Using Agentic AI to do simulation

Is it possible for Agentic AI to perform OPAL-RT simulations?

Yes—agentic AI can be used to operate and automate OPAL-RT simulations, but it should act as an orchestration and analysis layer rather than replace the deterministic real-time simulator.

A practical architecture would be:

AI agent interprets the objective
For example: “Run a sweep of load resistance from 5 Ω to 50 Ω and identify the point where the DC-link voltage becomes unstable.”

Deterministic automation layer executes the workflow
A Python, MATLAB, LabVIEW, C++, or TestStand application can:

Load and compile the RT-LAB model

Start, stop, reset, and monitor execution

Set exposed model signals or runtime parameters

Run test sequences and parameter sweeps

Collect logs, scopes, and measured signals

Generate pass/fail results

RT-LAB documentation describes a comprehensive API for custom online applications and dynamic modification of model signals or parameters during execution. https://opal-rt.atlassian.net/wiki/spaces/RD/pages/45880058/RT-LAB+User+Guide+Introduction

The AI agent analyzes the results
The agent can detect violations, compare results against requirements, identify anomalies, propose the next test, and produce a report.

Safety and authorization gates remain in place
For HIL or power-HIL systems, the agent should not be allowed to arbitrarily change hardware-facing signals, protection thresholds, or operating conditions without limits and approval. The execution layer should enforce:

Allowed parameter ranges

Rate limits on changes

Emergency-stop handling

Hardware interlocks

Test approval and audit logging

A defined safe state after communication loss