ABSTRACT
Approximately 70% of Japan’s railroad tunnels are over 60 years old. To ensure safety, regular inspections once every two years are obligated by law and soundness judgement is conducted by skilled inspectors. However, the soundness judgment is partly based on qualitative criteria of each inspector, and this could increase variability of judgements. In addition, it is expected to be more difficult to secure skilled inspectors in the future due to the declining birthrate. Therefore, we used deep learning to automatically judge the soundness of cut and cover tunnels for the purpose of improving efficiency of the inspections. In this study, the accuracy of automatic judgement verified by using the actual soundness judgement in the inspection and the accuracy was confirmed to be approximately 60% if the safe side judgement was treated as correct.
