Modern production plants are often so complex that the interrelationships can only be captured incompletely by classical modeling. Optimization potentials can then only be tapped with the help of data-supported machine learning methods. ML is therefore being used increasingly and with great success, for example to increase product quality, reduce the use of resources or avoid unplanned machine breakdowns through predictive maintenance.

The opportunities are matched by major challenges: There is a lack of experts who are equally at home in ML methods as well as production and automation technology. Reusable components for ML-based systems in the production environment are in short supply. There is no established procedure for large, heterogeneous project teams, and adaptation during operation to changing conditions (wear, properties of input materials, structural modifications to the process) must be guaranteed.

Process model plus continuous chain of software tools

"In the ML4P lighthouse project, six Fraunhofer Institutes are jointly developing a standardized process model and the associated tools for the use of ML in production to overcome the challenges," says Prof. Dr.-Ing. habil. Jürgen Beyerer, head of the Fraunhofer Institute for Optronics, Systems Engineering and Image Exploitation IOSB and ML4P project leader. "Our approach to AI engineering is oriented in many respects to systems engineering. The broad experience of the participating institutes enables us to simultaneously detail the basic method and develop the appropriate, continuous chain of interoperable software solutions".

The tools are used to systematically capture the relevant knowledge and data of a production plant, formalize it and prepare it for the use of the ML method spectrum. Furthermore, they can detect and evaluate existing optimization potentials, select the most suitable ML methods according to the specific application and use them profitably. "Using concrete processes and plants from the process and piece goods producing industry, we can directly check the practical suitability of our results", Beyerer continues. In addition to the Karlsruhe and Lemgo sites of the lead Fraunhofer IOSB, the Fraunhofer Institutes for Intelligent Analysis and Information Systems IAIS, for Factory Operation and Automation IFF, for Industrial Mathematics ITWM, for Materials Mechanics IWM and for Machine Tools and Forming Technology IWU are also involved.

"AI algorithms often make up only a fraction of the overall solution"

"As an important milestone of ML4P, we have now completed the associated process model and are pleased to present the short version to the public as a white paper within the framework of the Fraunhofer Solution Days," says Dr. Julius Pfrommer, research group leader at the Fraunhofer IOSB and team leader of the process model in the ML4P project. The detailed long version of the process model will be published as a reference book next year.

"Although the pure AI algorithms are of central importance for the use of ML in production, they often make up only a fraction of the overall solution," explains Pfrommer. Another important piece of the puzzle is provided by the process model, which can be used widely regardless of the specific application. It is divided into six phases with clearly defined results and uses two central documents or data structures that represent the current state of knowledge across all phases: the "machine learning pipeline diagram" and the "virtual process file".

Integrating data and expert knowledge

In addition, there is a role model that includes the disciplines, competencies and functions required in each phase. Pfrommer: "This comprehensively describes the path from the problem definition to the continuous operation of the ML-based system. In particular, it defines the knowledge management and interfaces required to enable scaling to large teams".

An important aspect is to specifically integrate the specifics and previous knowledge from the application domain, the researcher continues. "The expert knowledge from the engineering disciplines is a great treasure. You can't just put a neural network over existing models. Instead, a deep integration of existing tools from engineering disciplines with the AI procedures must be achieved. This is the only way to ensure that AI can do a good job even in those areas where it has little or no data and experience from the past".

The ML4P White Paper will be presented to the public for the first time on October 26 at the booth of Fraunhofer IOSB at the "Fraunhofer Solution Days". This virtual trade fair brings together numerous exhibits and specialist presentations on innovations and application-oriented research results from the broad spectrum of the Fraunhofer Institutes on four theme days. Participation is free of charge, registration and further information at www.fraunhofer.de/solutiondays.

The white paper will then be available for download at www.iosb.fraunhofer.de/ml4p. Editorial offices can also receive it in advance for internal use - simply send an e-mail to presse@iosb.fraunhofer.de.