Surrogate Modelling of Bridge Deterioration Using Machine Learning and Defect Data

Bridge maintenance relies heavily on group-level bridge deterioration models. However, current models that use inspection ratings struggle with data reliability and the subjective nature of inspector evaluations. This study proposes a new surrogate deterioration model for groups of bridges that utilizes data from visual inspections and leverages automated machine learning. The model measures bridge component deterioration using indicators obtained from these inspections, incorporating traffic flow, structural configurations, and defect data. A two-layer surrogate model is then developed using automated machine learning techniques. Validated using data from multiple bridges in China, this surrogate model demonstrates high predictive accuracy with little need for adjustment, effectively forecasting deterioration at both the individual and group levels across different maintenance approaches.