Point

We have developed a system capable of machine learning of a difference in hitting sounds from an infrastrucutre caused by an inspection hammer. The system can sense anomaly points and anomaly degrees automatically. It is also capable of automatic creation of an anomaly map.

“AI hammering test system” allowing artificial intelligence to assist in hammering test

Background

In recent years, social infrastructure has been getting older, and future demand for inspection is expected to increase rapidly. As a primary inspection, a visual check and a hammering test are performed. However, such inspection depends on the experience and sense of inspectors. Skilled inspectors tend to decrease in number, and it becomes difficult to secure inspectors.

Outcomes and Methods

A hitting sound waveform and hitting position information are simultaneously acquired using an acoustic sensor and a range scanner, and the presence or absence of an anomaly is automatically determined by an anomaly sound detection technology. When an anomaly point is detected, it is presented to an inspector in real time. Also, immediately after the end of the hammering test, an anomaly map is automatically generated. A normal hammering sound model is constructed by hitting normal points before work, and the normal model is updated by on-line learning during the inspection. As a result, even if a sufficient amount of data is not available, an inspection can be performed.

Future Plan

We will repeatedly perform a verification test on actual structures to enhance completeness. Although an inspection object is currently a flat surface, it will be expanded in the future. We will conduct research and development to improve usability toward practical use.