Your free CFD pipeline — the whole chain, $0 in licenses

CFD runs in four stages. There's a free, open-source tool for every one — and they run on the workstation you already own. Green = your pick. The boxes on the right are swap-ins for the solve step.

Why this matters: the survey paper you just read makes the same point — open-source CFD isn't a downgrade, it's the same physics (OpenFOAM matches Ansys Fluent in head-to-head tests). For someone on a tight budget with a strong GPU, this is the whole game: free tools, your own hardware, no per-seat fees. The only thing you give up is hand-holding — and SimScale's free cloud hours cover you when OpenFOAM's command-line gets annoying.

What each stage is — click to open it up

Open a stage to see your pick, the free swap-ins, and when to use which.

1 · Draw it — the geometry

Your pick: FreeCAD — free, parametric (change a number, the shape updates).

Free swap-ins: OpenSCAD (shapes from code), Blender (organic shapes/render), OpenVSP (NASA's aircraft-shape tool — handy for drones/aero), SALOME (all-in-one).

Why it's first: nothing meshes or solves without a clean shape — the geometry is the foundation.

2 · Mesh it — cut the shape into cells

Your pick: Gmsh — free, scriptable, the standard open mesher.

Free swap-ins: Netgen, SALOME, enGrid (good boundary-layer grids), or OpenFOAM's own blockMesh / snappyHexMesh.

Why it matters: the mesh decides accuracy and run-time. Too coarse = wrong; too fine = slow. This stage is where most of the skill (and most of the headache) lives.

3 · Solve it — run the physics (the engine)

Your pick: OpenFOAM — free, C++, ~100 built-in solvers (flow, heat, turbulence, combustion, multiphase). It's the engine inside SimScale and many paid wrappers.

Free swap-ins, by job:

· SU2 — aerodynamics + adjoint optimization ("tell me how to reshape it to hit the target") → racing, drones, the radiation-robot intakes.

· FluidX3D — lattice-Boltzmann, GPU-fast, runs right on your graphics card via OpenCL.

· Elmer — multiphysics (fluid + heat + structure together).

· Flowsquare — dead-simple 2D, draw the setup in MS Paint (good for learning).

· SimScale — cloud, ~3,000 free hours, zero install, in a browser.

4 · See it — turn numbers into pictures

Your pick: ParaView — free, the industry-standard CFD post-processor (the same one the paid tools lean on).

Free swap-ins: VisIt, MayaVi.

Why: a solve spits out gigabytes of numbers. This is where you actually see the airflow, the heat, the pressure — and decide what to change.

+ Optimize the shape — let the computer design the best one (generative)

What it is: instead of you drawing a shape and testing it, the computer generates the best shape. You give it the goal (least pressure drop, most flow) and the limits (fit this box, this material), and it explores hundreds of CAD-ready options and hands you the winners. For fluid parts (valves, intakes, manifolds) this is "generative fluid paths."

Free way (your pipeline): SU2's adjoint optimization — it computes exactly how to nudge the shape toward the target, then loops with OpenFOAM. That's the "auto-reshape to target" box up top.

Paid GUI version: Autodesk Fusion 360 Generative Design — the MFG501970 class you saved (Fernandez, AU2022). Same idea, friendly interface, but it needs an Autodesk subscription + cloud credits. Learn the concept and workflow from it; do the runs free with SU2.

Where it fits you: the radiation-robot harvester intakes, racing aero / ducts, and any "best-shape" engineering-spec work.

+ Later: the AI layer — make it near-instant

What it is: once you've run a few real simulations, you train a surrogate model (a neural net) on the results. It then predicts new runs in seconds instead of hours — the "AI surrogate / physics-informed" trick the survey describes.

Free tools: PyTorch / TensorFlow on top of your OpenFOAM / SU2 output.

Where it fits you: this is the bridge between CFD and your ternary/27D system — the model that learns the physics so the system can reason fast. A later step, not now.