Subprojects

Motivation:

Sand casting is important for large aluminium components, complex shapes and internal structures requiring sand cores. Most foundries in this segment are small or medium-sized enterprises. Their innovation relies on machinery developments and, to a large extent, their own experience and expertise. As important transport-sector suppliers, they must respond to industrial trends, innovation, legislation and global economic and political developments.

Alongside recent economic and political challenges, the energy transition, supply-chain carbon footprints and AI in production will strongly influence business success. For medium-sized foundries, underdeveloped and often externally managed IT infrastructure is a major barrier. Resources and expertise for digital transformation, comprehensive data systems and AI-based process and quality optimisation are often limited.

Approach:

Kempten's subproject belongs to the sand casting cluster and develops AI and machine learning models, in-process traceability and optical quality inspection systems. Model-based decision support and AI methods improve efficiency, flexibility, resilience and quality control, contributing to the wider transformation of automotive foundries.

Cross-process and, in some cases, cross-sector machine learning services support data-driven optimisation. Integration into BREOS CORE enables shared production and quality documentation and practical use of the technologies. Industrial partners and BDG support transfer through guidelines, standards and innovative training concepts for different user groups.

Results:

The cluster develops technologies for complete traceability through marking and tracking components, moulds and cores. For clay-bonded sand moulds, it considers sand properties, machine moulding, external influences and timing. Generic methods store, manage and analyse hybrid datasets, linking components to production parameters.

Surface fingerprints provide unique core identification. Integrating traceability into quality management links process and quality data for optimisation and visualisation. Data from existing equipment and mobile measuring stations supports predictive models of core and mould quality. Assistance systems derive and clearly present preventive recommendations for machine operators.

Quality data, including X-ray images and surface scans, supports AI and machine learning models for digital visual inspection of castings. Data is collected, linked and evaluated across work packages. Collaboration within the sand casting cluster develops practical digital solutions to improve production efficiency and quality.

Motivation: 

ReGAIN digitally connects foundry production systems using Catena-X. Identifying training needs and developing continuing education concepts ensures employees can use new technologies and processes effectively. BDG aims to share the results within the consortium and beyond the sector through communication and the involvement of additional foundries and users.

Approach: 

The project first assesses specific training needs in participating foundries. Tailored programmes combine theory and practical training, supported by digital learning formats for different audiences. An advisory committee gathers feedback from additional foundries and users in other sectors. Talks, publications, workshops and conferences promote transfer, while digital guidelines and a learning platform make technologies and concepts accessible across the industry.

Objective: 

BDG aims to secure competitiveness and support sustainable development through targeted training and effective technology transfer. Tailored education expands employees' skills, while guidelines, standards and communication activities promote adoption of innovative technologies and processes within and beyond the foundry sector.

Motivation:

The pandemic and war in Ukraine have intensified major challenges, while the lasting effects of climate change demand timely action. Available technologies must be used to develop sustainable, efficient solutions.

German foundries face particular demands. Their ability to produce complex, lightweight, technically sophisticated castings places substantial requirements on the entire process chain.

Digitalisation can enable demand-oriented production. AI models developed with universities offer significant potential for analysis and optimisation. However, their ongoing development and maintenance can become uncertain when researchers leave their institutions after graduating.

BREOS CORE addresses this with data analytics applications and a powerful runtime environment for AI models and their frontends. Necessary interfaces and a scalable cloud-native service architecture provide a robust, sustainable platform.

As an operating environment for these models, CORE supports long-term use of digital and AI technologies in foundries. Its architecture and scalability help optimise manufacturing processes and prepare them for the future.

Approach:

BREOS develops a module for aggregating, preparing and archiving data between foundry source systems and CORE apps. Integrated into the platform, it provides the database model for storing and enriching data.

The platform is extended to integrate AI models into demonstrator apps, including interfaces, different data models and a generalised frontend with visualisations.

Traceability methods for components, sand moulds and cores are developed and tested. Existing data models are extended for AI applications to improve casting quality and production efficiency, including melt supply, core quality prediction and die temperature control. The models are integrated and visualised in CORE.

Partner collaboration creates synergies for comprehensive process optimisation and quality assurance in casting production.

Objective:

BREOS aims to deliver a maintainable platform suitable for production use, providing a lasting home for foundry AI algorithms. Work includes data processing and analysis, backend functions and AI data models for die casting, alongside traceability methods for sand casting components, moulds and cores.

