Table of Contents
What is CFD? Industrial Applications
Computational Fluid Dynamics (CFD) is the branch of engineering that uses numerical methods and algorithms to predict how fluids — liquids and gases — behave under prescribed conditions. Instead of building a physical prototype and placing it in a wind tunnel, engineers create a virtual model, define boundary conditions (velocity at inlets, pressure at outlets, wall temperatures), and let a computer solve millions of coupled equations to predict velocity, pressure, temperature, and turbulence throughout the flow domain.
The result is a detailed, three-dimensional picture of the flow that is either impossible or prohibitively expensive to measure experimentally. A CFD simulation of an automotive underbody, for example, gives velocity and pressure at every point in space — something no array of physical probes could achieve without disturbing the very flow being measured.
Industries That Rely on CFD Today
External aerodynamics (drag & lift on body panels), underhood cooling, HVAC cabin airflow, brake cooling ducts, exhaust systems, and EV battery thermal management.
Wing and fuselage aerodynamics, jet engine internal flows (compressors, turbines, combustors), hypersonic re-entry heating, and spacecraft propulsion systems.
Wind loads on tall buildings and bridges, pedestrian-level wind comfort, natural ventilation design, fire and smoke simulation, and pollutant dispersion in urban environments.
Wind turbine rotor aerodynamics, gas turbine combustion, nuclear reactor thermal-hydraulics, geothermal reservoir flows, and pipeline transport of oil and gas.
Blood flow in arteries and around stents, respiratory airflow in the lung, drug delivery via inhalers, heart valve hemodynamics, and microfluidic lab-on-chip devices.
Mixing in chemical reactors, spray drying, heat exchanger optimization, pump and compressor design, distillation columns, and two-phase pipe flows with boiling.
I keep hearing "CFD" but I don't understand what it actually does differently from just measuring things with a wind tunnel. Why would a company use software instead of a real experiment?
Wind tunnels give you real data — absolutely valuable. But CFD gives you the full three-dimensional field: pressure and velocity at every point inside the flow, not just at the location of physical sensors. More importantly, in a typical car development programme you need to evaluate maybe fifty aerodynamic variants before production freeze. Building fifty physical prototypes and booking fifty tunnel sessions would cost millions and take a year. CFD lets you run those fifty cases overnight on a cluster. The modern approach uses both: CFD early in design to screen options cheaply, tunnel testing to validate the final candidate.
Makes sense for cars. But what about biomedical — surely you can't put sensors inside a patient's artery anyway?
Exactly — that's CFD's unique superpower. You physically cannot place fifty pressure sensors inside a living coronary artery. And inside a jet engine combustor operating at 1600°C, no physical sensor survives long enough to be useful. CFD operates safely where hardware cannot. That's why it has become indispensable: it is the only window into flows that are either inaccessible or too extreme to instrument.
Governing Equations: The Navier-Stokes System
All CFD solvers — from OpenFOAM to ANSYS Fluent to a research code — ultimately solve variants of the Navier-Stokes equations. For incompressible, Newtonian flow with constant viscosity, the governing system consists of three coupled partial differential equations:
1. Continuity Equation — Mass Conservation
For an incompressible fluid, the velocity field must be divergence-free: what flows into any control volume must immediately flow out. In compressible flows, this becomes the full form \(\partial\rho/\partial t + \nabla\cdot(\rho\mathbf{u}) = 0\), where density \(\rho\) now varies with pressure and temperature.
2. Momentum Equation — Newton's Second Law for Fluids
Reading left to right: the left-hand side is the rate of change of momentum (inertia). On the right: pressure gradient forces, viscous diffusion forces, gravitational body force, and any additional source terms \(\mathbf{f}\) (such as rotating frame forces or MHD body forces). The nonlinear convection term \(\mathbf{u}\cdot\nabla\mathbf{u}\) is what makes turbulence so notoriously difficult — it creates a continuous cascade of energy from large to small scales.
3. Energy Equation — First Law of Thermodynamics
Where \(\Phi\) is viscous dissipation and \(\dot{q}\) is a volumetric heat source. This equation is solved whenever temperature matters: conjugate heat transfer, combustion, natural convection, and all compressible flows require the energy equation to be active.
Why We Cannot Solve These Equations Exactly
Turbulent flows contain eddies spanning from the largest geometry scale (meters in a wind turbine blade) down to the Kolmogorov microscale (micrometers) where kinetic energy is dissipated as heat. Resolving all scales simultaneously — Direct Numerical Simulation (DNS) — requires a mesh with \(N \sim Re^{9/4}\) points. At a Reynolds number of \(10^6\), that is trillions of cells — computationally infeasible for industrial geometry. Therefore, in practice we use turbulence models that approximate the effect of unresolved small scales through additional transport equations.
