Pepsi’s Project Bonsai uses computer vision to keep a watch over Cheetos attributes.

Pepsi (NASDAQ: PEP) goes to great lengths to ensure that its highly popular Cheetos Snacks always have the optimal crunch, lightness, and shape. Consistency of flavor, texture, and color is achieved through careful monitoring of the extruders that produce the delicious snack. Engineers need to constantly monitor variables such as the speed of the snack cutting tool, the ratio of cornmeal to water in the mix, snack lengths, bulk density, and other attributes. For the company, Cheetos is a billion-dollar brand that is produced in 22 countries in more than 50 flavors. No surprises then, that Pepsi decided to deploy AI for quality control of Cheetos’ production. (Microsoft)

Project Bonsai

PepsiCo developed an AI solution powered by Microsoft Project Bonsai (NASDAQ: MSFT) that would control the all-important, and highly complex, extruder machines. Microsoft’s Project Bonsai is a low code AI development platform for intelligent control systems.

Used on a pilot plant, the Project Bonsai Solution successfully adjusted the machine to obtain Cheetos of the right quality and consistency.

The solution uses computer vision and data inputs such as density and length of the output to automatically adjust the extruders.

“This is the future for process controls,” says Sean Eichenlaub, a senior principal engineer at PepsiCo. “We’re using AI-based automation to improve the consistency of our products.”

A consistent production process delivers high quality and maximizes output.

It avoids the wastage of resources, time, and money because a product that is not up to the mark is unsaleable.

Another advantage: the AI system monitors the production continuously – a huge improvement over the current practice of checking the product manually at predefined intervals.

In the manual system, the production could be out of line for a while until the next sampling comes around.

Training the AI

It was too risky to train the algorithm on a live Cheetos production line on considerations of plant safety as well as scheduling.

Project Bonsai, therefore, used a simulated environment to train the algorithm. Plant operators coordinated with the system developers to replicate as far as possible the conditions seen in real life plant production.

Additionally, the developers used reinforced learning to teach the algorithm what to do, and what not to do, during the simulation.

The developers also built-in rewards for the platform when it achieved the targeted Cheetos qualities.

The simulation allowed the developers to save substantial time while training the algorithm. The solution could simulate a day’s run in just 30 seconds. By using the Azure cloud, the engineers could also implement multiple simulations simultaneously.

Outlook

Pepsi now intends to use the Project Bonsai solution on a live Cheetos production line.

Further, at some stage, the company will allow the AI to run the production autonomously. Hopefully, the rewards will be consistent quality, more throughput, and less wastage.

Significantly, Pepsi is marrying its expertise in food manufacturing with newfound AI capabilities. “Now that we’ve seen what Project Bonsai can do with one product, we are eager to find ways to improve processes across the entire company and product line,” says Eichenlaub.

Related Story:   UC Davis Awarded $20M Grant For Researching Next-Gen Food Systems