Our digital colleague will sort it out: self-learning assistance systems in the packaging industry

This particular colleague lends a hand whenever there is a problem with a machine during the production process— in the event of disruptions, downtimes, capacity being overloaded or rejects being produced. In these situations, they quickly find out what is causing the fault interrupting production. In a matter of seconds, they evaluate all the data at their disposal—all the manuals, information and experience they have been fed by their colleagues. Their name is MADDOX, and they are an assistance system, programmed on AI-basis, for manufacturing and processing machinery used in the packaging industry.

MADDOX will step in whenever a disruption occurs. The algorithm then assesses whether this or a comparable fault situation has occurred before and quickly develops solutions to solve the problem for the machine operators’ benefit. This is particularly helpful for new, inexperienced colleagues, says Andre Schult, the CEO of Peerox, the firm that developed MADDOX. But by no means is it only rookies who stand to benefit: “The assistance system ensures that ramp-up times are accelerated, productivity is increased by reducing downtimes, and the skills shortage in the manufacturing sector is addressed,” says the 40-year-old, who has a mechanical engineering background. He and his business partner Markus Windisch founded the company in Dresden five years ago. Since then, knowledge management has been in high demand. There are now 19 members of staff working at the start-up, whose head office is located between Dresden University of Technology (TU Dresden) and software specialist Robotron.

A strong dose of dynamism is required with major customers such as pharmaceutical giant BAYER, wine supplier Rotkäppchen-Mumm, dairy company Sachsenmilch, and packaging businesses such as LoeschPack and Theegarten PacTec. They now rely on the “digital colleague”, as Schult calls it, in their or their customers’ production. And the new colleague does their job extremely well. This is because MADDOX ensures that the machine runtime for a standard system in round-the-clock operation increases by an annual average of 6%, by reducing downtimes due to disruptions and the duration of these downtimes accordingly: “With MADDOX, we are reducing both the frequency and duration of disruptions. The number is falling because we help to correct disruptions long-term by tackling their root cause, and the duration is falling because there is less of a need to spend long periods waiting for maintenance staff. Even with improvements of just a few minutes per disruption and fewer disruptions per day, the lower end of the efficiency enhancements we can deliver would be an average technical downtime reduction of 300 hours per year per system,” Schult calculates. The chief executive knows what he is talking about— after graduating from TU Dresden, he spent over ten years at the renowned Fraunhofer Institute for Process Engineering and Packaging (IVV).

It is one of Europe’s leading institutes for applied research in the fields of process engineering and packaging with a focus on food, packaging, product performance, processing machinery, and recycling and the environment. More than 300 researchers and other members of staff are employed at Fraunhofer IVV across its Dresden and Freising locations. The Institute is headed by Jens-Peter Majschak, under whose tutelage Schult worked. For many years, he worked closely with the head of the former Centre for Processing Machinery and Packaging Technology Applications—until setting up his own company in 2019 with Windisch, who studied electrical engineering at TU Dresden.

And these roots continue to shape the basic understanding around interaction between humans, learning and non-learning machines, and machine intelligence. After all, only the knowledge and experience gained by machine operators and then fed into the programming can be effectively implemented by systems such as MADDOX. As a result, people’s willingness to put their own knowledge into an automated system is a crucial factor in how successfully self-learning assistance systems can be implemented. This can be viewed from two angles. The first consists of operators, technicians, engineers and UX experts being ready to develop a digital colleague via a form of training on the job. The second involves individuals sharing their own knowledge without feeling afraid of their unknown digital colleague: “Fears, concerns and the question of whether this new colleague could potentially become a rival to someone’s own job are a psychological factor for employees when developing and rolling out intelligent assistance systems—and this factor should not be underestimated. At least half of whether a successful outcome is achieved comes down to this psychology. In this respect, we work closely with psychologists to instil trust in machine learning,” Schult says, speaking from experience. If this is done, he adds, it will increase staff commitment and motivation to contribute their own ideas and solutions to ongoing development efforts: “They become experts with broader responsibilities.”

And Peerox is also producing its own successes: After being named 2019 Start-Up of the Year, it won the 2021 DLG International FoodTec Award.