Numerical Methods: FVM, FEM, and FDM
Three main discretization strategies are used in CFD, each with distinct characteristics:
| Method | Full Name | Core Idea | Conservation | Typical Applications | Representative Codes |
|---|---|---|---|---|---|
| FVM | Finite Volume Method | Integrate governing equations over discrete control volumes; compute fluxes at cell faces | Locally and globally conservative by construction | Standard for industrial CFD; handles unstructured meshes and complex geometry naturally | OpenFOAM, ANSYS Fluent, STAR-CCM+, CFX |
| FEM | Finite Element Method | Multiply equations by test functions and integrate (weak/variational form); high-order possible | Weakly conservative; requires care for advection-dominated flows | Multiphysics coupling (fluid+structure), biomedical flows, microfluidics, low-Re incompressible | COMSOL, Abaqus CFD, FEniCS, deal.II |
| FDM | Finite Difference Method | Replace derivatives with Taylor-series differences on structured grids; conceptually simplest | Depends on scheme choice | Structured-grid DNS/LES research codes; atmospheric and ocean modeling | NGA2, Nek5000, WRF (weather) |
In industrial practice, the Finite Volume Method dominates because its cell-centered flux formulation is inherently conservative — an essential property when mass and momentum must balance to machine precision across the domain. FVM also applies directly to arbitrary unstructured meshes generated around real-world geometry: turbine blades, car bodies, pipe networks.
Pressure-Velocity Coupling: The SIMPLE Family
In incompressible CFD, pressure does not have a dedicated governing equation — it appears only as a gradient in the momentum equation. Algorithms in the SIMPLE family (Semi-Implicit Method for Pressure-Linked Equations) handle this by iterating between a momentum predictor step and a pressure-correction step that enforces mass conservation. Variants include:
- SIMPLE — steady-state cases; widely used in ANSYS Fluent and OpenFOAM's
simpleFoam - PISO (Pressure-Implicit Split Operator) — transient cases; larger time steps allowed
- PIMPLE — combination of PISO + SIMPLE outer iterations; default in OpenFOAM's
pimpleFoam
Why is FVM used for CFD but FEM is used for structural analysis? Can't you just use FEM for everything?
Good question. FEM was originally developed for solid mechanics, where you want to minimize a potential energy functional — the weak form naturally falls out of that. It works fine for Stokes (very slow, viscous) flow. But for advection-dominated fluid flow at higher Reynolds numbers, standard Galerkin FEM has stability issues with the convection term. You need stabilization techniques (SUPG, GLS) to avoid spurious oscillations. FVM, by contrast, was designed for advection: upwind differencing schemes are built directly into the flux computation. That's why the CFD community converged on FVM. That said, modern FEM codes like deal.II or FEniCS with proper stabilization can absolutely do CFD — it's just not the mainstream path.
The CFD Workflow Step by Step
Import CAD geometry; repair surface defects (gaps, overlaps, non-manifold edges); suppress small features (bolt holes, chamfers) that would force unnecessary mesh refinement. Define fluid/solid regions. Tools: SpaceClaim, Pointwise, CADfix, ANSA.
Discretize the fluid domain into cells. Structured (hexahedral) meshes give higher accuracy per cell but are difficult to generate for complex geometry. Unstructured (tetrahedral or polyhedral) meshes automate easily but require more cells for the same accuracy. Critical decision: near-wall mesh. For low-Re wall treatment, target y⁺ < 1; for wall functions, y⁺ between 30 and 300. Getting y⁺ wrong is one of the most common sources of turbulence model error. Tools: ANSYS Meshing, Pointwise, snappyHexMesh (OpenFOAM), ICEM CFD, Fluent Meshing.
Assign inlet velocity profiles (or mass flow rates), outlet pressure, wall roughness, heat flux or temperature, turbulence intensity, and species concentrations for reactive flows. Choose turbulence model. Select steady-state or transient. Set discretization schemes (first-order is stable but diffusive; second-order is more accurate but can diverge on poor meshes). Convergence criteria: residuals typically targeted below 10⁻³ to 10⁻⁴ for velocity and continuity, 10⁻⁶ for energy.
The solver iterates through the discretized equations, updating velocity, pressure, and scalar fields each iteration. Monitor residuals and key integral quantities (drag force, Nusselt number, pressure drop) in real time. For parallel computing, domain decomposition splits the mesh across CPU cores using MPI. Industrial simulations run on HPC clusters with hundreds to thousands of cores — a typical automotive external aero simulation may use 1000 cores over 12 hours.
Extract velocity contours, pressure distributions, streamlines, iso-surfaces (e.g. Q-criterion vortices), and surface integrals (drag, lift, heat transfer coefficients). Always validate against experimental data or analytical benchmarks. Never report CFD results without a mesh sensitivity study showing that your quantities of interest are independent of further refinement. Tools: ParaView (free, open source), EnSight, Tecplot, ANSYS CFD-Post.
