Data-Driven Bridge Maintenance Cost Estimation Framework for Annual Expenditure Planning

Abstract

Bridge maintenance costs have increased due to the growth in the number of facilities and their extended periods of use. The distribution of the limited maintenance budget by considering the conditions and properties of the bridge is highly required. While much research exists to provide rough estimates of maintenance costs based on indicators, there has been a lack of a framework that incorporates historical management data at the bridge element and repair method levels. This study proposed a framework for estimating bridge maintenance costs using historical records collected in the Korean bridge management system (BMS), involving several phases of data integration. First, unit cost estimation models by repair methods were generated using the extreme gradient boosting algorithm, utilizing repair project records. A total of 31 models were developed with the average weighted f1-score of 0.82. Subsequently, these models were employed to update expected repair costs for each bridge defect recorded in the inspection reports. As a result, the updated individual costs were aggregated to estimate the required expenditure of the bridge maintenance based on its condition grade, which can be used to estimate the annual maintenance expenditure for the following years. The proposed framework underwent validation focusing on bridge elements such as deck, pier/abutment, and pavement. The error rates differed with 25.37% for deck, 8.27% for pier/abutment, and 27.18% for pavement, influenced by data availability and variation element characteristics. This framework contributes to the body of knowledge regarding bridge maintenance by incorporating historical data with data-driven approaches. As a result, it assists decision-makers in gaining a better understanding of the required cost information, ultimately enhancing the decision-making process.