[5K3-OS-38b] OS-38

2:00 PM - 2:15 PM JST(5:00 AM - 5:15 AM UTC)

[5K3-OS-38b-01] Degradation Diagnosis based on Current Sensor Waveforms by Robust One-Class Learning Time-series Shapelets (ROCLTS)

〇Ken Ueno¹, Akihiro Yamaguchi¹, Jin Machida¹, Yuko Kobayashi², Misato Ishikawa², Hitoshi Kobayashi³, Ayaka Takemura³, Takahiro Kokubo³, Tetsuya Asayama³ (1. System AI R&D Dept., AI Digital R&D Center, Toshiba Corporation, 2. Mechanical Systems R&D Dept., Advanced Devices R&D Center, Toshiba Corporation, 3. Machinery and Equipment Development Dept., Production Innovation Technology Center, Toshiba Corporation)

Keywords:

Degradation Diagnosis,Machine Learning based-on Waveforms,Anomaly Detection,Condition-Based Maintenance (CBM),Current Waveform Analysis

Labor shortages have increased the need for more efficient maintenance and inspection of equipment in industrial plants and manufacturing facilities (e.g. picking robots). Conventional spectral analysis techniques such as FFT often fail to detect subtle precursor-level degradation when analyzing equipment conditions through current waveforms. This study proposes an early degradation detection method for timing belts of picking robots based on our proposed Robust One-Class Learning Time-series Shapelets (ROCLTS) algorithm, which can train a degradation model using only normal motor current waveform data. Experimental evaluations demonstrate that the proposed approach successfully identifies early-stage degradation patterns that FFT cannot capture. We show that ROCLTS successfully detects timing belt degradation, which would be especially useful for monitoring physical AI systems.