ORCA: AI-powered assistance systems for automated simulation studies
The ORCA project is developing an assistance system that helps production planners to automatically generate complete simulation studies. Users formulate their optimisation queries in natural language; specialised large language model agents then define the objectives, carry out experiments and evaluate the results. A knowledge graph stores identified cause-and-effect relationships, enabling both individual recommendations for action and the reuse of knowledge already acquired.
The aim
Simulation studies in digital twins are regarded as a key technology for evaluating ‘what-if’ scenarios in production planning and control. In practice, however, they are often used only sporadically, as the manual effort involved – from defining objectives through modelling to interpreting results – is considerable. Particularly for those with little experience, such studies take weeks, whilst decision-making windows are shrinking to a matter of days. Previous automation approaches cover individual sub-processes but do not independently translate natural-language queries into complete, validated simulation studies. This is precisely where ORCA comes in.
The approach
ORCA combines specialised LLM agents into a socio-technical multi-agent assistance system. These agents classify experiments and objectives, carry out data retrieval, handle the design of experiments, and implement and control the actual simulation studies in line with established process chains such as VDI 3633. A knowledge graph interprets the results and provides explainable recommendations for action, whilst also serving as the basis for sustainable knowledge management. The innovation lies in the combination of LLM-supported semantic interpretation and deterministic validation, which enables technical risks—such as misinterpretations of the models or challenges in agent coordination—to be managed in a targeted manner.
SimPlan as a research partner
As a research partner, SimPlan contributes its expertise in simulation, optimisation and software integration to the project. The focus is on building, adapting and extending the material flow and plant simulation models, as well as their agent-compatible interfaces. In addition, SimPlan is responsible for the technical integration of the simulation environment into the overall demonstrator, which includes the development of agent-based APIs and a GUI for end-user interaction. SimPlan is also actively involved in the development of agent strategies, the set-up of the test environment and the use case studies.
Project partners
Funded by
The research project is funded by the Baden-Württemberg Ministry of Economic Affairs, Skilled Trades and Tourism.
Funding began in August 2026 and is expected to continue until the end of July 2028.
The project is managed by VDI/VDE Innovation + Technik GmbH
