Current & Past Projects

ToVeBiMi

Multimodal deep learning toolkit for cross-application improvement of image quality in microscopy and optical inspection.
More information to follow.

AI for Sustainable Plastics – Enabling Circularity

Stricter requirements under various regulations, such as the Commercial Waste Ordinance (GewAbfV), the Packaging Act (VerpackG), or the End-of-Life Vehicles Ordinance (AltfahrzeugV), aim to reduce the amount of waste sent to incineration and increase material recycling rates. For plastic packaging, which currently accounts for just under 60% of the total volume of municipal solid waste, the new Packaging Act (VerpackG) mandates a material recycling rate of at least 55% by 2030. However, as things stand today, 53% of lightweight packaging (LVP) waste is sent to waste-to-energy plants or used as substitute fuel in cement plants for economic reasons. Recovered, sorted plastic bales are often defective. The resulting low-quality, gray, odorous regranulate is primarily used in the construction and agricultural sectors. The specific goal of KI-Sustain is to use artificial intelligence (AI) to increase the usability of plastic recyclates from municipal and commercial waste by optimizing material purity, thereby closing loops in the plastics industry and increasing recycling rates.

SkalCon - Improved scalability of condition monitoring across different product variants through feature translation using generative AI for condition monitoring

Further information to follow.

INSIGHT - Intelligent AI segmentation and inference of healthy and pathological tissue and single-cell images

More information to follow.

PREBAT - Predictive electrode evaluation for sustainable battery cell manufacturing

More information to follow.

Human-Centered AI System

As part of the project, a Human-Centered AI Workstation (HCKIS) is being developed that utilizes various levels of abstraction in artificial intelligence. The goal is to enable intelligent and interactive collaboration between humans and smart, networked machines at various levels. Through the targeted use of artificial intelligence (AI), a product is to be manufactured through interaction between AI, robots, and humans. In this context, the AI is not only intended to act autonomously but also to be able to respond flexibly to human interactions.

The central goal of the project is to intensify cooperation in the field of networked artificial intelligence while simultaneously creating interdisciplinary connections within a process. The AI fulfills several tasks in this context. An LLM model functions as an independent agent that intervenes as needed and can control the systems connected to the workstation.

Specifically, as part of the development of a joint technology platform, a scenario will be created in which new, innovative ideas for the future of work will be tested. The 360° AI Workplace serves as a demonstration platform for interrelated AI methods, particularly in the area of real-time process analysis and quality assurance. Here, “360°” means that humans are fully assisted by AI systems during a manufacturing order. The aim is to test innovative concepts that enable more precise, efficient, and adaptive control of production and manufacturing processes. In this way, the project not only makes an important contribution to the further development of smart production systems but also opens up new perspectives for human-machine interaction in an increasingly digitized working world.

HCKIS 2 One-Pager359 KBPDF

AI for Sustainable Plastics – Enabling Circularity

Stricter requirements under various regulations, such as the Commercial Waste Ordinance (GewAbfV), the Packaging Act (VerpackG), or the End-of-Life Vehicles Ordinance (AltfahrzeugV), aim to reduce the amount of waste sent to incineration and increase material recycling rates. For plastic packaging, which currently accounts for just under 60% of the total volume of municipal solid waste, the new Packaging Act (VerpackG) mandates a material recycling rate of at least 55% by 2030. However, as things stand today, 53% of lightweight packaging (LVP) waste is sent to waste-to-energy plants or used as substitute fuel in cement plants for economic reasons. Recovered, sorted plastic bales are often defective. The resulting low-quality, gray, odorous regranulate is primarily used in the construction and agricultural sectors. The specific goal of KI-Sustain is to use artificial intelligence (AI) to increase the usability of plastic recyclates from municipal and commercial waste by optimizing material purity, thereby closing loops in the plastics industry and increasing recycling rates.

KI-Sustain One-Pager300 KBPDF

Use of AI in Planning, Budgeting, and Forecasting

The goal of the SKIPS project is to integrate artificial intelligence (AI)—specifically machine learning (ML) and neural networks (NN)—into the controlling processes of small and medium-sized enterprises (SMEs). By utilizing external data sources such as market trends and economic data, the accuracy of forecasts is expected to improve significantly. This innovative approach enables the efficient use of internal data and the identification of complex relationships. This increases the efficiency of planning and budgeting processes, leading to time and cost savings and sustainably improving decision-making in SMEs.

