As part of a doctoral fellowship funded by the Phoenix Contact Foundation, the TMDT chair is conducting research together with Phoenix Contact GmbH on AI-based optimization approaches in the production and assembly of printed circuit boards. Phoenix Contact is primarily active in the production of electronic connection elements and control cabinets in the industrial sector. A central research topic is the development and implementation of an AI-based anomaly detection in visual quality inspection. For this purpose, state-of-the-art Deep Learning methods are developed to automatically identify rarely occurring defects during assembly. The methods will be implemented and validated in the real production line.

The occurrence of water damage in the basement of a house and the associated costly repair and renovation work are a major annoyance for homeowners and residents. In order to be able to take countermeasures at an early stage and at comparatively low cost in the future, options are required for continuous monitoring of the condition of the insulation layer and corresponding early detection of moisture in cases of damage. With the aim of implementing a first proof-of-concept, the TMDT is conducting research together with Achim Wunderlich Bauunternehmung GmbH & Co. KG on the use of digitalization in the construction industry. The integration of IoT sensors during the construction phase is intended to enable continuous quality monitoring even after completion of a construction project. In the course of the project, moisture sensors will be integrated into basement elements built specifically for this purpose. The sensor values are then continuously monitored and evaluated using a custom-developed dashboard. In real experiments, different influencing factors, including the quality of the mortar and the position and size of the damage, are examined with regard to their effects on the sensor measurements.

In the production of windshields, cutting-edge solutions such as augmented reality head-up displays (AR-HUD) are placing new quality demands on the existing production. New technologies of digitalization as well as artificial intelligence promise great potential for process optimization. As part of the project, the TMDT is conducting research together with Saint-Gobain Sekurit and HotSprings GmbH on AI-based quality predictions in the manufacturing process. One of the main goals is to use collected process data and trained machine learning models to make inline quality predictions and to derive quality-improving and process-stabilizing action recommendations for the process.

Recommender systems are encountered every day in online stores, on streaming platforms, and on social media platforms. An essential functionality is to help customers find new and interesting offers according to their known but especially unknown preferences. In this project, the TMDT is conducting research together with the start-up Breinify (San Francisco, USA), the leading provider of software for predictive personalization at the customer level, on the next generation of Deep Learning-based recommender systems. In this context, the solution concept aims at responding to highly fluctuating variables and adapting recommendations to these changes. During the project, the developed concepts will be tested in production environments and used in real-world scenarios, thus generating important insights for further research as well as for the AI scientific community.

The manual reworking of rubber screens for screening machines in industrial production is associated with high time expenditure, physical strain, and health risks for employees. Especially with high quantities and a wide variety of variants, purely manual reworking reaches its economic and ergonomic limits. Against this background, TMDT is collaborating with Haver & Boecker OHG on the development of a camera-based, robot-assisted system for the automated post-processing of rubber screens based on computer vision. To this end, relevant screen areas are captured by a camera, deep learning models are used for visual fault detection, and robot trajectories are derived from the identified image areas. These trajectories are then automatically traversed by a collaborative robot with a suitable processing tool in order to precisely rework the affected areas.

The monitoring and maintenance of high-voltage overhead lines poses major challenges for manual inspection by human specialists due to the length of the lines and their free-hanging position. Automated AI-supported monitoring of overhead lines and image-based fault detection for maintenance cases represent an attractive alternative to manual inspection. Against this background, TMDT is conducting research into the development of methods for condition assessment and early fault detection of overhead lines through the use of AI-supported methods for the automated evaluation of visual inspection data recorded with camera drones for safe and efficient maintenance. In order to create the necessary conditions for this, an upstream data readiness radar project was carried out, in which TMDT assessed and evaluated the existing data stock. The quality, completeness, and usability of the existing data as well as the processes of data acquisition and storage were analyzed in order to identify existing gaps and derive targeted measures for the sustainable and structured further development of the database.

In modern warehouses, numerous pallet storage, retrieval, and relocation processes take place, which are usually carried out by automated robot systems. Depending on the type of pallet and its specific position in the warehouse, these processes can take varying amounts of time. The aim is to position pallets in such a way that the travel and processing times required by the robot are minimized and the overall efficiency of the warehouse is maximized. To this end, TMDT is collaborating with Ingstep GmbH to research solutions based on deep reinforcement learning in order to optimize these processes in an adaptive and data-driven manner, thereby minimizing the duration of retrieval operations. This is achieved, for example, by continuously carrying out relocation processes based on the current order situation, so that goods to be removed from storage are stored close to the removal location in anticipation of future removal processes. In this way, an intelligent warehouse agent permanently organizes and reorganizes the warehouse and flexibly adapts the allocation of storage space to external circumstances.

In order to remain competitive on the global market, manufacturing companies must make their production processes as efficient as possible. This efficiency is measured using various key performance indicators (KPIs), such as costs, throughput times, and adherence to delivery dates. However, modern manufacturing processes are often highly individualized and subject to specific technological and organizational requirements. The TMDT is therefore working with Bruckmann Steuerungstechnik (BSG) GmbH to investigate how common planning and control rules can be applied to specific use cases. The aim is to systematically analyze and evaluate their effects on the resulting production plan and on various KPIs. In this way, production plans are optimized and flexibly adjusted, taking into account the current order situation, machine utilization, personnel availability, inventory status, and raw material availability on the global market.