The last couple of years have seen Artificial Intelligence transforming the software programming landscape. Now the same transformative power is coming to system engineering. In our latest webinar, we explored the cutting-edge intersection of AI and system simulation, diving deep into Agentic Modeling and Simulation using Dyad and its open-source Julia foundation.
If you spend your days building block-diagram models, dealing with initialization errors, or painstakingly verifying system requirements, agentic AI is about to make your workflow significantly faster.
Here is a recap of how AI agents are changing the game for model-based design.
What is Dyad?
Before diving into the AI capabilities, it helps to understand the foundation. Dyad is a commercial system simulation tool built on top of high-performance, open-source Julia libraries. Distributed as a VS Code extension, it bridges the gap between scientific machine learning and traditional system modeling.
Key features include:
Multi-Domain Simulation: Supports 0D, 1D, and 2D continuous and discrete components.
Advanced Integrations: Interacts with 3D models via reduced-order modeling and seamlessly handles FMI imports/exports.
Industry Footprint: Actively used across aerospace, automotive, and energy utilities, and recently integrated into the Ansys (Synopsys) digital twin portfolio.
Agentic Modeling in Action: Building an Aerospace Brake System
To demonstrate the power of the Dyad AI agent, we didn't just build a simple toy model. We tasked the agent with developing a proof-of-concept civil aircraft wheel brake system based on the rigorous SAE ARP 4754B development standard.
Here is how the AI agent tackled the engineering workflow from scratch:
1. From Literature to Architecture
Instead of manually dragging and dropping blocks, you can simply prompt the agent. We asked it to review public literature related to the SAE standard and propose a system architecture.
The agent autonomously read the Dyad documentation.
It formulated a plan combining hydraulic actuators, mechanical brake assemblies, and wheel dynamics.
It then generated the declarative Dyad code to build the foundational lumped-parameter model.
2. Self-Correcting Compilation
Building a model is rarely a one-shot process, and the AI agent mimics the iterative workflow of a human engineer. When the agent attempted to compile the proposed code, it encountered typical syntactical errors and missing dependencies.
Autonomous Debugging: The agent recognized the errors, added the necessary hydraulic and rotational component libraries, and fixed connection types (e.g., swapping a translational flange for a rotational spline).
Physics Validation Initialization: It actively checked if the number of equations matched the number of unknowns. When it encountered a discontinuity (a sign flip causing the wheel to accelerate instead of decelerate), it ran a forensic debugging protocol to correct the physics automatically.
3. Automated Requirements Verification
In standard aerospace programs, models exist to verify requirements. Once the model compiled and simulated successfully, we tasked the agent with testing it against standard industry specifications.
The agent built customized test harnesses to verify critical metrics: stopping time, energy conservation, maximum brake torque, and Rejected Takeoff (RTO) energy absorption.
It simulated the edge cases to verify safety requirements, such as ensuring the system did not cause wheel lockup under full brake application.
Finally, the agent automatically extracted the relevant variables and generated the necessary plots to visually prove the requirements had been met, saving engineers the tedious task of manual data visualization.
Why This Matters for Systems Engineers
The demonstration highlighted a massive shift in how simulation models can be built. By treating an AI agent as a modeling assistant, you can:
Accelerate Proof-of-Concepts: Quickly turn standard specifications or PDFs into functioning baseline models.
Automate Tedious Tasks: Offload the frustration of balancing equations, hunting down initialization errors, and formatting plots.
Focus on Complexity: Let the AI handle the boilerplate architecture so you can focus on adding advanced multi-domain complexities and refining system controls.
Dyad's Agentic AI goes beyond basic programming. It understands physical laws, works within your system's limits, and automatically validates your engineering designs autonomously. Watch the full webinar to learn more.
Interested in trying this out for yourself? Dyad is available via the VS Code extensions marketplace. For enterprise inquiries, specific project collaborations, or access to our weekly workshop sessions, contact the Dyad sales team today.
