44 Informatik, System- und Elektrotechnik
Refine
Year of publication
Document Type
- conference proceeding (article) (171)
- Article (88)
- conference proceeding (presentation) (17)
- Part of a Book (11)
- Master's Thesis (11)
- Preprint (10)
- conference proceeding (summary, abstract) (6)
- Book (5)
- conference proceeding (volume) (5)
- Doctoral Thesis (4)
Language
- English (296)
- German (45)
- Multiple languages (2)
Publication reviewed
- peer review (236)
- nicht begutachtet (68)
- begutachtet (39)
Keywords
- Digitale Spiele (6)
- Fernwärme (6)
- Game Studies (6)
- Gesellschaft (6)
- Künstliche Intelligenz (6)
- Automated Driving System (5)
- ECG (5)
- Lastprognose (5)
- Maschinelles Lernen (5)
- deepDHC (5)
Institute
- Fakultät Elektrotechnik (116)
- Fakultät Informatik (100)
- Fakultät Maschinenbau (89)
- IFM - Institut für Fahrerassistenz und vernetzte Mobilität (69)
- IEAT - Institut für Energie- und Antriebstechnik (30)
- Promotionszentrum (30)
- IPI – Institut für Produktion und Informatik (29)
- IDF – Institut für Datenoptimierte Fertigung (19)
- IKR – Institut für angewandte Künstliche Intelligenz und Robotik (19)
- IMS - Institut für Maschinelles Sehen (14)
The present document was created as part of the European funding programme AI4Green and details the development of a numerical kinematic model for a DELTA-robot. The resulting model is subsequently employed to train a reinforcement-learning (RL) algorithm. By evaluating power consumption, the RL agent can be guided to exploit the workspace in a more energy-efficient manner.
This work aims to increase the energy efficiency of a delta kinematic robot using reinforcement learning (RL) by examining and optimizing the robot's movement sequences between variable positions. An RL environment was developed to train a model that is able to automatically generate an energy efficient trajectory between two random points in a 3D-workspace, while avoiding a static cubic obstacle with random position and size. Using a digital twin, the robot's limits and energy consumption were simulated during training. By balancing the reward function between the task of reaching the goal and minimizing energy consumption, a final AI model was created that generates trajectories using around 28% less energy than manually created ones.
In manufacturing, 3D CAD models define the geometry of a part while 2D technical drawings specify its manufacturing requirements, yet the two are rarely linked automatically. Associating drawing annotations with their corresponding geometric surfaces in the 3D model is a prerequisite for automated Product Manufacturing Information (PMI) extraction and remains unsolved. We formulate this as self-supervised CAD-face localization: given a point marker on a projected surface in a technical drawing, the model identifies the corresponding face in the 3D CAD model. In deployment, such markers are obtained from PMI arrow tips, making our approach a composable component of an end-to-end PMI transfer pipeline. A Graph Neural Network encodes 3D geometry while a CNN or vision transformer encodes 2D drawing crops, fused via cross-attention. Training uses synthetic data of varying complexity, requiring no manual annotation. Our best configuration achieves 98.04% accuracy on the face matching task, demonstrating successful cross-modal correspondence learning.
This volume constitutes the refereed proceedings of the 8th International Workshop on AI System Engineering: Math, Modelling and Software, AISys 2026 and the First International Workshop on Optimisation of Industrial Production with AI Algorithms, AI4IP, co-located with the 37th International Conference on Database and Expert Systems Applications, DEXA 2026, which took place in Graz, Austria, during August 11–13, 2026.
The 14 full papers were thoroughly reviewed and selected from a total of 29 submissions. They are organized in topical sections as follows: AI System Engineering: Math, Modelling and Software; and Optimization of Industrial Production with AI Algorithms.
A coupled nonlinear simulation framework for a drone carrying a payload suspended via rope is presented. The tethered configuration introduces pendulum-like dynamics coupled to the drone’s motion. The framework is intended to support the development of strategies for safe trajectory-following with collision avoidance.
