Delivered and ongoing use cases across our core competencies — from industrial computer vision through metallurgical data analytics and predictive maintenance to intelligent robotics and bioacoustic monitoring. Each project combines scientific depth with industrial applicability, developed in close collaboration with industry partners and research institutions.
Manual verification and counting of heterogeneous packaging units in manufacturing and outbound logistics processes introduce inefficiencies and error potential. The objective was to establish an automated monitoring solution to ensure accuracy, traceability, and process reliability.
A vision-based AI-Framework was designed, developed, and evaluated to monitor packaging processes in real time. The solution combines object detection, classification, and body posture tracking, together with a tablet-based visualization, tailored to varying packaging units with automated counting mechanisms and integration into existing production workflows.
The AI-System enabled reliable, automated counting and monitoring, reducing manual efforts and error rates. Process transparency and consistency were improved, supporting more efficient and scalable outbound operations, all together with a strong conceptual focus in terms of transferability to other industrial packaging processes.
Fragmented data landscapes in metallurgical processes limit the effective use of production data for quality assessment and optimization. The objective is to establish a unified digital infrastructure as the foundation to apply advanced data analytics next to diverse use-case specific AI-driven process monitoring.
A scalable data infrastructure was implemented within the field of metallurgical processes, to integrate, process, and fuse heterogeneous data sources in order to facilitate a more structured and guided data acquisition procedure, all together with a proper user authentication and permission framework. On top of this foundation, data analytics and AI-Models were developed for strength analysis, hardness estimation, and various data visualization.
The project resulted in a unified, centralized and structured data environment supporting consistent data access and analysis. AI-driven insights improved material evaluation processes, while enhanced visualization increased transparency as well as decision-making efficiency.
Early detection of mechanical defects, such as bearing damage, is essential for ensuring system reliability and product quality. Ideally, defects should be identified as early as possible, starting with the quality assessment of supplied components, continuing during the production process, and extending to the early detection of failures at the customer site. Traditional inspection methods are often reactive and difficult to scale across these stages.
An acoustic-based anomaly detection system was developed using advanced signal processing and deep learning techniques applied to recordings acquired throughout production and assembly, as well as during operation at the customer site. Leveraging acoustic data from production test runs and field operation alike, the AI model reliably detects and reports anomalies irrespective of the underlying fault type, thereby identifying any deviation from normal operating behavior.
The developed solution enables reliable detection of acoustic anomalies across production and field operation, supporting early identification of abnormal mechanical behavior throughout the product lifecycle. This improves quality assurance, strengthens in-process monitoring, and allows failures to be recognized at an earlier stage at the customer site. As a result, the approach contributes to higher product quality, increased system reliability, and reduced reliance on reactive inspection methods.
Conventional robotic systems rely on manually programmed behaviors, limiting flexibility and adaptability in dynamic environments. The objective is to explore learning-based approaches enabling robots to autonomously acquire and generalize skills.
Current research centers on video-based imitation learning and deep reinforcement learning approaches. The models are being developed to learn task execution from demonstrations, through trial-and-error interaction, as well as via hybrid methods that integrate both paradigms. The aim is to achieve robust, transferable, and environment-independent behavior across a wide range of application scenarios.
The current research establishes a foundation for robotic systems that can acquire and generalize skills more autonomously, reducing dependence on manually engineered behaviors and static environments. By combining imitation learning and deep reinforcement learning, the approach supports improved adaptability, greater robustness across varying environments, and enhanced transferability to new tasks and scenarios. This contributes to more flexible and scalable robotic systems for dynamic real-world applications.
Accurate identification of bat species is essential for ecological monitoring, biodiversity assessment and conservation, yet manual classification of ultrasonic recordings is time-consuming and requires expert knowledge.
Ultrasonic audio data was processed and analyzed using deep learning techniques to derive species-specific data-driven acoustic features in order to build classification models being able to robustly distinguish across different ultrasonic call signatures of various bat species.
The developed deep learning models achieved robust bat species classification from ultrasonic recordings, showing strong performance across multiple species and demonstrating the ability to capture species-specific acoustic paradigms. The results indicate that AI-based analysis can support efficient and scalable acoustic wildlife monitoring.
Manual and fragmented operational processes limit efficiency, transparency, and the effective use of machine data in industrial environments. The objective was to digitalize workflows and enable structured, data-driven decision-making.
Industrial applications were developed and maintained within the given hardware/software ecosystem, focusing on workflow digitalization and system integration. This included the processing, analysis, and structured preparation of machine data, along with visualization components for operational insights.
The implemented solutions significantly improved process efficiency and transparency while enabling reliable access to and structured interpretation of machine data. This established a solid foundation for data-driven decision-making and scalable digital and AI-driven advancements.