Bayesian model updating for steel structures with non-ideal boundary conditions using image generation networks and frequency response functions

• Published in: Mechanical systems and signal processing Vol. 234; p. 112849
• Main Authors: Liu, Jiming, Duan, Liping, Jiang, Yuheng, Lin, Siwei, Miao, Ji, Zhao, Jincheng
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.07.2025
• Subjects: Bayesian model updating; Frequency response function; Image generation network; Non-ideal boundary condition; Structural finite element model; Bayesian model updating; Non-ideal boundary condition; Image generation network; Frequency response function; Structural finite element model
• ISSN: 0888-3270
• DOI: 10.1016/j.ymssp.2025.112849

•Ex-cDCGAN is proposed to concurrently predict entire frequency response function curves across all measurement points.•The DREAM-based Bayesian model updating using frequency response functions and image generation networks is implemented.•Two feasible strategies are presented to facilitate model updating for real structures with non-ideal boundary conditions.•The implemented algorithm is adopted to update the finite element model of a complex steel structure. The application of Frequency Response Function (FRF)-driven Bayesian model updating to Finite Element (FE) model of real structures, particularly building structures, faces three primary challenges: inconsistent inference outcomes across various updating schemes, complexities arising from non-ideal boundary conditions, and the detrimental effects of significant measurement noise. To address these issues, an image generation network, termed Ex-cDCGAN, is first proposed to facilitate comparative trials for selecting the appropriate updating scheme. Ex-cDCGAN is capable of predicting entire FRF curves across all measurement points simultaneously, thereby eliminating the need to retrain surrogate models for each new updating scheme during comparative trials. Subsequently, the proposed surrogate model is incorporated into an advanced FRF-driven Bayesian model updating algorithm. Furthermore, the strategies of Feasible Range Constraint (FRC) and Principal Component Analysis (PCA) dimensionality reduction are adopted to enhance the applicability of the implemented algorithm in updating tasks with non-ideal boundary conditions and low Signal-to-Noise Ratio (SNR) measurements. Eventually, the combined efficacy of these strategies and the updating algorithm is validated through both numerical and experimental cases, demonstrating the applicability of our approach for FE model updating of real structures.