08. April 2021, 11:00 Uhr | Tobias Schlichtmeier
In Germany, around twelve million tons of food end up in the trash every year. More than 30 percent of this is already destroyed during the production process. Fraunhofer IGCV wants to change that with the help of artificial intelligence.
Germany has committed to the United Nations goal of reducing food waste by 2030 by half. There is a great need for action, because up to twelve million tons of food end up in the garbage, along the entire value chain from the field to the plate. Around 52 percent of the waste is caused by private households, as a 2019 study by the Thünen Institute found out. However, the study also shows that around 30 percent of losses already occur in food production and processing. The remaining 18 percent is caused by wholesalers and retailers trade and out-of-home catering. In the »REIF« project, which stands for »Resource-Efficient, Economic and Intelligent Foodchain«, 30 partners, including the Fraunhofer Institute for Foundry, Composite and Processing Technology IGCV, are working on a long-term method to combat food waste. The focus here is on designing an AI ecosystem that involves stakeholders at all stages of the value chain.
With the help of the ecosystem, the researchers aim to reduce losses. After all, cheese, rolls, meat, and the like can be produced more efficiently with data-based algorithms. In addition, the researchers want to optimize sales and production planning, process and plant control using machine learning (ML) methods.
The causes of avoidable waste are manifold. They range from overproduction and fluctuations in raw material quality to visual requirements that the food does not meet. In the project, the partners focus on dairy products, meat, and bakery products. Losses occur in the products primarily because they are perishable goods.
»We are bringing AI into the entire value chain and into the area of production. For this, we adapt and select the appropriate algorithms depending on the use case«, says Patrick Zimmermann, a scientist at Fraunhofer IGCV. »We are investigating the planability and controllability of all areas in terms of their optimization potential - from production in agriculture to sales in the supermarket«.
The potentials, however, are very different, he says. Zimmermann illustrates this with the example of a meat mixer. »The temperature and duration of the mixing process influence the best-before date of meat products. By using AI algorithms to minimize the energy input through the mixing process, we can extend the best-before date and consequently optimize the selling time in the supermarket and reduce food losses«.
At the plant level, the highest food waste occurs during startup, as the optimal parameters must first be found, and thus waste is produced first. »For example, we are trying to use smart sensors and self-learning AI algorithms to perfect the foaming process in the production of cake layers on the very first try«, the researcher explains.
In the long term, the project partners want to establish an IT ecosystem and set up a virtual marketplace. Among other things, companies will be able to make their implemented AI algorithms available to all participants here in the future. Another goal is to network the data of all companies involved in the project to increase value creation in the complex network of the food industry.
Via the online marketplace, the project partners can exchange their data. Production companies can thus better manage their manufacturing processes by benefiting from sales forecasts generated based on purchases. Data collected by supermarkets is incorporated into the forecasts. By combining a wide range of factors such as customer behavior, inventory levels and best-before dates, supermarkets can introduce targeted dynamic price adjustments for specific products.
This ensures retailers maximize profits while reducing waste and overproduction. In this way, the entire supply chain benefits. The end customer would also benefit: In rainy weather, supermarkets could lower the price of barbecue meat early to prevent it from sitting on the shelves. Forecasting systems designed in this way could also be offered via the online platform. The project partners are currently in the design phase, and the first practical tests will start shortly.