Using Convolutional Neural Network (CNN) for Automated Defect Classification in Tunnel Lining Inspections - NAT2024

Tunnels and underground infrastructures assets, like all assets, deteriorate with time. At some point, they reach the end of their useful life. Therefore, frequent inspections, higher levels of maintenance, and rehabilitation are needed to address loss of life and safety concerns. The repair cost of tunnels will increase if maintenance of tunnels is not satisfactory and timely. Historically, tunnel inspections have primarily relied on visual and manual procedures. Due to their time-consuming nature and susceptibility to human error, there is a demand for alternative automated techniques that can enhance efficiency and reliability in tunnel inspection. By utilizing a tunnel scanning system, such as laser scanner and thermography, or photogrammetry system, tunnel inspection cost, time, and personnel can be reduced. The goal of this study is to utilize deep convolutional neural networks (CNNs) for automated identification of defects in tunnel lining inspection. This paper presents a framework to classify two types of defects in tunnel lining (water leakages and cracks) using CNNs. The CNNs were trained and tested using 4,608 images. The Precision, Recall, and F1 Score were each 99.6%, confirming the viability of this approach in the automated defect identification in tunnel lining.