- Anomaly detection
- Numerical analysis
Automatic monitoring technology to detect decline in accuracy of AI image inspection on manufacturing lines
This new automatic monitoring technology increases the reliability of AI image inspection on the front lines of manufacturing.
- Toshiba is introducing AI image inspection systems using neural networks in the front lines of manufacturing.
- Inspection accuracy declines due to changes in manufacturing conditions and changes in equipment over time, so maintenance and management of inspection systems is essential.
- Detects declines in inspection accuracy by monitoring changes in Mahalanobis distance between learning images and inspection images in an intermediate layer, where image features are quantified.
Applications
- Image inspections in general, welding inspections, etc.
Benchmarks, strengths, and track record
- Provides an AI model deterioration detection method that is effective in the autonomous maintenance and management of AI inspection models.
- This method for detecting changes in the distribution of feature space enables highly accurate detection of deterioration in AI models with no “correct” labels, in comparison to traditional methods.
Inquiries
Inquiries to the Corporate Manufacturing Engineering Center of Toshiba Corporation
Please include the title “Toshiba AI Technology Catalog: Automatic monitoring technology to detect decline in accuracy of AI image inspection on manufacturing lines” 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:
- Miyuki Uchida, Taisuke Washitani: Study of model deterioration detection in AI image inspections; Information Processing Society of Japan, 84th National Conference Proceedings; Mar. 2022 (in Japanese).
- Miyuki Uchida: Monitoring Technique to Automatically Detect Deterioration in Accuracy of AI Visual Inspection in Production Lines, TOSHIBA REVIEW FRONTIERS OF RESEARCH & DEVELOPMENT, Vol. 77 No. 6, Nov., 2022.
