Development of an Optimized Condition Estimation Model for Bridge Components Using Data-Driven Approaches

To support bridge management systems (BMSs) in establishing strategic maintenance plans to preserve the condition of aging bridges, it is important to estimate a bridge component's future condition reliably. To this end, many data-driven studies have attempted to apply diverse algorithms and explore major factors influencing the condition of specific components. Despite these efforts, it is still difficult to construct a robust and generally applicable condition estimation model for bridge components regardless of the characteristics of the BMS data because BMS data become heterogeneous and complex by period, region, or country. Therefore, the objective of this study is to develop an optimized condition estimation model for bridge components using data-driven approaches. To achieve the main objective, the proposed model included the following elements: (1) outstanding algorithm selection by comparing the performance of diverse algorithms; and (2) influential variable identification by utilizing the recursive feature elimination (RFE) method according to the permutation variable importance. Based on a case study to estimate the condition grades of decks on concrete-girder bridges using the Korean Bridge Management System (KOBMS) data, extreme gradient boost (XGBoost) was selected as the optimal algorithm, and influential variables were identified such as "bridge age" and "first past condition grade of deck." Finally, the optimized model based on the integrated results of the algorithm selection and the influential variables identification showed good performance with an average weighted average F1 score of 0.876. The outcome of the research will contribute to reliably estimating the future condition of bridge components by constructing an optimal model suitable for each BMS data and supporting strategic maintenance decisions based on the expected components' condition for proactive bridge management.