Simulate heat

You want to run a thermal study on a design and see how heat moves through it. This guide covers a quick conduction solve, higher-fidelity conjugate and natural-convection studies, and fast surrogate approximations for iteration. It assumes an MCP-connected agent (see the quickstart).

Quick vs. high-fidelity

Pick the cheapest study that answers your question:

• Quick — run_thermal solves conduction only. Use it to get steady-state temperatures and find hot spots in seconds while you are still moving parts around.
• High-fidelity — simulate_conjugate_heat and simulate_natural_convection couple the fluid to the solid. They are slower but capture how air actually carries heat away, which conduction-only solves miss.

1. Run a baseline thermal solve

Start quick. A conduction-based solve gives you steady-state temperatures with the heat sources you’ve defined.

“Run run_thermal on the current design with the heat sources defined and report the peak temperature.”

Use this to confirm placement and find where the heat concentrates before you pay for a coupled solve.

2. Add fluid–solid coupling

When you need real cooling behavior — airflow carrying heat off the solid — run a conjugate heat-transfer study.

“Call simulate_conjugate_heat to couple the cooling airflow with the solid.”

3. Model passive cooling

If the design has no fan and relies on buoyancy-driven airflow, simulate natural convection.

“Run simulate_natural_convection for the enclosure and tell me if it stays under the limit.”

4. Iterate fast with surrogates

High-fidelity solves are too slow to repeat for every design tweak. A surrogate is a cached, learned approximation: train it once on your solve data, then predict near-instantly for new configurations.

If a surrogate already exists, predict directly:

“Call predict_surrogate for this configuration.”

If none exists yet — or you have fresh solve data to learn from — train one first, then predict from it:

“Run train_surrogate on the thermal results, then use predict_surrogate for the next variants.”

Use the surrogate to sweep variants quickly, then confirm the final candidate with a full simulate_conjugate_heat or simulate_natural_convection solve.

View the results

The viewer at /app/ shows the simulation field overlays — temperature gradients and flow — rendered on the model, so you can inspect hot spots directly instead of reading numbers.