Modeling and Predicting Deterioration of Concrete Bridge Elements Using Machine Learning

Prediction of bridge elements’ deterioration is an essential task in maintenance planning to preserve bridges in a functioning and safe state. Several stochastic and data-driven approaches are used in the literature to estimate deterioration of bridge elements. However, these approaches rely on expert judgments of deterioration parameters or subjective selection of deterioration factors. The objective of this paper is to predict health index (HI) of concrete bridge elements using machine learning methods based on National Bridge Inventory (NBI) and the National Bridge Elements (NBE) databases. To this end, entropy-based mutual information analysis is applied to evaluate the influence of different deterioration factors, such as daily traffic, location, and age on bridge elements deterioration. Moreover, several machine learning models are developed to identify the best method to predict bridge elements’ HI. Based on predictive performance metrics results, random forest method had the best performance in terms of mean absolute error, root mean square error, mean absolute percentage error, and coefficient of determination metrics for all the elements. The primary contributions that this research adds to the body of knowledge are (1) application of entropy-based mutual information to evaluate linear and nonlinear impact of different factors on bridge elements deterioration, and (2) development of new machine learning models to predict deterioration of bridges with various characteristics including age, daily traffic, and location. The models developed and presented herein are expected to support decision makers in identifying optimal time of interventions to minimize maintenance costs while maximizing bridge performance.