[2I4-GS-2] Machine learning: Anomaly detection

[2I4-GS-2-04] Failure factor detection in production process

〇Yosuke Otsubo¹, Naoya Otani¹, Megumi Chikasue¹, Masashi Sugiyama^2,3 (1. Nikon Corporation, R&D Division, 2. RIKEN Center for Advanced Intelligence Project, 3. The University of Tokyo, Graduate School of Frontier Sciences)

Keywords: Production process, Failure factor detection, Simulator

Precision mechanical products consist of lots of parts and assemblies. Data accumulation systems are usually installed to monitor the production process, but even if the process data is available, mechanical experts must manually analyze the data to identify the factor when the process has a defect. In this study, we propose an effective detection method for the defect factor based on the production process simulator and density ratio estimation. Numerical experiments clarify the effectiveness and limitations of the proposed method. Furthermore, the application to actual data consistent results with design information given as domain knowledge.