How Japan's Sekisui House Automated Ceramic Wall Defect Inspections with VisionPro Deep Learning

Sekisui House uses Cognex deep learning to find defects on ceramic tile

Sekisui House Ltd. Is one of Japan’s largest house builders with yearly revenue close to $20 billion USD. The company, founded in 1960 and headquartered in Osaka, is known for its distinct Bellburn ceramic tile cladding made in its Shizuoka and Tohoku factories.

Bellburn ceramic tiles are used for outer wall siding in upscale home construction. They are not only attractive but durable, with self-cleaning properties, and represent the company’s philosophy of creating products “slow and smart.” The tiles are tempered and ceramic art techniques are incorporated in the production process. But just like in ceramic arts, there are sometimes small concaves or lines left in the exterior from the original molding process, which shapes the clay or other material. In order to eliminate cosmetic imperfections from the molding process it is vital to inspect every piece of tile.

Sekisui House 01

The volume of manual visual inspection, though, is impossible for a person to keep up with. It has also been difficult and time-consuming to formulate machine vision inspection rules to account for all the potential visual defects that could present during an automated inspection of the Bellburn tiles. Plus, not every imperfection occurring from external defects or color irregularity is cause for rejection. Some of the potential defects fall within the acceptable range to limit the risk of removing good product from production, which would impact the required volume of supply.

In order to ensure the required number of Bellburn tiles at the time of the final inspection, production plans had to be formulated based on past yield rates and inefficient manual inspection. This resulted in increased inventory, some of which became immobile stock.

To meet these inspection challenges, Sekisui House deployed VisionPro Deep Learning combined with a line-scanning camera and LED lighting to automate the cosmetic defect inspections throughout the production process. By automatically inspecting all pieces the homebuilder succeeded in improving overall quality in the manufacturing process, reduced surplus production, cut stock and costs, and stabilized the supply of Bellburn ceramic tile siding.

Deep Learning defect detection needs only a small amount of image data

VisionPro Deep Learning, a PC-based deep learning solution, identifies multifarious cosmetic defects  on the original plate based on a small number of sample good images. Generally, a massive amount of image data is necessary for an open source deep learning tool to understand acceptable flaws from unacceptable ones. However, VisionPro Deep Learning can do so from a sample set of about 100 images because its defect detection tool is purpose built for manufacturing.

After the firing step of the molding process, the line scanning camera and LED lighting captures external images for automated inspection. VisionPro Deep Learning  identifies defects such as bulges, depressions, color shades, or unwanted lines. Then, it analyzes these images allowing automation engineers to further refine the application based on acceptable or unacceptable cosmetic defects.

High processing speed for real-time detection on the production line

Sekisui House 02 

By detecting defects that occur during the molding process in real-time, faulty items can now be discovered at an earlier stage in the process. Quality management is done done more accurately in the step before the final inspection improving production planning accuracy. For example, differentiation of one 200 x 32 cm material takes approximately two seconds. External inspections at this speed on the production line is completely impossible with only human inspection. VisionPro Deep Learning inspects deformed spots selectively, allowing it to make distinctions at a high processing speed.

Success in reducing immobile stock

Introducing this inspection system resulted in a 40% reduction in immobile stock, which formerly accounted for 0.4% of the reserve production volume, leading to cost reductions.

Automation of the visual inspection through deep learning prior to the final process led to huge improvements in more accurate quality management.. It improved yields that makes accurate production planning possible and also allowing for a reduction in inventory.

Further cost reductions with horizontal expansion

After the success deployment at the Shizuoka factory, the deep learning defect detection system was also implemented at the Tohoku factory. In Tohoku the external inspection was incorporated after the clay material pressing stage. If defects are detected before the clay dries, the materials are recycled, eliminating waste. This led to successfully saving significantly on material costs.

Making stable supply a sure thing

The ability to detect defects in real-time through automation of visual inspection during production using VisionPro Deep Learning succeeded in reducing inventory and costs. Achieving the automation of external visual inspection of every tile piece using VisionPro Deep Learning made it possible to produce the exact number of ceramic plates required precisely when they are required.

Overall, the production process of Bellburn ceramic siding became more streamlined. With VisionPro Deep Learning automating these inspections Sekisui House was better able to achieve it’s motto of, “Creating freely designed residences one at a time, each custom-made from scratch to make the customer’s dream come true.”

James Furbush

More Posts on


Join MyCognex


Cognex representatives are available worldwide to support your vision and industrial barcode reading needs.

Contact Us