The payload is modeled as a rigid pendulum with a holonomic length constraint enforced via Lagrange multipliers and stabilized using the Baumgarte method, ensuring numerical consistency while preserving physically correct rope forces. The quadrotor model is calibrated up to the attitude-control level, enabling direct specification of thrust magnitude and direction consistent with state-of-the-art autopilot interfaces, while retaining actuator limitations and operating close to the physical origin of the pendulum dynamics.
Simulations demonstrate that the model reproduces physically plausible system behavior, validating its suitability as a basis for control design. System identification experiments confirm frequency-dependent coupling between drone and payload, including tension loss phenomena and nonlinear interaction effects.
The presented framework establishes a foundation for future research into trajectory tracking in the presence of pendulum dynamics, fast and safe payload placement during swing motion, and integrated collision avoidance under simultaneous consideration of pendulum motion, drone trajectory, and actuator constraints.
Accurate road condition assessment requires the fusion of geometric and visual information from LiDAR and camera systems. In this work, we investigate the feasibility of multimodal registration between LiDAR-derived height and intensity maps and RGB image data for road surface analysis under real-world conditions.
Our experiments show that classical feature- and intensity-based registration methods are not suitable for this application. The homogeneous nature of road surfaces, the lack of stable cross-modal correspondences, and the limited structural quality of LiDAR intensity data prevent robust and reproducible alignment.
As a result, we rely on a time- and velocity-based alignment strategy that leverages sensor synchronization and vehicle motion. This approach provides a stable coarse registration but does not allow precise multimodal fusion and is sensitive to temporal offsets and dynamic measurement conditions.
These findings indicate that reliable multimodal fusion cannot currently be achieved using classical methods in real-world road scenarios. We therefore outline learning-based approaches, such as cross-modal feature learning and self-supervised methods, as promising directions for future work.
Grid-connected multi-string battery energy storage systems (BESS) operating under sustained high-power conditions face heat accumulation, power derating, and cooling-related efficiency losses, leading to trade-offs among thermal management, efficiency, and system availability. To manage these competing objectives, this work proposes a mixed-integer nonlinear programming (MINLP)-based multi-objective optimization (MOO) framework for optimal power allocation across parallel strings under air cooling. The optimization incorporates an equivalent circuit model (ECM) with SOC-, temperature-, and C-rate-dependent internal resistance, temperature-driven derating, and an experimentally derived inverter loss model. High-fidelity control is achieved by co-simulating the optimizer with an electro-thermal BESS simulation that captures spatial thermal dynamics of the battery pack. The simulation provides online state of charge (SOC) and temperature feedback and is validated against industry-measured battery pack temperatures, revealing core temperatures up to 4.5 °C higher than those predicted by conventional 0D average thermal models calibrated to pack surface measurements. Simulation results show that the baseline MINLP controller with evenly weighted objectives achieves approximately 3% higher round-trip efficiency (RTE) and system availability compared to the industrial benchmark droop control while minimizing thermal derating. Linearized optimal control without thermal optimization results in higher maximum string temperatures up to 40 °C and 4% higher derating losses. A Pareto-based objective prioritization improves energy performance by up to 5% while keeping temperatures below the 35 °C derating threshold. System scalability analysis shows that operating an optimal subset of strings outperforms conventional full-string operation, achieving 5.5% higher RTE with comparable availability and thermal performance. The framework is released open-source for adaptation to custom BESS applications.
Neben ihrem Unterhaltungswert, können Videospiele, Personen welche unter psychischen Belastungen leiden, als virtueller Rückzugsraum dienen. Einen Zugang zu Spielenden zu finden, welche in diese zur Vermeidung von Konfrontation ihrer Lebensumstände nutzen, kann mitunter eine Herausforderung darstellen.
Wirkungsvolle Kommunikation, sowohl im Umgang mit Mitmenschen als auch in der Psychotherapie, wird als ein einflussreicher Faktor auf dem Weg zur Verbesserung persönlicher Umstände gesehen. Die Formulierung einer Aussage kann maßgeblich beeinflussen, wie deren Inhalt wahrgenommen wird, hängt jedoch ebenfalls von den persönlichen Erfahrungen und Assoziationen eines Zuhörenden ab.
