CFD Setup Prompts: How to Ask AI Better Engineering Questions

When engineers ask AI tools broad questions like "What CFD setup should I use for a pipe?", the results are almost always disappointing. Generic prompts yield generic, Wikipedia-level answers that lack the physics constraints and rigorous validation requirements of real-world engineering.

AI cannot replace solver documentation, rigorous validation, or engineering judgment. However, when properly constrained, AI can be an excellent tool for structuring your thinking, generating pre-simulation checklists, finding missing assumptions, and drafting review questions.

This guide provides a framework and copy-ready prompts for CFD engineers, CAE analysts, managers, and automation developers to safely interrogate AI about simulation setups.

[!WARNING] Never ask AI to decide these alone:

• Turbulence model choice (always verify against literature and flow regime).
• Wall treatment and $y^+$ targets.
• Material constants (always use verified property tables).
• Boundary condition realism.
• Convergence acceptance criteria.
• Safety-critical design conclusions.

What to Provide Before Asking AI

To get an engineering-grade response, your prompt must include:

• Flow type: Compressible, incompressible, multiphase, reacting?
• Objective quantities: Pressure drop, drag coefficient, heat transfer rate?
• Geometry assumptions: 2D, 3D, axisymmetric, symmetry planes?
• Operating conditions: Velocity, temperature, pressure ranges.
• Material model: Constant density, ideal gas, non-Newtonian?
• Expected flow regime: Laminar, turbulent, transition, Mach number?
• Solver/software context: open-source CFD software, commercial CFD software, open-source CFD software?
• Mesh constraints: Polyhedral, hexahedral, boundary layer requirements?
• Validation data: Is there analytical data or empirical reference data available?

---

1. Defining Physics and Objectives

Low Risk: Structuring a checklist for a defined problem.

Bad Prompt:

"How do I simulate flow over a wing?"

Better Prompt:

"I am setting up a 3D steady-state incompressible CFD simulation for an aircraft wing section at $Re = 3 \times 10^6$ and $Ma = 0.15$. The primary objective is to accurately predict the drag coefficient and the onset of flow separation. Generate a comprehensive list of the physical phenomena I need to model and the critical solver settings I need to define before generating the mesh. Do not recommend specific solver toggles; instead, list the engineering decisions I must make."

Why this works: It restricts the AI from giving a tutorial and instead forces it to output a decision checklist based on specific flow conditions.

---

2. Boundary Condition Review

Medium Risk: Identifying conflicting or ill-posed boundary conditions.

Bad Prompt:

"What boundary conditions are best for a pump?"

Better Prompt:

"I am modeling a centrifugal pump using water at $20^\circ C$. My proposed boundary conditions are: Mass Flow Inlet (10 kg/s) and Pressure Outlet (0 Pa gauge). The walls are adiabatic with a no-slip condition. Identify any potential numerical instabilities or physical inconsistencies with this combination of boundary conditions. Suggest two alternative boundary condition pairings and explain the physical implications of each."

Why this works: It provides a concrete proposal and asks the AI to play the role of a reviewer looking for numerical reflection or continuity issues.

---

3. Material Properties and Operating Conditions

High Risk: Material models directly impact governing equations.

Bad Prompt:

"What is the density of air?"

Better Prompt:

"I am running a natural convection simulation inside an electronics enclosure where the air temperature will vary between $20^\circ C$ and $85^\circ C$. Should I treat the air as incompressible using the Boussinesq approximation, or should I use an ideal gas law formulation? List the mathematical assumptions of the Boussinesq approximation and calculate the maximum expected density variation to help me justify this decision."

Why this works: It asks for the mathematical assumptions and the specific calculation needed to justify the engineering decision, rather than asking the AI to just pick one.

---

4. Solver Setup Completeness

Medium Risk: Catching forgotten solver parameters.

Bad Prompt:

"Check my open-source CFD software setup."

Better Prompt:

"I am preparing a steady-state simulation using the SIMPLE algorithm for a highly swirling flow. I plan to use 2nd-order upwind for momentum and 1st-order for turbulence quantities initially. What are the common numerical diffusion risks associated with this discretization scheme in swirling flows? Create a step-by-step solver progression plan to move from a robust 1st-order initialization to a stable 2nd-order final solution."

Why this works: It demonstrates domain knowledge and asks the AI to construct a safe numerical progression plan.

---

5. Mesh and Near-Wall Planning Handoff

Medium Risk: Translating physics into mesh requirements.

Bad Prompt:

"How fine should my mesh be?"

Better Prompt:

"I need to resolve the boundary layer for a flow over a flat plate where the expected skin friction coefficient $C_f$ is approximately $0.003$ and the free-stream velocity is $15 \text{ m/s}$. I am using a $k-\omega$ SST model and require a $y^+ \approx 1$. I have a target expansion ratio of 1.2. Write a short engineering brief I can hand to the meshing engineer detailing the required first cell height, the estimated boundary layer thickness, and the number of inflation layers required to cover the boundary layer."

Why this works: It provides all the necessary fluid dynamic inputs and asks the AI to format the output as a handoff document.

---

6. Pre-Simulation Review Checklist

Manager / Team-Lead Prompts

Prompt 1 (Completeness Review):

"I am reviewing a CFD simulation brief from a junior engineer. The objective is to predict pressure drop through a complex manifold. The brief specifies an incompressible, steady-state $k-\epsilon$ simulation with a polyhedral mesh. What are 5 critical probing questions I should ask the engineer during our design review to ensure they have considered flow separation, mesh independence, and boundary condition placement?"

Prompt 2 (Validation Strategy):

"Our team is proposing a new CFD methodology for simulating heat exchangers. Before we approve the computational expense, generate a rigorous verification and validation (V&V) checklist based on ASME V&V 20 guidelines. What specific experimental data points must we demand from the test lab to validate the thermal and hydraulic predictions?"

---

7. Developer and Automation Prompts

Prompt 1 (Python Script Skeleton):

"I am writing a Python script to automate the setup of an open-source CFD software controlDict file. Generate a Python function skeleton that takes endTime, deltaT, writeInterval, and maxCo as arguments and writes a formatted dictionary string. Do not use any external dependencies or libraries other than the standard library. Include type hints and basic validation (e.g., deltaT must be greater than 0)."

Prompt 2 (Checklist Automation):

"I want to build a simple bash script that checks a CFD project directory before submission to the HPC cluster. Write a bash script that verifies the existence of a mesh/ directory, checks if the boundary_conditions.yaml file exists, and greps for the keyword 'INCOMPLETE' in all text files. The script should exit with an error code if the checks fail."

---

8. Asking AI to Identify Missing Information

Master Prompt: The CFD Setup Review

Keep this prompt saved for whenever you are about to finalize a simulation setup document.

"Act as a senior CFD aerodynamicist reviewing a simulation proposal. Here is my current setup:

• Objective: [Insert Objective]
• Flow Regime: [Insert Regime]
• Boundary Conditions: [Insert BCs]
• Turbulence Model: [Insert Model]
• Mesh Strategy: [Insert Mesh Strategy]

Identify any missing boundary conditions, conflicting physical assumptions, or poorly justified solver choices. Do not solve the problem for me. Instead, output a bulleted list of the exact engineering parameters or physical justifications I need to add to my proposal before it is considered rigorous and complete."