Bridge damage characterisation using machine learning: methods and advances

Abstract

Bridge deflection is a descriptive proxy for potential bridge deterioration and damage. It can be used to determine bridge condition and link this to actionable damage states for timely and accurate damage mitigation and adaptation. While design guidelines mandate strict deflection control at the design stage, primarily for serviceability, there are currently no assessment guidelines or available framework to facilitate bridge damage identification based on bridge deck deflections. A thorough review of the literature revealed that the main reason that deflection is not used as a damage proxy is the complex mechanics underlying its development. A state-of-the-art review is presented to efficiently characterise global bridge damage related to deck deflections. The approach goes beyond existing methods by striving to reveal the causes of bridge deck deflection and their interdependencies, offering a clearer understanding and interpretation of the factors driving this phenomenon. Given the significant uncertainties around deflection causes and the impracticality of complex, tedious analyses for large bridge portfolios, Machine Learning (ML) is proposed as a scalable solution that reduces modelling effort, enhances explainability, and can successfully correlate deflections with damage levels. It then proposes a conceptual Physics-Based ML approach that correlates deflection patterns with actionable damage states, offering a roadmap for future research to enhance bridge damage characterisation. Unlike prior studies, it integrates all major deterioration mechanisms and their interactions into a unified deflection analysis.