The charm of "Authentic Fried Rice" is introduced in "Part 1: The Fairy of Fried Rice Report on HONKAKUITAME-CHAHAN (fried rice)! Part 1: The Fried Rice Fairy's In-depth Report on the Appeal of HONKAKUITAME-CHAHAN (fried rice) Please take a look at this page as well!

In order to manufacture high quality products more efficiently, Nichirei Foods has been developing and introducing AI-based inspection systems since around 2017. Tsutsui of the Equipment Development Group, which is responsible for the design and introduction of all equipment in the plant, told us about the benefits of introducing AI into the production plant.

The ability of people to make instantaneous judgments and work well is so high that AI and machines have yet to catch up. However, it is not true that concentration can last forever, and there is a non-zero possibility of fluctuations in accuracy, such as when one is not in good physical condition. AI, on the other hand, is tireless, and it works at a constant quality level. By understanding the difference between the two and using AI well, we can reduce the burden on employees while producing high quality products.

However, introducing the equipment into a frozen food factory requires overcoming a high hurdle.

For example, the Funabashi Plant, the site of this study, manufactures a variety of rice products, including Nichirei Foods' flagship product, HONKAKUITAME-CHAHAN (fried rice). Fried rice produces steam containing oil, so the camera used to send image data to the AI needed to be equipped with various countermeasures.

After much trial and error, a removable transparent plastic cover was installed, which can be quickly replaced if it becomes foggy. In addition, a cool air conditioner was installed to circulate air to protect the camera from the heat. With these innovations, we found that we could continue to take high-precision photographs for a long time.

After overcoming the first step, the development of equipment to be introduced into the HONKAKUITAME-CHAHAN (fried rice) production line began in earnest in 2019. Teams were formed for each of the three inspection processes that place a heavy burden on employees, and studies began.

The first was to inspect whether or not the fried rice had been chewed into the crimped part of the product package.
The second was to inspect the green onion as an ingredient to see if any foreign substances were contained.
The last process is to check and remove burnt fried rice.

For many years, the process of checking and removing burnt fried rice was done by human eyes and hands.
With the new system, a camera is installed in place of the human eye. The AI recognizes burnt rice from the data captured by the camera and transmits the information to a device (arm) that removes the burnt rice.

Tsutsui recalls, "It was a long road until the AI was able to correctly identify burnt food."

The work to improve the AI's identification accuracy requires perseverance, which Tsutsui describes as just like a school teacher.

Nevertheless, Tsutsui patiently worked to improve accuracy. Over a period of six months, he trained the AI on several hundred images, adjusting the color of the photos to make it easier for the AI to distinguish burn marks. The efforts bore fruit, and the AI gradually became able to distinguish burn marks with a high degree of accuracy.

There was another requirement for an AI-based burn removal device. That is a function to check for uneven quantity of ingredients in fried rice.

We wanted to help maintain quality by adding an alert function that, like an inspector, would request the equipment to check the previous process if it found an unbalanced amount of ingredients.

Tsutsui pondered how to respond to this request. This is because, due to the nature of the dish, fried rice is a mixture of differently shaped ingredients such as rice grains, eggs, and roasted pork, making it difficult to instantly produce the proportions of the ingredients.

Tsutsui decided to create a new program that would instantly determine the percentage of area for each ingredient from the images and alert the customer if there were any abnormalities.

In addition, with the cooperation of the factory, he operated the line for one hour after the day's production was completed, just for verification purposes. We let the AI learn from the anomalies, for example, by running a test product with an unbalanced amount of ingredients. The time it took to process the data was rapidly reduced, and alerts could be sent out with a high degree of accuracy.

During the trial introduction, the communication between the camera, the device for judging burns, and the device for removing burns did not go smoothly, and the team had to work hard to make adjustments. The difficulties were overcome by working together with the rice line leaders, technical group members, and other members of the introduction team who knew the plant well.

In February 2024, the burn removal device began full-scale operation on the production line.
Tsutsui smiles, "It was really a laborious process, so by the time the learning process was complete, I had grown attached to the equipment as if it were my own child."

The story of the introduction of the burn removal device, as seen by an employee working at the Funabashi Plant, will be introduced in Part 3: AI Changes the Manufacturing Site Today.