TOMOMI RESEARCH List of News Articles

The award ceremony for the "40th Kanagawa Industrial Technology Development Award," held on December 25, 2025, was featured on the official blog "Kuroiwa Diary" of Kanagawa Prefectural Governor Yuji Kuroiwa. Our company received the "Future Creation Award" for our 3D×AI visual inspection system "TR-300" at this award ceremony. We perceive our efforts as being recognized as part of Kanagawa Prefecture's industrial technology. Moving forward, we will implement our research results as technologies that create value in manufacturing sites, not just limited to proof of concept (PoC).

Our high-speed, continuous AI visual inspection system "TR-300" has received an Encouragement Award at the 38th "Small and Medium Enterprises Excellent New Technology and New Product Award" (organized by the Resona Foundation for Small and Medium Enterprises Promotion, co-organized by Nikkan Kogyo Shimbun). It was also featured in the Nikkan Kogyo Shimbun on April 15, 2026. The TR-300 is a visual inspection system that synchronously controls multiple line-type LED lights to capture fine surface irregularities in three dimensions and uses AI to determine defects. The inspection results are displayed as a heatmap. It can inspect glossy materials where defects are difficult to see under normal lighting without stopping the production line. ■ Intended applications: Steel plate forming processes, automotive parts, sheet and film work such as laminated films, textiles, and paper products. ■ Points of evaluation: Uniqueness of technology integration, practical level of completeness, and adaptability to challenges in manufacturing sites (labor shortages, variability in inspections, difficulty with glossy materials). The TR-300 has also received the "Future Creation Award" at the 40th Kanagawa Industrial Technology Development Award.

When there are many false alarms in AI visual inspection, the cause is not solely the algorithm. Our company has fixed the parameters of the AI anomaly detection engine "E3 ENGINE" and conducted verification by changing the input from normal camera images (LIVE) to 3D images using the 3D imaging technology "FORESIGHT STEREO," compiling the results into a white paper. ■ The same 128 samples (35 good products and 93 defective products), imaging device is TR-100 ■ The overlapping band of scores for good and defective products was reduced by 84% from Δ0.37 to Δ0.06 ■ False alarms at the optimal threshold were halved from 12 to 6 (while missed detections increased from 0 to 4) An example where two good products, indistinguishable by visual inspection, received scores that differed by three times due to differences in reflection illustrates "whether AI was seeing shape or appearance," supported by concrete data.

We have compiled a report on the results of imaging the steel plate sample entrusted to us by the customer using the 3D×AI visual inspection system "TR-300." The steel plate, covered with black scale patterns, has three linear scratches and two dents hidden beneath. When comparing the normal image (LIVE) obtained from a single imaging session with the extracted 3D Surface, we found that: - The contrast of the pattern decreased from 66.4% to 0.2% (since the pattern is color and not topography). - Shallow scratches that were almost invisible in the LIVE image appeared as two lines in the 3D Surface. On the other hand, the scratches where the base metal was exposed were clearer in the LIVE image. The TR-300's feature is that it allows for the verification of both color-visible scratches and shape-visible scratches with two aligned images. Comparison data and images of the five scratches can be viewed in the related catalog (imaging report).

TOMOMI RESEARCH Co., Ltd. released the synthetic NG data generation software "E3 Anomaly Synthetic Generator" on October 27, 2026. This Windows software creates pairs of defective (NG) images synthesized from good product images and texture images, along with ground truth masks that indicate the location of defects at the pixel level. ■ Even before mass production, evaluation data for inspection AI can be prepared without any defective products. ■ Defects can be selected from six types: linear, dents, foreign objects, chips, point-like, and lump-like, and can be synthesized by specifying their size. ■ Except for the defects and their surroundings, the original image matches at the pixel level. The standard price is 398,000 yen (excluding tax, one-time purchase for one license, including first-year maintenance). A 14-day free trial version is available. * The synthesized defects are not the same as actual defective products and do not guarantee detection rates or accuracy.