Bridge Damage Prediction Using Deep Neural Network

Abstract

In bridge management practices, detecting damage and taking proper maintenance actions in a timely manner are significant issues. Due to the limited professional manpower and budget, providing a guide concerning potential problematic conditions is important to support on-site bridge inspections. The aim of this study was to estimate the number and severity of damage occurrences on a bridge deck using the Korean bridge management system (KOBMS). In this research, we considered identification, structural, inspection, and environmental factors and developed a deep neural network (DNN) model using 15,309 data, and we determined 36 influencing factors. The DNN model successfully predicted the number of damage occurrences on bridge decks and their severity with about 94.68% accuracy, confirmed by inserting external environmental data and span information. The findings emphasized the benefit of using machine learning algorithms when analyzing bridge conditions, and it showed potential for application to network-level decision making for preventive maintenance.