Examination of crack damage prediction in concrete bridges using machine learning algorithms

Abstract

The issue of infrastructure deterioration has been an ongoing concern, and research based on the inspection results of bridges regarding the evaluation of structural damage has been continuously conducted. In this study, we employed a machine learning algorithm designed for classification prediction to forecast the extent of crack damage in concrete bridges. By utilizing this machine learning-based model for predicting crack damage, the aim is to extract the intricate factors influencing damage in concrete bridges, with a long-term goal of establishing efficient maintenance and management methods for these bridges. In this research, we implemented crack damage predictions for the main girders of concrete bridges using various machine learning algorithms. As a result, methods known as gradient boosting, specifically XGBoost and LightGBM, demonstrated commendable prediction accuracy. Additionally, through the predictive model using LightGBM, we conducted the selection of significant explanatory variables and visualized the impact on crack damage extent using SHAP, further discussing these findings.