In this study, two methods of machine learning were investigated with the aim of using AI to identify the huge amount of output data obtained from ACM sensors. Specifically, the data obtained from the ACM sensors for implementation were analysed using supervised and unsupervised methods of machine learning. As a result, classification was carried out with high accuracy （95％） using supervised learning, but unsupervised learning, which is considered to be effective in actual operation, was considered to require further study because of low accuracy （60％）.

Hydrogen penetration into PC steel bars in utility poles occurs in narrow environments such as inside concrete cracks. However, hydrogen penetration phenomena are generally evaluated under a uniform environment, such as immersion tests in alkaline solutions. In this study, we constructed a system for evaluating hydrogen penetration into pure iron in an environment that simulates cracks in the concrete of a utility pole. It was found that a non-uniform hydrogen penetration phenomenon occurred in this narrow environment.