Motivation:

Germany's automotive and supplier industries, particularly energy-intensive foundries, face pandemic-related pressures, higher energy and raw material costs linked to the war in Ukraine, and climate change. Industrial electricity prices rose by 50% compared with 2020, while European gas prices increased tenfold between October 2020 and October 2022. The sector's roughly 600 foundries and 70,000 employees need more sustainable, resource-efficient processes. Digitalising complex process chains can make production more resilient, sustainable and flexible.

A key foundation is stronger digital connectivity between foundries, suppliers and customers in greenfield and brownfield settings. Faster information supports operational improvements and problem-solving, improving economic and environmental performance. This requires technical solutions such as Catena-X and organisational methods for obtaining and routing the right data.

Approach:

After identifying required data exchange connections, ReGAIN partners receive Eclipse Dataspace Connectors (EDCs), providing gateways to Catena-X and cloud systems.

Shared data models provide a common language. Suitable Catena-X models are selected from requirements analysis, with new models developed where necessary, ensuring recipients understand the data as intended by senders.

The subproject connects foundries across company boundaries along the value chain.

Objective:

Its objective is secure, demand-oriented data exchange between foundries, suppliers and customers. The resulting Foundry Hub provides a foundation for connecting ReGAIN partners and their activities.

Particular attention is paid to AI services. Initial training and later retraining require substantial casting process datasets, with Catena-X providing the exchange infrastructure.

A consistent data model compatible with platforms such as AWS and Microsoft Azure allows aggregators, translators, backups and AI models to be integrated into the Catena-X Foundry Hub. Standard interfaces, appropriate communication, data governance and cross-sector knowledge transfer support broad participation, higher output and lower energy consumption.

Motivation:

Process-chain traceability based on individual sand casting features, with localisation and classification of quality defects.

Approach:

Demonstrators are developed and installed in foundries, connected to relevant interfaces and controls. Cast codes on mould surfaces uniquely identify the resulting castings. Other demonstrators read these codes, create natural surface fingerprints of cores and recognise them during production, forwarding results to higher-level systems.

Objective:

The work package implements IPI and 3Dvisionlab research in production environments, enabling cross-process tracking of castings and their production components. Defect classification demonstrators identify process-related quality issues. They apply 3Dvisionlab methods to localise and classify shrinkage cavities, pores, cold shuts and other defects.

Motivation:

Approach:

Objective:

Motivation:

The University of Kassel's GTK subproject uses AI to improve aluminium die casting efficiency and quality. Sensor-based digitalisation of melt cleaning and release-agent application enables optimisation that reduces energy and resource use. Continuous in-situ quality monitoring and a user-friendly visualisation interface support precise automated impeller control. Infrared thermography measures die surface temperatures to improve casting quality and support the industry's sustainable transformation.

Approach:

GTK's research is organised into the following areas.

I. Optimising die casting through AI-assisted impeller melt treatment

Casting trials and FLOW-3D simulations generate and analyse process data. Sensors capture impeller operation, while a simulation model assesses how process parameters and rotor geometry affect treatment time. AI is trained to recognise relationships. An EMC HYCAL hydrogen analyser continuously measures melt hydrogen content as the target variable. Real-time interfaces supply process data to the AI for automatic, operator-independent selection of impeller parameters and precise hydrogen control.

II. Optimising die casting through AI-assisted, demand-oriented release-agent application

Die surface temperature strongly affects casting quality. Foundries often use spraying to regulate it, but experience-based programming is not always optimal. Non-uniform temperatures are usually measured only locally and intermittently using ReGAIN – Subproject thermocouples. This subproject uses infrared cameras to capture both die halves before and after spraying. In-situ emissivity map corrections improve temperature data for the AI, which determines where and for how long release agent should be applied..

Objective:

GTK integrates AI into casting and melting. Sensor- and AI-assisted melt preparation aims to reach a defined hydrogen content in the shortest feasible treatment time, improving energy use, resource efficiency and quality. Demand-oriented spraying targets the full die surface temperature distribution to reduce porosity, shrinkage and incomplete filling. Because process conditions make reliable full-field cavity temperature measurement difficult, the project develops improved infrared measurement strategies, including in-situ emissivity correction.

Results:

The models are designed for manufacturer-independent use. Continuous melt quality monitoring, AI evaluation and automated impeller settings aim to improve quality and consistency while reducing costs. Universal integration into existing impeller processes is a priority. Demand-oriented spraying regulates die temperatures more precisely and may reduce release-agent consumption. This can improve component quality and extend tool life across aluminium die casting applications.