Um einen passenden Bezug zu Betroffenen aufzubauen, könnte ein auf persönliche Eigenschaften abgestimmter Kommunikationsstil einen Ansatz bieten, um Spielende möglichst wirkungsvoll zu erreichen und ggf. die Vermittlung von positiven Impulsen innerhalb des Spielgeschehens, langfristig auch über dieses hinaus zu verstärken.
Dies könnte sowohl zur Erleichterung der Kommunikation im Umgang mit Problemen mentaler Gesundheit und der Etablierung von Akzeptanz in Serious Games, als auch im allgemeinen Einsatz zur Verbesserung der Spielerfahrung und der Spielerbindung eingesetzt werden.
Im Rahmen der Arbeit wird ein Konzept vorgestellt und ein Prototyp für ein adaptives Dialogsystem entwickelt, welches Persönlichkeit mithilfe eines Large Language Modell (LLM) möglichst wissenschaftlich fundiert abbildet und Spiel-Dialoge in der Unity Game Engine abhängig des Spielerprofils umformuliert. Der gewählte Ansatz beruht auf psychologischen Grundlagen und Kommunikationsforschung, Erfahrungen aus dem Game-Design, sowie aktuellen Erkenntnissen im Umgang mit Intelligent Virtual Agents (IVAs) im Bereich der Human-Computer-Interaction (HCI). Zudem wird im Rahmen der Arbeit untersucht, inwiefern LLM zur Umformulierung von Dialogen im Spielkontext eingesetzt und in bestehende Prozesse der Spielentwicklung integriert werden können. Besonderes Augenmerk wird darauf gelegt ein universell anwendbares System zu entwickeln, welches durch die Umformulierung bestehender Inhalte nicht in die narrative Struktur und kreative Vision des Spielkonzeptes eingreift.
Many production lines require active control mechanisms, such as adaptive routing, worker reallocation, and rescheduling, to maintain optimal performance. However, designing these control systems is challenging for various reasons, and while reinforcement learning (RL) has shown promise in addressing these challenges, a standardized and general framework is still lacking. In this work, we introduce LineFlow, an extensible, open-source Python framework for simulating production lines of arbitrary complexity and training RL agents to control them. To demonstrate the capabilities and to validate the underlying theoretical assumptions of LineFlow, we formulate core subproblems of active line control in ways that facilitate mathematical analysis. For each problem, we provide optimal solutions for comparison. We benchmark state-of-the-art RL algorithms and show that the learned policies approach optimal performance in well-understood scenarios. However, for more complex, industrial-scale production lines, RL still faces significant challenges, highlighting the need for further research in areas such as reward shaping, curriculum learning, and hierarchical control.
Automated sheet metal handling in the automotive industry using robot manipulators is a standard in modern production. However, the desire of automotive companies to speed up the production process on the assembly line and at the same time to reduce expensive hardware components poses new challenges for robotics. Excessively rapid movement of flexible parts or sheet metal can either lead to its plastic deformation or increase the decay time of the vibrations to such an extent that it is necessary to wait before the part can be further processed, for example by welding. The traditional approach to solving these problems is to add fixing points for the part and/or to allow for waiting times at the end of the robot movement to ensure that the sheet metal vibration subsides. This means that more effort than necessary has to be put into the hardware setup or a poor cycle time has to be accepted. In our work, we propose to improve virtual commissioning systems by conducting a deeper analysis of the dynamics of sheet metal parts during its movement by the robot. To achieve this, a three-step optimization strategy is proposed. The first step is a structural transient analysis of the thin metal part under disturbances that arise during its movements. The second step is the creation of a substitute model, which is trained on the basis of the data obtained in the first phase and considerably reduces the computing time required compared to the time needed for a finite element method (FEM) simulation. Subsequently, this is used to optimize robotic handling efficiency by optimizing the robot trajectory. By addressing the challenges posed by sheet metal dynamics, enhanced process control, reduced cycle times, and ultimately, improved manufacturing outcomes are anticipated.
