Added good product learning AI to the high-speed, high-precision inspection AI "Roxy AI"! It compensates for the blind spots of the high-speed, high-precision defect detection AI!

An AI for good product learning that compensates for the blind spots of high-speed, high-precision defect detection AI.

Roxy AI's proud defect detection AI accurately detects even very small defects and has contributed to the automation of inspections for many customers. However, the defect detection AI struggles with detecting defects that require a wide view of the image and unknown defects, which sometimes does not meet expectations. The newly added good product learning AI excels at determining defects that the defect detection AI struggles with. Since the good product learning AI only learns from good product data, it can be used in environments where defects do not occur. There are three AIs that support the defect detection AI, including the good product learning AI: - Good product learning AI: Detects widespread defects, unknown defects, and mistakes - Image generation AI: Generates high-quality training data for defect detection AI - Zero learning measurement AI: Achieves size measurement and count without learning By combining these, flexible and high-precision inspection automation can be realized. A one-month trial is also available. Please give it a try.

• Visual Inspection Equipment

As shown in the sample video, Roxy AI is highly accurate. Here, we introduce the concept of Roxy AI, which enables high-precision inspections.

- A simple AI focused on necessary functions It can only detect defects, identify types of defects, and detect unknown defects. It cannot measure the size of defects, nor does it have functions like OCR. High accuracy is achieved with a proprietary AI optimized for simple tasks. - Learning features of both normal and defective products By learning both normal and defective products together, it accurately understands the boundaries. Learning only with normal products is appealing to customers who do not have defective products. However, to make a decision on implementation, isn't it necessary to collect a certain number of defective products? If so, I would like to improve accuracy by making a bit more effort to collect defective products. This is Roxy's thinking. - A thorough process to eliminate inappropriate training data If there are errors in the training data, the AI will learn those errors as well. By detecting and organizing inappropriate data in various ways, we prevent the creation of a distorted AI. - Annotation that eliminates mistakes regardless of who does it Annotation involves simply clicking near the defect. Even if there is a slight deviation, it does not affect accuracy. No special technique is required. All that is needed is the experience of the inspector to accurately determine whether it is a defect or not.

• Visual Inspection Equipment

Roxy AI, which can be mastered even by non-engineers. See how it differs from other appearance inspection AIs.

Unless it is an extremely simple test, a perfect AI cannot be created in just one learning session, so it is very important to improve accuracy. Many AI inspection products create AI using one of the following methods: - Utilizing AI know-how to repeatedly tune parameters to improve accuracy - A simple one-click process for anyone. If the accuracy is poor, data is endlessly collected Just as studying from incorrect or ambiguous textbooks does not improve grades, the quality of learning data is critically important for improving AI accuracy. At Roxy AI, we enhance accuracy by organizing the learning data. Roxy AI has many unique features, such as: - Automatic extraction of potentially inappropriate data (during/after learning) - Highly efficient annotation that is error-proof for anyone - One-click deletion/replacement of inappropriate data - One-click augmentation of missing data - Extraction of inconsistencies between learning data and test data There are many other innovations as well. Please download the PDF to see the details.

• Other inspection equipment and devices

Equipped with standard image generation AI, paving the way to zero defective products!

The improvement of AI performance requires a certain amount of training data. However, in products where defects are rare, there are challenges in sample collection. Roxy AI has equipped itself with an image generation AI that solves this problem! The feature-rich image generation AI lowers the barriers to sample collection. ■ Roxy AI Image Generation AI Features - Reproduces various defects You can freely specify the occurrence location, shape, and type of defect to generate realistic defect images. Defects in areas that are unlikely to occur in reality can also be easily created. - Thoroughly eliminates false positives Generates normal images with a texture that is indistinguishable from real ones. By learning more normal images, it thoroughly reduces false positives. - Capable of learning across different products Allows for collective learning of defect images that occurred in different products. It is also possible to repurpose the trained image generation AI for other products. We are confident in the quality of the generated images. Please check its capabilities with sample images. We encourage you to develop your company's original image generation AI. It will surely be effective in improving the accuracy of visual inspections and reducing the burden of sample collection.

• Visual Inspection Equipment

The phrase "high precision," which is used casually, has a solid reason behind Roxy AI's high precision.

