How SoftwareOne’s image recognition app helped Coca-Cola bottler identify 220,000 spare parts

Exploring image recognition for spare parts

Could the advances in image recognition technology, as popularized by Google Lens, transform the process? If you take a picture of the Eiffel Tower, Google will tell you what it is, where it is on Google Maps, the nearest Métro station, the opening times, and ticket prices. Could Coca-Cola HBC develop a tool that would recognize a spare part and tell engineers where to find it just by having a photograph of that part in your database?

“I admit I was skeptical,” says George Stamatiou, Digital Product Manager in Manufacturing at Coca-Cola HBC. “The commercial function had run an AI image recognition project on the way that we position bottles in our refrigerators and how you can then capture which bottles are missing. My initial thought was that this would never work for spare parts which are so different in terms of size and technical characteristics compared with a bottle. But finally, the technology and SoftwareOne proved that this is feasible.”

Rising to the challenge of managing spare parts inventory

To get an understanding of the vendor landscape, Coca-Cola HBC initiated a brief RFI and selected SoftwareOne’s Data and AI Center of Excellence to do an initial Proof of Concept. “These projects are also a risk for the vendor,” says George, “because the application has to be developed from scratch and you don’t know if it’s going to work.”

The first PoC demonstrated a high level of accuracy but with a small number of photographs, so the pilot phase was extended. On reaching 97 percent accuracy for the top match for 200 parts (this was where only 3% of the items were not found as the immediate first match, however, the AI solution would often find it as the second- and third-best matches and would always find the item in the top five best matches) Stamatiou and his team knew the SoftwareOne solution was viable and began to roll it out beyond the bottling plant used for the pilot.

Scaling the project

By the end of September 2024, 37 plants had been onboarded, with another 11 to follow in October. The five Coca-Cola HBC sites in Egypt will be added in 2025.

“The first task [of the onboarded plants] was to start gathering photographs of the spare parts to build a database,” says George.  “This is a huge effort. To give you an idea, one of our bigger plants stocks 4,000 parts and they have to be photographed a minimum of five times to capture all the angles and dimensions. Not a trivial undertaking at all.

“My concern that there would be serious resistance to this proved unfounded because the plants have made significant progress in building the database. Some plants even created their own shooting environment and set up a photo studio to get a better result. They also had ideas on how to improve the process and we take those on board as we add new features and functionalities to the app.

“It’s clear the engineers see the value in this and are confident that the application will reduce the time they need in a maintenance activity without opening books, opening systems, navigating the ERP, or more.”

How much time is saved and the value the app will release cannot become apparent until it is operational across all the plants. “We have an approach to this,” says George. “We have a KPI for the ‘equipment performance loss’ that reduces our production line efficiency. We have assumed that with this app we improve the KPI by 0.05 percent. On this basis, we get payback on the project in less than two years.”

Improving efficiencies on the production line

As the database grows, so will the production line efficiencies as delays and interruptions of 20 minutes or more are pared down to a minute with the aid of SoftwareOne’s image recognition app.

“It’s worked well,” says George, “working with SoftwareOne has been a great experience. It was a pleasure collaborating with the very experienced SoftwareOne consultants.

“It was quite a technologically advanced project, I would say. So the work that the team did with implementing this AI and Machine Learning tool, I think it was a really successful implementation in terms of embedding the latest technology trends into a really useful digital solution.”

Other divisions inside Coca-Cola HBC are looking at the SoftwareOne experience with interest.

“The Quality department wants to do a similar AI project for the image recognition of microbes I am not sure how advanced they are with this initiative and when they will be ready to have this discussion with SoftwareOne,” George concludes. “But based on our experience and the outcome of our AI project, I fully recommended SoftwareOne.”

Renato Luketin, Group Excellence Maintenance Manager at Coca-Cola HBC is excited by its results. “The app identifies the SAP code and tells you where in the company the spare part is stocked,” he says. “Before, you would have to email other sites on the off-chance that they had the part you needed, and if you didn’t hear from them, you’d have to source it externally. Time and money wasted.

“An unexpected benefit of the SoftwareOne solution is that it confirmed that our SAP coding across all the sites is remarkably accurate. Out of 20,000 parts captured the app identified just three inconsistencies, which we corrected immediately.

“The app saves us time and helps us minimize disruptions to production. During our peak season, from April to September in particular, the app plays a crucial role in maintaining operations. The joint effort with SoftwareOne is essential in securing the smooth running and efficiency of our production lines,” concludes Renato.