Motivation:

HEITEC PTS experimentally extends its X-ray systems with optical surface inspection. AI examines correlations between surface and radiographic data. The aim is to identify some defective parts using lower-energy optical inspection before X-ray testing, increasing throughput or reducing the number of X-ray examinations and lowering overall energy use.

As an inspection equipment manufacturer, HEITEC PTS also contributes to data exchange standards from a user's perspective. X-ray data is more complex and voluminous than ordinary sensor data, often reaching 1–10 GB per minute.

Approach:

Project activities:

Suitable optical sensors are selected and tested, then evaluated in a laboratory setup designed for smooth integration into a production inspection system. The setup establishes the technology's capabilities and limitations.

Following feasibility testing and integration design, a sensor is installed in a customer's operating inspection system to generate real data. Digital commissioning using twins of all components identifies compatibility issues in advance and minimises production disruption.

AI methods evaluate surface quality and investigate whether surface measurements reveal defects also detected by X-ray inspection.

HEITEC PTS develops a model for efficient handling of inspection data and adds a database module to its equipment software for storing results. The module is extended to provide Catena-X-compliant export.

Objective:

Customer demonstrators gather richer component datasets without modifying the production line. AI investigates relationships between surface and X-ray results. If optical data can reliably identify defective parts, their energy-intensive X-ray inspection can be avoided, saving energy and potentially increasing inline inspection throughput.

A further objective is to make Catena-X exchange mechanisms suitable for X-ray inspection data. This is challenging because 3D inspection can generate 1–10 GB every minute, far more than typical process sensors.

Motivation:

Approach:

Objective:

Motivation:

Approach:

Results:

Motivation

The European Green Deal, disruption of energy markets by the war in Ukraine, rising prices and the transition to renewables and a circular economy create major challenges for medium-sized foundries. Complex process chains need deeper digitalisation. As a small- and medium-batch producer, LGL requires highly flexible equipment, which can incur greater energy losses than large-series production. Digital analysis targets melt quality, while dynamic energy simulation identifies improvements.

Approach:

Process, energy and quality data are digitally collected to analyse melting. Operational data collection is partly digitalised and melt quality is integrated into a shared database. A measurement concept extends existing energy metering. Anonymised data supports partner development of ReGAIN services for melt quality (IDECO, GTK Kassel, TU Braunschweig), production optimisation and dynamic energy simulation (TU Braunschweig, Fraunhofer IFF). LGL analyses energy and material flows, defines measuring points and records material inputs. Demonstrator analysis and assistance systems are tested and validated, including energy-optimised scheduling.

Objective:

The subproject demonstrates improvements through assistance systems, focusing on energy-flexible operations and melt quality. Simulation and AI support energy-optimised manufacturing, while specific recommendations help skilled operators select better process parameters.  

Motivation:

The project combines real production data from multiple sources with virtual manufacturing knowledge from digital twins for near-real-time quality prediction and process control. Data transfer, sensors, analytics, AI and digital twin integration are central. A major focus is visualising this knowledge at the die casting machine, making complex relationships between settings, machine characteristics, measurements and component quality intuitive. Close industrial and research collaboration supports practical implementation.

The subproject develops technologies that advance digitalisation and Industry 4.0 in German foundries. Predicting virtual component quality within a production cycle through direct links between die casting machines and simulation models represents a new capability for the sector.

Collaboration across disciplines and industrial partners supports transfer to other processes, including core production in work packages 2.x and 3.4.x. Practical solutions are made available to science and industry.

Approach:

Work starts with automated exchange of real process data through a defined interface such as OPC UA. An app makes simulation knowledge available at the workplace. A workflow then links measured production data with simulated statistical quality data to predict quality at the machine and give early recommendations. Finally, AI and machine learning models from real production are applied to simulation data to create virtual assistants.

Objective:

The objective is to integrate and visualise simulation and digital twin knowledge in die casting, permanent mould casting and core production. AI and machine learning analyse expert knowledge embedded in model setup, definition and results. Assistants use real production data to predict quality within the manufacturing cycle and provide recommendations to operators.

Motivation:

The Green Deal, disrupted energy markets, higher prices and circular-economy requirements demand efficient casting production. PINTER GUSS produces hundreds of different components in small and medium batches, requiring substantial flexibility. With its partners, it develops a fully connected process chain, extending established operating and production data collection to complex process data. Linking these data to each component creates a digital product passport and connects process parameters to quality. Statistics, AI and machine learning help explain complex interactions. The subproject demonstrates comprehensive digitalisation in an established medium-sized foundry.