There is an explanatory video, so please take a look. The phrase "high precision" is often used casually. Ask various companies: What is the reason for your high precision? Roxy AI achieves high precision with three key commitments. Without any one of them, precision cannot be achieved. ■ A simple AI focused on necessary functions It can only detect defects and identify defect types. It cannot fill in defective parts or read text. Even so, it can meet the needs of many workplaces. Because the tasks are simple, it not only makes high precision easier to achieve, but also offers many advantages such as easy annotation and fast learning. ■ Learning the characteristics of both normal and defective products together No need for defective samples. This is indeed an attractive phrase. However, many people may feel there are limits to learning from normal products. By properly learning the boundary between normal and defective products, we aim for high precision. ■ A thorough process to eliminate inappropriate data An AI trained on incorrect data is useless. By organizing the training data, we create conditions that make it easier for the AI to learn. There are many features available for this purpose.

• Image Processing Software

CLT, which has been widely adopted in the construction field, is now expanding its applications into the civil engineering sector. The "Hyper CLT Mats" are civil engineering CLT products that utilize wood CLT as temporary materials. Traditionally, steel plates have been the mainstream choice for temporary roads, access routes, and heavy machinery yards. The Hyper CLT Mats introduce a new option of "wood" to that market. The 5x10 mats weigh approximately 230 kg, making them lightweight and easy to install and remove with small machinery, with a maximum capacity of 40 mats on a 10-ton truck. They also have advantages in terms of workability and transportation efficiency. Moreover, they offer added value that can only be achieved with wood, such as glare reduction, noise control, CO2 fixation, and consideration for aesthetics. This product is expected to be a new proposal for not only construction companies but also building material firms, lumber companies, and material trading companies. It is a registered technology with the Ministry of Land, Infrastructure, Transport and Tourism under NETIS, with the registration number "TH-250021-A." The use of wood is not limited to construction anymore. The utilization of Hyper CLT Mats is expanding as a new market that connects civil engineering and wood.

Fukuda Corporation has been pursuing leak detection for 60 years and is dedicated to advancing the future of leak testing. 【 Supporting Various Leak Tests: Air Leak Test / Gas Leak Test (Hydrogen Gas, Helium Gas) 】 We develop, manufacture, and sell leak testing equipment to inspect airtightness and sealing performance. We will propose suitable testers and devices for each inspection target. 【 Target Industries and Examples of Inspection Targets 】 - Automotive Industry: Engines, FC components, valves and piping, various parts - Electronic Components Industry: Smartphones, keyless switches, various small sensors - Pharmaceuticals / Food / Cosmetics Industry: Bottle containers, syringes, vials, eye drop containers, pillow packaging, PTP packaging We will propose the most suitable leak testers for each industry. 【 ISO Certification 】JIS Q 9001:2015 (ISO 9001:2015)

This is a collection of links to Fukuda's technical columns. It covers various topics related to leak tests and more... Please check the related links for more information.

In civil engineering sites such as mountainous areas and construction sites, it may be necessary to use wooden boards on slopes or inclined access roads. Hyper CLT boards are civil engineering materials utilizing wood CLT, which can be laid not only on flat ground but also on sloped terrain. They can be used for temporary roads, access routes, heavy machinery entry paths, and measures against muddy conditions, creating a driving environment that takes advantage of the unique surface characteristics of wood. The 5×10 board weighs approximately 230 kg, and the 5×20 board weighs about 460 kg, making them lightweight and easy to install and remove even with small machinery, which is a notable feature for maneuverability in narrow sites. This wooden board is registered as a technology under the Ministry of Land, Infrastructure, Transport and Tourism's NETIS system as "TH-250021-A," and can also be utilized for safety measures and innovative ideas in public works.

Metal cracks are one of the concerns in equipment maintenance operations. Repairs using conventional welding methods can alter the base material around the crack due to heat, leading to residual stress and deformation, which always carries the risk of "crack recurrence" or "secondary damage." Hinode Waterworks Equipment Co., Ltd.'s "Mechanical Stitching Method" is a new technique that allows for the repair of metal cracks without applying heat. By using special bolts and reinforcement plates, the crack is physically removed without affecting the base material with heat. Repairs can be performed even in environments where open flames cannot be used. Since no heat is applied, the disassembly of equipment is minimized. For more details, please contact us or download the catalog.