17. Inspection support system using images of tunnel walls

Railway tunnels include many that were constructed before World War II or during the period of Japan's rapid economic growth in the 1950's to 1970's. Currently, these tunnels are properly maintained through regular inspections by experienced engineers.
However, with a projected decrease in the number of engineers in the future, there is a growing need for automation and reduced reliance on specialized skills.

Given this situation, we have developed a technology that utilizes digital technology to reduce periodic inspection time and minimize the need for personnel.
We have developed a technology that uses AI to extract individual deteriorations from images of tunnel walls and identify the overall health of the tunnel and areas requiring special attention that should be inspected intensively (Figure 1).
In constructing this AI, we trained it using a database of deterioration data from railway tunnels across Japan. In addition to detecting cracks, the AI can also extract indicators necessary for assessing condition, such as rust stains, water leakage, water marks, and repaired areas, with over 90% accuracy. Additionally, we developed a portable projection device that can display identified areas requiring special attention in the tunnel wall (Figure 2).
The shape of the projected mesh can be corrected based on cross-sectional shape, and the mesh can be moved horizontally based on the distance traveled.

In simulated inspections, the amount of work done at desk was reduced to about 1/10, and the speed of on-site inspection was improved to approximately twice as fast, achieving reductions in inspection time and personnel requirements (Figure 3).