- Client: Griffith Foods
- Services Provided: Data Science & Machine Learning
Griffith Foods wins 10% of production capacity with AI-driven scheduling tool
Griffith Foods is an American product developer of customized food ingredients. We got in contact with the team at the production facility in Herentals, that asked us to develop an intelligent production sequencing tool to maximize productivity on the factory lines.
Our team developed a tool that optimizes the cleaning schedule of their production lines to regain production capacity. Thanks to the tool, the number of so-called wet cleanings (30 – 90 mins. standstill of the line) decreased by 5%. This way, 17 production days are won (in just one production facility) on an annual basis. In addition, they make their site greener by saving 1000 m³ of water per year and are using less detergents. All of this has contributed to a payback period of just 10 months.
An answer to our continuous growth
Griffith Foods produces all kinds of customized food products for their customers. Rising demand led to a search for innovative solutions and tools to boost production capacity. The cleaning of the production lines quickly came into view, says Oscar Sluiter, Senior Director Global Supply Chain at the company: “To guarantee the quality of each product and to ensure, for example, that there is no unwanted transfer of allergens, color, taste and other product properties take place, the lines have to be cleaned between the production batches. Absolutely necessary, but on average we lose about 165 production days per year, across all lines. the idea came to see if we could reduce the number of wet cleanings, because they are not needed between each batch. Sometimes a faster, dry cleaning is sufficient.”
The objective of this project was to minimize the cleaning time of the manufacturing equipment between production of different additives. If cleaning takes less time, then there is more time for production, and this way the company can increase their revenue. So far, planning the products on a production line was still quite some manual work, because planning is different every day. So a second objective was also to help the planners in doing their job in a more efficient way.
High-level problem statement
Not every product is the same. They differ in color, odor, texture and allergens. It’s important to take these differences into account during production because Griffith Foods wants to avoid that products get contaminated by other products in terms of color, allergens, … Since the beginning of production, Griffith Foods ensured the quality through a thorough wet cleaning between each product. This takes between 30 – 90 minutes, each time. And it’s not always necessary. If products are compatible then a dry cleaning is enough. Therefore, Griffith Foods asked Arinti to write an algorithm and create a tool that minimizes the number of wet cleanings based on the differences and equalities between the products. This can save a significant amount of time and effort per day!
“The collaboration with Arinti is going well and the tool is working fine. That translates to a roll-out across all our sites. Currently, two other factories in Europe are already planned for the next test. Ultimately, all our 21 locations are planned.” – Oscar Sluiter, Senior Director Global Supply Chain
It was decided to use algorithms to determine more accurately when cleaning with water and detergents was required. Arinti was called in to develop the algorithms, fine-tune them after several testing phases and develop the user interface. The Griffith Foods site in Herentals would become the home of this pilot project. “It was not an easy job,” Oscar Sluiter continues. “A number of iterations have been done to see what works best. We first started from a model that works based on allergens, odor, taste, color per recipe, among other things. Some of these parameters turned out to be difficult to quantify, so we switched to use the ingredient lists with data from our ERP system as a basis for the algorithms.”
Technical solution – Sorting optimization
An algorithm was written according to a logic provided by Griffith Foods. Basically, we matched products together based on their equalities. For example: if two products are yellow, contain small flakes and contain fish as an allergen, then they are compatible and can be produced after each other with a dry wash in between. However, if the next product doesn’t contain fish, then we have to schedule a wet wash between the product that contains fish, and the product that doesn’t contain fish. Basically, the algorithm looks at the combinations possible and starts building sequences of products that only need dry washes in between. Of course these sequences need to be glued together with wet washes. And that’s how we come to a daily plan for production on the different machines in the plant.
Algorithms save time, water and detergents
Choosing to focus on the ingredients turned out to be the right way forward. After training the models and developing the front-end, the planners at Griffith Foods received a simple, useful tool that allowed them to optimize cleanings. They also use to tool to assess the impact on the rest of the planning when an urgent order has to be squeezed between the normal schedule. “It used to take many hours to calculate all cleanings for a day without compromising the quality of our products,” explains Oscar Sluiter. “Now that happens in a few minutes, it is more accurate and the planner gets a clear suggestion. As a result, we already have five percent fewer wet cleanings and we want to increase that to at least ten percent. We have already gained back 17 production days per year with the current version of the tool. In addition, this already saves us 1000 m³ of water every year and we use less detergents, which is great for the environment. That was also an important parameter for a successful project. ”
Fluent collaboration leads to international roll-out
The references of Arinti convinced Griffith Foods to work with them. “Arinti did not just dump a lot of technology into our organisation. They listened to our demands and developed a tool that works for us. Our planners find the tool easy to use. In addition, there is an app that helps our employees control which batches are produced on which machines and that we can use to change dependencies ourselves when databases change location. That the collaboration with Arinti and the tool they developed works great, translates to further roll-out plans for all our sites. Currently there are two other factories within Europe on our schedule for the next test. We’ll use their findings to improve the logic of the algorithms and to increase the time savings possible even further. Ultimately, all our 21 locations around the globe will be using this tool, “concludes Sluiter.