DEVELOPMENT OF A METHOD TO DETERMINE THE CAUSE OF CRACKS IN PAVEMENT USING AN EXPLAINABLE AI

Abstract

 In Japan, the time to repair and renew a vast stock of aging roads is approaching. Road administrators need to establish a pavement maintenance cycle that allows for efficient maintenance and repair in the face of difficulties such as fiscal constraints and a shortage of engineers. In this study, with the aim of providing a pavement diagnostic support tool for road administrators, we developed a method to determine the cause of cracks in pavement using an explainable AI (XAI) based on road surface images from vehicle-mounted cameras and basic data on pavement characteristics. As a result, we found that using a composite model incorporating a convolutional neural network and random forest improved the accuracy of determining the cause of cracks, and that it was possible to visualize the basis of the results. Furthermore, we applied this method to pavement over a wide area in Hokkaido and demonstrated that it is possible to map the causes of cracks along a route (project-level evaluation) and to analyze the regional characteristics of cracks (network-level evaluation).