The researchers have developed a system to learn, through machine learning, differences in the hammering echoes of infrastructure, so as to automatically detect anomalies and to generate an anomaly map automatically.

AI hammer testing system to assist hammer testing with artificial intelligence

The waveform of an impact sound and information on where a hammer impact is applied are obtained by using acoustic and range sensors. Unusual sound analysis technology determines whether there is an anomaly. Upon detection, the anomaly is shown to the inspector in real time. An anomaly map is then generated upon completion of the hammer test. A normal sound model is trained by using impacts at sound locations before testing and is updated through online learning during the test. Therefore, testing can be performed even if sufficient training data are not available in advance.

With the aging of social infrastructure in recent years, the need for inspection is growing rapidly. Visual inspections and hammer testing are performed as primary inspections. However, these inspections rely on the experience and senses of inspectors, and the number of skilled inspectors is decreasing, making them difficult to find.

Demonstration tests on existing infrastructure will be performed to improve the system. Currently, hammer testing can be performed only on flat surfaces. The scope of application of the testing system will be extended. A product development organization for social implementation of the system will be set up in FY 2017.