SKIPS One-Pager369 KBPDF

Interpretable Anomaly Detection for Sustainability in Energy, Water, and Circular Economy

The early identification of unusual operating patterns in the energy, water, and circular economy sectors enables the maintenance of sustainable and environmentally friendly operations of technical facilities. In practice, the selection of an AI model is not based solely on performance metrics, but also takes into account the interpretability and, consequently, the implicit trustworthiness of the models. The project’s goal is to develop AI methods for anomaly detection in industrial time series that are not only powerful but also ensure interpretability.

FiAZ 2.0 One-Pager443 KBPDF

Deep Learning Super Resolution for Improved Imaging of the Finest Structures in Industrial Quality Assurance

Technically, the method is to be implemented using two modules. The first is an adaptation of a deep learning-based super-resolution network for materials microscopy, and the second—depending on the specific application—is a specific convolutional neural network for segmentation or defect detection in materials. First, the low-resolution microscopy image is upscaled by a selected scaling factor (e.g., factor 4) using the super-resolution network. For example, when determining grain boundaries in copper, this makes fine, poorly resolved grain boundaries more distinct, which simplifies their identification.

SuReQA One-Pager369 KBPDF

General Multimodal Inspection Using a Generative Pre-Trained Transformer

In industrial applications, traditional computer vision systems are often the only viable solution due to a lack of training data. However, developing them is time-consuming and costly, as it requires deep domain expertise and numerous communication loops between programmers and domain experts. At the same time, the demand for flexible, visually customizable image processing solutions is growing—especially among small and medium-sized enterprises. AMI aims to bridge this gap: Using multimodal AI that combines text and images, it automatically generates understandable and editable image processing workflows from just a few examples. This makes it possible for the first time to efficiently solve complex visual inspection tasks and adapt them flexibly without any programming knowledge—enabling accessible, explainable, and certifiable industrial image processing.

AMI One-Pager274 KBPDF

Development of Self-Learning Systems in an Industrial Setting

Machine learning (ML) systems used in industrial applications for quality control are typically tailored to the specific application. This means that the training data must be collected specifically for the given problem and, in particular, carefully labeled. This process is both time-consuming and costly. The goal is to create a system that incorporates algorithms and models to simplify this process. The aim is to enable recurring, similar, and complex analysis tasks to be handled automatically with as little effort as possible. By detecting anomalies, outliers, or deviations, the process is improved, as the labeling of training data can be made significantly easier.

SeSy One-Pager1.00 MBPNG

Transferable Framework for Interpretable Anomaly Detection in Industrial Time Series Data

In industrial applications, the selection of an ML model is based not only on performance metrics but also on the interpretability, traceability, and—implicitly—the trustworthiness of the models. The goal is to develop AI methods for anomaly detection in industrial time series that are more powerful yet still interpretable.

FiAZ One-Pager346 KBPNG

AI-Assisted Process Stabilization for 3D Printing and Polishing

The goal of the KIUPRO project is to use AI methods—specifically machine learning—to stabilize both 3D plastic printing and the computer-controlled polishing process based on process sensor data, thereby achieving higher component quality. The result will be a machine learning model capable of predicting quality parameters for various processes based on sensor data.

KIUPRO One-Pager551 KBPNG

AI-Supported Marketplace for Regionally Generated Renewable Electricity Based on Dynamic Pricing Models

The goal of this research project is to develop and implement an integrated platform that uses AI to forecast consumption and generation data, thereby optimally balancing regional/local demand and regional/local supply of electricity from renewable energy sources on a marketplace through dynamic pricing models.

One-Pager: AI-REN Marketplace569 KBPNG

AI-Supported Process Optimization for the Recycling of Post-Industrial Waste

The goal of the project is to use artificial intelligence (AI) to enable the recycling of post-industrial waste into new components in a way that ensures the quality requirements for a new product are met.

One-Pager AI ProWitA431 KBPNG

Improving detection accuracy by expanding state classes for transient airborne and structure-borne sound signals using artificially generated process data.

The goal of the TRANGAN project is to develop a generally applicable method for artificial data generation. This method is intended to enable the creation of artificial transient process data. The impact of this approach lies in the further development of known methods for GAN-based data generation—from quasi-stationary signals to the generation of transient signals. This significantly simplifies the development of AI-based condition monitoring systems that require high classification accuracy and shortens development time by reducing the number of tests required.

TRANGAN One-Pager412 KBPNG

Human-Centered AI System

The goal is to establish a Human-Centered AI System [HCKIS] that enables smart, networked machines to interact with humans at various levels using AI systems. As part of the project, a joint technology platform will be established to demonstrate, in particular, methods for real-time process analysis and quality assurance using artificial intelligence.

One-Pager HCKIS389 KBPNG