Agentic workflows for simulation-driven decision support

Running simulations typically requires specialised knowledge, familiarity with complex software, training to setup the models, and access to technical documentation.

This creates a practical challenge: valuable models exist, but they’re difficult to use in day-to-day operations. As a result, many decisions are still made based on experience rather than data.

Agentic workflows: A simulation co-pilot in action

To address this, an agentic workflow—a modern expert system that acts as a simulation co-pilot. It helps process experts interact with complex models and documentation using natural language, making it easier to work with simulations without needing deep technical training.

The system’s most valuable feature is its ability to automatically configure and execute simulations for various operating scenarios. This allows users to explore how different conditions affect process behaviour without manually setting up models or navigating technical interfaces.

To support this, the system includes a Retrieval-Augmented Generation (RAG) component.  The RAG handles information retrieval from knowledge sources (like model and operation documentations) ensuring the responses from framework are grounded. This allows the framework to act as an interactive user guide.

Users can interact with the system to:

• Automatically configure and execute simulations for various operating scenarios
• Explore the impact of different inputs and operating conditions based on simulations
• Access relevant information about process operations and models

This makes simulation tools more usable in everyday operations.

Proof of concept use case: Language model-based expert system

For the proof of concept, we used a metallurgical process setting involving the Søderberg electrode model developed by SINTEF and ERAMET Norway (in NextGenSøderberg project). The expert system integrates existing model documentation, simulation tools, and operational guidelines into a unified interface. Operators can ask questions like “What input parameters can I change in the simulations?” and receive actionable responses, including simulation results, relevant documentation excerpts, and a dynamic interface to update parameters.

This demonstrates how language model-based expert systems can support operational decision-making without requiring users to become simulation experts themselves.

The solution is designed to run securely on-premises, ensuring full data confidentiality while enabling integration into existing workflows.

Business impact

By simplifying access to modelling tools and operational knowledge, the system delivers clear benefits:

• Reduces training time for using simulation tools
• Accelerates decision-making with simulation-backed insights
• Improves efficiency by integrating simulations into daily workflows
• Ensures data confidentiality through secure on-premises deployment

Key benefits

• Easier setup and execution of simulations
• Enhanced decision support for industrial operations
• Better process understanding and responsiveness

Acknowledgement

This was a use-case developed under the EuroCC2 project hosted by the National Competence Center (NCC) of Norway (https://www.sigma2.no/nb/nasjonalt-kompetansesenter-hpc)

Top image created by Treesa Rose Joseph using Microsoft Copilot.