Approach:

A digital product passport requires traceability throughout the process chain. Kempten University, 3Dvisionlab and CMO-SYS develop technologies for tracking castings and sand cores. Demonstrators are integrated into production and support further data collection.

Detailed process analysis requires digital acquisition of machine, process, sensor and quality data . These extend existing production data systems and are supplied to partners in anonymised form. Casting defects must be quantified and classified to relate process data to quality. 3Dvisionlab and CMO-SYS develop automatic optical surface inspection; PINTER also supports Fraunhofer IIS's X-ray work. ReGAIN services address melt quality, production and energy optimisation, and core and mould production. Demonstrators are tested and validated. The product passport also provides a basis for future carbon footprint calculation and a high-quality circular economy.

Objective:

The subproject aims to create a digital product passport that supports analysis tools and assistance systems in other subprojects. PINTER GUSS demonstrates comprehensive process-chain digitalisation in an established medium-sized foundry, aiming to reduce energy consumption and production scrap . The digital product passport also lays the foundation for future carbon footprint calculation and is an essential step towards a high-quality circular economy.

Motivation:

Digitalising manufacturing across sites enables earlier problem detection and improves productivity, quality and flexibility. Greater production efficiency supports the long-term viability of German manufacturing sites.

Approach:

Information flows along the manufacturing chain are modelled, measuring points identified and sensors installed. The resulting data is grouped and processed using trained AI methods to provide insights for targeted optimisation.

Objective:

The objective is a digital representation of manufacturing across sites and along the production chain. It acts as an assistance system for employee decisions and improves the design of new processes.

Motivation:

Many long-established foundries operate equipment of different ages with heterogeneous interfaces and data systems: a typical brownfield environment. Consistent process data collection can create significant value, especially in highly automated die casting with permanent dies. Machine learning extracts process knowledge to improve quality, predict scrap in near real time and reduce downstream costs. Models can also assess peripheral equipment and identify developing trends or failures early.

Approach:

With Wollin and AWEBA, a test setup and research die support AI development for release-agent spraying. A second model builds on spraying results and additional sensor data, such as ejector forces, to generate component-specific predictions for reliable demoulding. Numerical simulation and AI aim to reduce required machine data and create a generic model deployable after a short training period wherever the necessary data is available. Following research trials at the Soltau Light Metal Centre, the models are validated in production at G. A. Röders. Successful models are combined in a cloud system accessible through BREOS CORE.

Objective:

The resulting spraying assistant adjusts release-agent quantities by region according to die temperature distribution and thermal balance. It responds to process fluctuations and improves resilience through digital process representation. Combined with demoulding prediction, it aims to reduce scrap and improve efficiency, profitability and sustainability. Cloud integration supports access with minimal hardware dependence, while visualisation makes results easier to interpret.

Motivation:

Approach:

• Prediction of production stoppages
• Control of the impeller treatment process
• Prediction of casting quality
• Prediction of melt composition during alloying, modification and grain refinement.

Results:

Motivation:

Approach:

Objectives and results:

Motivation:

Germany's predominantly medium-sized die casting industry faces higher energy costs, increasingly complex and larger automotive components, stricter quality requirements, international competition and a growing shortage of skilled workers. Substantially improved automation is needed to address these pressures.

Die casting involves many interdependent parameters, requiring new AI assistants and machine learning methods. Existing equipment often lacks suitable interfaces, data models and sufficiently detailed measurements, so these foundations must first be established.

As a release-agent application specialist, Wollin investigates how to digitalise a process that has largely operated without feedback control. The aim is to provide data for higher-level optimisation and feed model results back into production.

Approach:

With TU Braunschweig, typical spraying faults are categorised and reproduced on real spraying systems. A test rig using an AWEBA die measures their effects and automatically generates sufficiently large training datasets.

In parallel, a data model describes release-agent application across the die geometry for different spray tool configurations, motion programmes and process parameters. It initially supplies target values and later measured values to the partners.

Existing partner spraying systems are extended to implement AI optimisation results automatically.

Wollin also contributes to OPC UA standardisation of the resulting models and interfaces, making future integration into industrial equipment easier.

Objective:

The objective is to capture and digitally represent release-agent application sufficiently for AI-based control. The process is currently largely set up manually and operates without continuous feedback. Practical demonstrations with TU Braunschweig and G. A. Röders show the approach. Since spraying strongly influences scrap and component quality, this can improve profitability and resilience across die casting.