Explainable machine learning model for classifying vehicle-impact damage of reinforced concrete bridge columns

Abstract

This study aimed to develop a machine learning model to predict the damage state of reinforced concrete bridge columns after vehicle collisions. To achieve this, a numerical model of the columns was constructed in LS-DYNA to realistically simulate their lateral impact response through the calibration of concrete and steel material models under impact loads with the experimental results of column specimens available in the literature. The developed numerical model was then used to simulate vehicle collisions with full-scale bridge column, enabling a comprehensive analysis of column damage across diverse impact scenarios. Using design and vehicle parameters (input) used in the scenarios and post-collision damage state of the columns (output), five classification-based machine-learning models were developed. Among these models, the extreme gradient boosting model with Bayesian optimization (accuracy of 92 %) was selected as the optimal machine learning model based on feature selection, normalization, data splitting (training versus test), data balancing, and hyperparameter tuning. Shapley additive explanations were implemented to offer insights into the contribution of each input variable to the final prediction. The analysis showed that the column diameter, vehicle velocity, and longitudinal reinforcement ratio, in order of influence, significantly impacted the column damage state.