Learn more about the use case:

Erium GmbH

Expert systems


Integration of expert knowledge in machine learning for the optimization of production processes

What drove the decision to develop an AI-based solution?
The increasing complexity of industrial processes often leads to suboptimal settings and conditions, which then result in high material consumption and high error and rework rates. Since these demanding processes can no longer be mastered by one person alone, companies must distribute a large amount of domain expertise and process knowledge across a large number of experts, engineers and technicians. This scattering of knowledge and expertise leads to difficult collaboration and, because of this, suboptimal production processes. The only solution to these problems is a collaboration between the human and the machine. Eriums software enables this partnership by graphically modeling the process knowledge and expertise of a company's employees so that it is understandable to the computer. The software user can then combine this knowledge with machine learning. This partnership leverages employee expertise with AI and enables not only a better understanding of processes, but also their optimal control!

What were your expectations or requirements for the AI solution?
The biggest challenge when applying AI in industry is data availability. Data collection is particularly expensive in industrial processes because it has to run in parallel with the profitable operation of the plant. In addition, production volumes are low in many areas, so the usual AI systems don't have enough data to train on. The new solution therefore had to be capable of calculating and predicting complex results despite less data.
Erium has addressed this difficulty by using expert knowledge and Bayesian methods. The knowledge of the process experts is used as a substitute for data and the information content is integrated by Erium's system so that the required amount of data is 10 to 50 times lower. In addition, Bayesian methods automatically provide an uncertainty estimate that allows the results to be evaluated with little data. Based on this estimate, the user can avoid collecting an unnecessary amount of data.

Which partner did you rely on with which technology for the development?
Erium's founders, Dr. Theo Steininger and Dr. Maksim Greiner, developed the basis for Erium's AI system during their doctoral research at the Max Planck Institute for Astrophysics. Their expertise in Data Science was then requested by BMW to optimize an assembly line using AI. It was in this context that Erium's approach could be tested for the first time. After the successful project, Erium designed its AI system as a product so that companies can quickly and easily apply it to their processes.

The use of AI solutions must also be accompanied by a change in competences. How have you dealt with this challenge in respect of your employees?
Erium's AI solution works as a tool to empower the employees of a company in their daily work so that they can optimally manage the company's processes with the help of AI. That is why our clients' employees are always strongly and actively involved in the development of the solution from the very beginning. This leads to a progressive introduction to the field of AI, a deeper understanding of the approach and an interest in these new tools of the future industry. In addition, Erium's software has an intuitive user interface that not only simplifies working with the system, but also facilitates collaboration between employees.
If necessary, Erium offers services such as consulting, seminars and practical workshops to introduce a company and its employees to the field of AI.

In what timeframe did your company implement the AI solution?
The development of the system's software base took place over several years during the doctoral research of Erium's founders. The first implementation at a client's site took several months, as it was preceded by a feasibility study.
Meanwhile, because the system only needs to be customized, Erium can now deliver a running prototype in less than 10 days. Implementation, in turn, takes insignificantly longer.

A conclusion to your AI solution:
Erium's solution is a system that enables industrial companies to quickly and easily combine their domain expertise with AI. The software enables the graphical modeling of experts' process knowledge so that it can then be used as the basis for powerful machine learning algorithms. This partnership between human expertise and artificial intelligence enables deep analysis and optimal control of complex industrial processes.

One final piece of advice/tip to other entrepreneurs looking to apply AI:
Setting concrete goals is extremely important for successful AI projects. In addition, data availability must be considered from the beginning, as poor or little data limits the possible solutions. Finally, smaller companies must learn to successfully apply stochastic methods (such as Bayesian methods), as these methods allow modeling using very little data.

More info at: