Data-driven structural condition assessment for high-speed railway bridges using multi-band FIR filtering and clustering. (July 2022)

• Record Type: Journal Article
• Authors: Li, Yiwei; Ding, Youliang; Zhao, Hanwei; Sun, Zhen
• Is Part Of: Structures. Volume 41(2022)
• Page Start: 1546
• Page End: 1558
• Publication Date: 2022-07
• Subjects: High-speed railway bridge -- Structural condition assessment -- Influence line -- Dynamic load factor -- MFIR filtering -- GMM clustering; Structural engineering -- Periodicals; 624.1
• DOI: 10.1016/j.istruc.2022.05.071
• Languages: English
• ISSNs: 2352-0124

Abstract

Highlights: Multi-band FIR filtering extracts the train-induced response features of high-speed railway bridges using monitoring data. Gaussian mixture model clustering reveals the distribution pattern of response features. The data-driven approach leads to the dynamically updated and real-time bridge condition assessment. An operational long-span steel-truss bridge is assessed via the proposed method as a validation. Abstract: The monitoring data makes it feasible to inspect and assess the structural condition of bridges in real time. Among the diverse in situ data of high-speed railway bridges under varying operational environments, dynamic responses caused by passing trains can offer insight into the mechanical properties of the bridge structure. Based on the train-induced response features of influence line (IL) and dynamic load factors (DLF) extracted from raw measured data, a comprehensive data-driven approach is developed for structural condition assessment of high-speed railway bridges, which is applied to a long-span steel-truss bridge as a validation. Considering the sparsity of the train-induced static response in the frequency domain, the multi-band finite impulse response (MFIR) filtering is used to extract the train-induced response features. The features are clustered via the Gaussian mixture model (GMM), and the two-level objective for structural condition evaluation and degradation warning of bridges can be achieved through the dynamically updated clustering results and probabilistic models. The results demonstrate that (1) MFIR filtering can effectively reject abnormal or interfering data and accurately extract the train-induced response features, and (2) the intrinsic nature and laws of features can be revealed by GMM clustering, which provides a statistical premise for the reliability analysis of bridge structures.