Prof. Dr. Tilman Traub

The automation of manufacturing processes has been a central component of the further development of production technology since the 1960s at the latest. Current developments focus not only on the automation of processes, but increasingly also on the intelligent automation of Decision-Making Processes. Preliminary work on this project developed a model (figure) showing how conventional, non-intelligently automated processes can be further developed into (partially) autonomous, digitally supported processes. The availability of meaningful data is crucial for the success of such development approaches. Providing this data is a major challenge, especially in the field of special-purpose machines or small-batch production. As a result, activities such as set-up, process coordination or troubleshooting in these processes are still heavily dependent on the employees' wealth of experience.

The prerequisite(s) for the eligibility, suitability, aptitude of a process for intelligent automation is the provision of data access accordingly (level 1). To assess the eligibility, suitability, aptitude of an existing process, a guideline will be developed in the first part of the project to assess, rate existing data access points as well as possibilities to create additional measuring points. In the second part of the project, an exemplary process at Aalen University of Applied Sciences will be equipped with an intelligent data evaluation system and thus serve as an illustrative example as a starting point for joint discussions in teaching and research.

Project duration: January 2025- December 2025

Supported, sponsored by: Kessler + Co. foundation for education and culture

In recent years, increasing efforts have been made in the manufacturing industry to make members of staff (pl. staff), employee (pl. employees) experience describable and conservable through sensor technology and intelligent data evaluation. Nevertheless, despite great progress, process setup and the implementation of corrective measures must often be carried out by experienced employment. The aim of the "RoboForm" project was therefore to use a deep-drawing process to test the extent to which a humanoid robot can be trained to handle a forming process. In addition to comparatively simple tasks in series production, complex tasks such as tool maintenance, fault detection on components and the initiation of countermeasures were also considered step by step. As a result, the robot was able to perform basic worker activities such as inserting semi-finished products, operating the tool and press and removing finished components largely independently. At the same time, there was still a great need for development, particularly with regard to more complex activity profiles and drawing conclusions. With the help of a digital twin of the forming process, an AI model was trained to predict component defects. The results provide important insights into the possible skills and limitations of humanoid robots in manufacturing at the present time. They show the potential for the use of humanoid robots, but at the same time point to a considerable need for development and adaptation in order to reliably ensure autonomous processes.

Project duration: March 2025-September 2027

Supported, sponsored by: Invest BW - funding code BW8_1064

Digitalization and the use of artificial intelligence (AI) are fundamentally changing industrial production. While AI methods have long since found their way into high-volume production, small series and special processes often lack the necessary data volumes to train intelligent systems. This is precisely where the (IntA)²KS research project comes in.

Small series and special-purpose machines are the backbone of many medium-sized companies in Baden-Württemberg. However, their production requires extensive empirical knowledge, for example during set-up, process coordination or troubleshooting. As this knowledge is usually stored in the heads of experienced specialists and is not available in digital form, the increasing shortage of qualified professionals poses major challenges for companies.

The aim of the project is to develop assistance systems with interactive learning algorithms that specifically incorporate empirical knowledge into the training of AI models. In this way, even processes with limited data sets are to be digitally supported and employees are to be relieved of the burden of decision-making.

The project combines classic machine learning approaches with knowledge transfer from experts. Software prototypes are developed, trained and tested in real processes in laboratory / lab and industrial trials. The algorithms learn not only from data, but also from human experience.

Project duration: September 2025-August 2027

Supported, sponsored by: Carl Zeiss Foundation, funding reference P2024-17-050