[Application Case] Jointly developed with Toyota Kyushu, a “Defect Prediction Detection System” where humans and AI collaborate to pass on the skills of skilled workers

2021/6/9

TRIART Co., Ltd. (Headquarters: Iizuka City, Fukuoka Prefecture, CEO: Kentaro Imazu, hereinafter TRIART) and Toyota Motor Kyushu Co., Ltd. (Headquarters: Miyawaka City, Fukuoka Prefecture, President: Osamu Nagata, hereinafter Toyota Kyushu) have started a trial of the “Defect Prediction Detection System” that detects ultra-fine defects, which skilled workers intuitively discover, in advance through collaboration between humans and AI at the Lexus production line of Toyota Kyushu’s Miyata Plant.

This system detects very slight shape deviations and differences in steel sheet elongation of panels mass-produced in the “press process,” where pressure is applied to steel sheets to form panels, based on images from a thermal camera installed inside the press machine. Such differences are critical indicators that can cause cracks or fractures after molding and lead to future product defects. However, detection has so far relied heavily on the experience and intuition of skilled workers, making it a field where skill standardization is difficult.

TRIART used its proprietary technology, Composite AI ‘4CAS’*1, which has many achievements in sensibility information processing based on image data, to generate ① a master image representing the standard shape of the panel from thermal camera images, and ② calculate the differences (such as expansion and contraction areas) between the produced panel and the master image. This enabled the evaluation of every panel produced every 5 seconds. This allows for identifying patterns of “what kind of cases lead to the next defect,” enabling more efficient defect prediction detection with the precision of skilled workers. The reliability and processing power of this evaluation were dramatically improved by a unique workflow where the task of generating the image ① does not rely solely on AI learning but involves workers specifying rough shapes midway, then returning to AI computation.

Could not retrieve the image from the original site.

In AI-driven DX (Digital Transformation), “full automation by AI” is often imagined, but in tasks where workers already possess skills or where human judgment is clearly superior, leveraging human power and designing workflows that complement AI can accelerate solving many problems. This is especially effective in manufacturing sites with many types of parts and lines where applying systems to all is difficult. This approach relies on Composite AI ‘4CAS,’ which does not black-box AI processes but consists of a series of control AIs. The system jointly developed by Toyota Kyushu and TRIART is a prime example of humans and AI collaborating equally in the physical manufacturing field.

Could not retrieve the image from the original site.

Toyota Kyushu’s Miyata Plant, as the world’s leading Lexus production base, has assigned skilled workers to each process and has continuously maintained top product quality*2. The plant is not only a place to produce products but also a place to create production technologies. We hope that new production technologies like this will create room for skilled workers to further hone their skills and sensibilities, becoming a foundation for offering more attractive products.

*1 Composite AI = Composite AI. It has a proven track record in sensibility information processing that is difficult to quantify, such as judgments made by human senses, and is also implemented in the difference comparison function of TRIART’s commercially available “Image and Drawing Comparison System MIIDEL.”

*2 The Miyata Plant is widely known as one of the world’s top automotive production sites, having won the highest Platinum Award five times in the factory category of the IQS (Initial Quality Study) conducted by the U.S. market research company J.D. Power, tying the world record.

This was also featured in the Japan Economic Newspaper.