The IMI Colloquium Report in January 8, 2025
2025.01.16
■Title : Anomaly Diagnosis AI on Waveforms for Infrastructure and Manufacturing
■Place : IMI Auditorium(W1-D-413) and Live streaming with Zoom
■Speaker: Dr. Akihiro Yamaguchi (Corporate R&D Center, Toshiba Corporation / Institute of Mathematics for Industry, Kyushu University)
■Attendance: 24(Students: 7; Staff: 16;Others:1)
At this colloquium, the speaker gave a lecture on the AI for abnormality diagnosis that automatically determines time-series waveform data collected from sensors with high accuracy, which is attracting attention in the fields of infrastructure and manufacturing. The goal is not simply to improve the accuracy of determination, but to achieve more reliable operation by diagnosing waveforms that are not black boxes and can be interpreted, and by comparing the basis for judgment by AI with the knowledge held by experts. This problem is important in a very wide range of fields, and its realization requires a high level of mathematical ingenuity.
In this presentation, the speaker introduced an original method that improves the Shapelets learning technique, an interpretable classification method, by adapting it to real-world problems. Specifically, the speaker reported that the proposed method solves problems such as, the small number of anomaly data compared to normal data, the difficulty of detecting unlearned anomalies, and the fact that the results depend on the timing of waveform data acquisition. The presentation also included concrete examples of judgment, allowing the audience to experience how mathematics plays an active role in the real world. It was impressive to see that the Q&A session was very active, and the academic staff and students were very interested in the contents of this colloquium.
