Development and Applicability Assessment of a Tunnel Face Monitoring System Against Tunnel Face Collapse

To avoid damage caused by rock collapse in tunnel faces, it is necessary to sound an alarm before a rock collapse occurs so that workers have time to evacuate from the tunnel face. However, the area around the tunnel face is a confined workspace where multiple workers and machines are mixed together, and visual monitoring by observers alone is not sufficient to ensure safety and time. To solve these problems, we have developed a system to identify cracks on the surface of shotcrete using deep learning based on semantic segmentation, a method of image analysis, to realize constant monitoring as a technology to support monitoring and judgment (Tunnel Face Monitoring System) (Uke et al. in ITA-AITES World Tunnel Congress, 2022 [1]). At mountain tunnel construction sites where this system is applied, work behaviors such as crack observing are likely to change depending on the work environment and surrounding conditions. Therefore, the observation of tunnel face was used as the target task to verify the effect of environmental changes on crack detection time. This paper reports on the AI construction method and the usefulness of this system, which was verified quantitatively and objectively in a demonstration case conducted with the budget of a project commissioned by the Ministry of Economy, Trade and Industry (METI) in 2021.