High Quality Material – Thanks to AI

Molding: The granulate stored under perfect conditions arrives in the injection molding machine.

Artificial intelligence was the key to success. But there was plenty of work ahead of the participants before they reached that stage. Process engineers in Schwalmstadt and a team of Freudenberg Sealing Technologies IT experts first had to learn to speak the same language. “Perhaps putting it a bit too simply, you can think of it as Analog 2.0 meets Industry 4.0,” Hermann said. “Two worlds came together in one project team.” Although communication was an obstacle at the start, the two groups were soon getting along better and better. To write an algorithm, it was important for the IT experts not only to understand certain plant processes – they had to completely immerse themselves in them. “We asked question after question to comprehend the problem and the process,” said Data Scientist Anna Lüders, a member of the in-house “Advanced Analytics Team.” It was her job to create the simulation model that enabled precise forecasts of moisture levels in TPU granules.

The Solution: A Digital Shadow  

A machine-learning model had to be developed to determine the TPU’s residual moisture. The model was based on the construction of a structured data transmission and storage system. Experts on a Freudenberg team devoted to MES and IIoT systems were responsible for the model. The database’s foundation was data that had been collected manually. To enrich the information with current data, the team installed sensors inside and outside the silos. They capture temperatures and atmospheric humidity in each case. Other parameters such as storage duration go into the database as well. The moisture content of the granules can be derived from these indicators. Manual measurements have been reduced. But they continue to be needed to compare real figures with the values from the algorithm that IT expert Lüders developed. It quickly became clear that the work had paid off. “Our samples show that AI forecasts of moisture content in the granules deviate by merely 0.005 percent from the results of the manual analytic measurement,” Hermann said. Certainly a negligible difference. 

By the way, “digital shadow” is the technical term for the kind of system set up in Schwalmstadt. It took the team a year to develop it from an idea to a successful implementation. Today, the moisture of the granules is calculated and visualized automatically. If limits are exceeded, the system triggers an alert so that the process technician on site can promptly intervene. This ensures that TPU granules with the proper quality reaches the production process – a fact that one major customer has been able to verify multiple times. Before production, the company always gets a sample from the batch it is due to receive. Since AI has been part of the monitoring process, the material has not been rejected a single time. “This has added to our customer’s confidence in us,” said Hermann, the project manager, “even though we have not changed the recipe for the material. With AI’s help, we are simply optimizing physical features.”