A Dynamic Dimension Reduction Method for Structural Health Monitoring Data Based on Sparse Bayesian Inference

With the rapid development of structural health monitoring systems and sensor technologies for civil structures, various methods have been developed for condition monitoring and performance assessment based on measured structural response data. However, structural responses under working conditions are influenced by environmental and operational factors. Some of these factors are strongly correlated with structural responses, while others show weak or even negligible associations. This inconsistency often leads to data redundancy and limits the effectiveness of structural performance assessment methods. To address this issue, this paper proposes a unified Bayesian sparse Gaussian process (BSGP) model that integrates adaptive feature selection directly within a nonlinear regression framework. The proposed method can predict the structural response under different environmental conditions and prunes irrelevant environmental features simultaneously. The effectiveness of the proposed BSGP model is validated by both synthetic data and a year‐long monitoring data of a full‐scale bridge. The model successfully pruned all five redundant channels in the simulation and demonstrated higher prediction accuracy than a standard Gaussian process model in the bridge case study. Furthermore, the analysis reveals that the sensitivity of influencing factors such as wind speed and humidity is time dependent, showcasing the capability of the proposed method to support adaptive monitoring strategies. The proposed framework provides a robust tool for intelligent data analysis in SHM, enhancing model interpretability and offering guidance for optimizing monitoring system costs.