as decoder to reconstruct denoised and complete stresses. In the training strategy, various incomplete data conditions, where stress data are lost continuously or discretely with different missing rates, are considered as training samples, making the established correlation mapping robust, accurate, and adaptive. The application on a high-speed railway truss bridge demonstrates that the proposed method can robustly reconstruct the complete stress data under different data loss conditions. The method can also be employed to assess the importance of any sensor combinations to the monitoring item, which shed light on the maintenance of sensor network.