How can medium-sized companies use AI in production to streamline processes and thus save time and costs? Using machine learning as an example, it is shown how the quality of individual production stages at PRINZ VERBINDUNGSELEMENTE GMBH can be predicted using AI and thus the economic sustainability of the company can be significantly improved.
The company PRINZ VERBINDUNGSELEMENTE from Plettenberg, founded in 1875, produces fasteners and formed parts using cold forming with around 180 employees (so-called “princes and princesses”). Whether in cars, window construction or household appliances, PRINZ GmbH’s current product range includes molded parts according to customer-specific drawings or according to standards, screws, coupling rods, pressed parts, bent parts, near-net-shape formed parts with and without complex further processing, such as machining or compression. In addition to developing new innovative products, such as pedals for bicycles, one of PRINZ GmbH’s challenges is improving the cost structure within the process for existing products.
For this reason, the presiding managing director Stephan Schwarz turned to the KI NRW Future Center to find out whether and to what extent an AI model trained on machine learning can help to optimize time- and cost-intensive production.
Marco Fries, Dr. Oliver Fix and Fatma Mendoza from the KI NRW Future Center during the factory tour at PRINZ GmbH in Plettenberg at the kick-off workshop
Approach and solution
During the 10-day intensive consultation of the “Zukunftszentrum KI NRW” project, the jointly defined goal was to investigate how well the AI can predict in a special machine during production for swaging wire into drill blanks whether “straightening” needs to be done in a downstream process and when not.
“The fact that we have a topic there that means people’s time and effort,
We were aware of that. There just wasn’t a solution yet.”Henrik Schwabe, AI project manager at PRINZ GmbH
Machine for upsetting drill blanks in the production of PRINZ GmbH
To ensure that the customer receives the desired quality of drill blanks, PRINZ GmbH takes test pieces from each batch and manually measures whether the compressed pieces are within the desired concentricity tolerance or not. If this is not the case, the entire batch is subsequently “straightened” in a downstream process on a special machine. This procedure is time-consuming and costly.
“It has often happened that people say, OK, we’ll fix it anyway. But it’s not just the work on the system, but also the working time of the employee in QA, because he spends half an hour on it every day. If we’re just talking about a month, then that’s 12 hours and 120 hours in the whole year.”
Henrik Schwabe, AI project manager at PRINZ GmbH
The task of the future center was initially to decide which parameters should be used to predict whether an AI model needs to be directed or not. In collaboration with the management and the established project team, it was decided to use the quality of the source material, the workplace and other data. Decisive focus was placed on the input material, the wire. This is delivered in so-called “coils” from two different suppliers and was divided into various parameters such as size, density, circumference, processing and surface, and based on the test and order data, the actual results achieved from the past 10 years were analyzed. (We deliberately avoided including other influencing factors such as weather, day of the week or employee quality in the analysis, simply to avoid the problem of handling sensitive data.)
“Coils” as input material in the production hall at PRINZ GmbH
Initially, it turned out that the existing data was initially not sufficient for the case described – a problem that we often encounter at the Future Center when introducing AI in the company.
“In the beginning, of course, the first hurdle was getting that much data. People thought that if we sent the (consultant) 30 to 40 data sets, then hopefully that would be enough to get a positive result somehow. We quickly realized that that wasn’t the case.”
Henrik Schwabe, AI project manager at PRINZ GmbH
As a result, an additional employee was needed to collect a sufficient number of past “cases” from various ERP data and thus enable the AI model to read them out. Only when the AI was able to analyze around 4,000 cases in which it was clearly visible which parameters of the wire condition had been straightened or not straightened in the past, could the quality of the AI model be assumed to be almost 90%.
Correlation of the parameters used in the AI model
The prediction goal was to predict when the condition of the wire would most likely result in rework through “straightening.” The AI was able to predict with different forecast probabilities, but the quality of the forecast depends on which forecast probability is sufficient to trust the AI’s prediction. As a result: If the forecast reliability has to be >90%, the AI can only do this for 38% of the data used. If the forecast reliability is reduced, the model can take more data into account. This trade-off must be determined by management.
Trade-off in forecast reliability
The focus of the development was also on a prototype, which can be further developed by PRINZ GmbH itself in order to create a solution that finds its way into real use. We used standard software for the development ourselves and “cast” the framework program that is responsible for data preparation into a Docker framework so that it is as easy to use as possible, following the “plug and play” approach. The PRINZ GmbH Future Center provides the necessary technical components to carry out the final touches, adapt the concept to other areas and use AI.
“Marco (advisor of the future center) has prepared it in such a way that even a layperson can use it once it has been explained properly.”
Henrik Schwabe, AI project manager at PRINZ GmbH
Live test and conclusion
It can be concluded that an AI that predicts when a certain production process is necessary and when not can represent an essential component in overall production planning. It is now up to the management and the responsible employees to decide to what extent the forecast can be trusted and the processes can be adjusted accordingly. The intensive advice provided by the Future Center has clearly proven that AI works and how it can be used. The forecast quality, which is the focus, can always be increased through a continuous supply of data, so that we can assume that PRINZ GmbH can continue to advance the process independently on the basis of our collaboration.
„We would like to bring this live into production. But there is still a long way to go because we first have to get the data out of the system automatically in order to be able to continue training the AI, and so far that is a manual effort that is simply not sustainable.”
Henrik Schwabe, AI project manager at PRINZ GmbH
Further steps
Due to the satisfactory results that were achieved during the intensive consultation, the company’s motivation is even greater to delve further into the topic of AI and to actively use AI as tested in the production process.
In addition, further measures are planned with the Future Center to impart further knowledge on the topic of AI and digitalization. We look forward to future collaboration and are excited to see how the project we were able to start together develops!
Back