- Anomaly detection
Time-series waveform anomaly detection technology*
Detect anomalies, learning based on normal time-series waveforms.
- Learning anomaly detection models and sub-waveform patterns (shapelets) by waveform-based machine learning.
- Currently applying the method to diagnose elevators, switchgears at substations, and other infrastructure facilities.
- One-Class Learning Time-series Shapelets (OCLTS)
Applications
- Anomaly detection for Gas Insulated Switchgear (GIS) at substation based on monitoring waveforms.
- Anomaly detection in elevator doors and brakes based on monitoring sensor waveforms.
- Also available for various types of infrastructure facilities and industrial devices.
Benchmarks, strengths, and track record
- World’s first anomaly detection technology that learns sub-waveform patterns (shapelets) using only normal time series data
- In performance verification using benchmark data, confirmed a 9% increase in accuracy compared to conventional technologies.
Inquiries
Please include the title “Toshiba AI Technology Catalog: Time-series waveform anomaly detection technology” or the URL in the inquiry text.
Please note that because this technology is currently the subject of R&D activities, immediate responses to inquiries may not be possible.
References:
- TOSHIBA SPINEX for Energy | Digital Transformation | Toshiba Energy Systems & Solutions Corporation (In Japanese)
- A. Yamaguchi et al, Applications of time-series waveform anomaly detection technologies in substation facilities diagnostics and improvements; The Database Society of Japan, Data Driven Studies, Vol. 3, Article No. 4, 2025. (In Japanese)
- A. Yamaguchi et al, Development of Advanced AI Technologies for Condition Diagnosis of High Voltage Switchgear in Substations. CIGRE Science & Engineering (CSE) Journal, 2022. (CIGRE Kyoto 2022 best paper)
- K. Ueno et al, Elevator Door Diagnosis with Low-cost Current Sensor based on Robust One-Class Learning Time-series Shapelets, Proc. of the 39th Annual Conference of the Japanese Society for Artificial Intelligence , 2025.
- A. Yamaguchi et al, Learning Evolvable Time-series Shapelets, IEEE International Conference on Data Engineering (ICDE2022), 2022.
- A. Yamaguchi et al, One-Class Learning Time-Series Shapelets, IEEE Big Data 2018.
- Press Release: - Develop highly explainable AI that detects abnormalities based on learning only from normal waveform data (In Japanese)