All CFD Topics Covered on This Site
This archive covers computational fluid dynamics from first principles through advanced industrial simulation techniques. Browse by sub-topic:
Reynolds number, Mach number, boundary layer theory, dimensional analysis, flow regime classification
Pipe flow, channel flow, lid-driven cavity, SIMPLE algorithm, pressure-velocity coupling
Mach number, shock waves, normal/oblique shocks, nozzle flow, transonic and supersonic regimes
k-ε, k-ω, k-ω SST, Reynolds Stress Models, wall functions, y⁺ requirements
Large Eddy Simulation, Smagorinsky subgrid model, WALE, Direct Numerical Simulation, turbulence spectra
VOF method, Euler-Euler, Lagrangian particle tracking, cavitation, free surface flows
Forced convection, natural convection, conjugate heat transfer (CHT), radiation in CFD
Airfoil analysis, drag/lift prediction, boundary layer separation, wake structure, wind engineering
Pipe networks, duct flow, heat exchangers, flow distribution manifolds, pressure drop analysis
Premixed and diffusion flames, species transport equations, reaction mechanisms, NOx prediction
Open-source CFD: case setup, solver selection, blockMesh, snappyHexMesh, parallelization, complete tutorials
ANSYS Fluent, STAR-CCM+, CFX, COMSOL — features, capabilities, strengths, and licensing models
Recommended Learning Path for CFD Beginners
I have a mechanical engineering degree and want to get into CFD for work. I've never run a simulation. Where do I actually start — just dive into Fluent tutorials?
The most common beginner mistake is jumping straight into Fluent tutorials without building the physics foundation. You'll produce results, but you won't know whether to trust them. In practice the engineers who become genuinely good at CFD follow a structured path:
Step 1 — Fluid mechanics fundamentals (2 weeks). You need to understand Reynolds number intuitively, know what a boundary layer is and why it matters, and be able to explain the difference between laminar and turbulent flow. Without this, choosing boundary conditions is guesswork.
Step 2 — Understand the governing equations physically. You don't need to derive Navier-Stokes from scratch, but you should recognize what each term physically represents — convection carries momentum; viscosity diffuses it; the pressure gradient drives flow. This understanding is what tells you, later, whether your results make physical sense.
Step 3 — First validation case. Set up Hagen-Poiseuille pipe flow or the lid-driven cavity — problems with known analytical or benchmark solutions. Compare your CFD output directly against the reference. If it matches, you set up the simulation correctly. If not, you learn exactly where you went wrong before it matters.
Should I learn OpenFOAM or ANSYS Fluent first? I've heard OpenFOAM is free but hard.
If you have a university Fluent license, start there — the GUI removes setup friction and lets you focus entirely on the physics. Once you understand what the solver is actually doing, transition to OpenFOAM. The command-line workflow forces you to understand every setting explicitly because there is no wizard filling in defaults. Six months of OpenFOAM cases will build deeper CFD understanding than clicking through Fluent menus for a year. In industry, the best CFD engineers are fluent in both: commercial tools for production runs (faster turnaround, better support), OpenFOAM for research, custom solvers, and cost-sensitive projects.
- Fluid mechanics fundamentals — Reynolds number, Bernoulli, boundary layer, laminar vs turbulent
- Physical interpretation of Navier-Stokes — what each term represents; incompressible vs compressible
- FVM basics and SIMPLE algorithm — how the solver actually turns PDEs into algebra you can compute
- First benchmark case — Poiseuille pipe flow or lid-driven cavity with analytical/benchmark reference
- Turbulence modeling — k-ω SST theory, y⁺ near-wall mesh requirements, wall function selection
- Mesh independence study — Richardson extrapolation; verify that key quantities stop changing on refinement
- Advanced topics by application — multiphase, conjugate heat transfer, compressible flows, LES
CFD Software Comparison
| Software | Type | Discretization | Best For | License |
|---|---|---|---|---|
| OpenFOAM | Open Source | FVM (unstructured) | General industrial CFD, turbulence, multiphase, HPC, research, custom solvers | Free (GPL/Apache) |
| ANSYS Fluent | Commercial | FVM (cell-centered) | Industry workhorse: multiphase, combustion, conjugate heat transfer, excellent support | Commercial |
| ANSYS CFX | Commercial | FVM (vertex-centered) | Turbomachinery, rotating machinery, steam and gas turbines | Commercial |
| Simcenter STAR-CCM+ | Commercial | FVM (polyhedral) | Large-scale automotive & aerospace, design optimization automation, co-simulation | Commercial (power token) |
| COMSOL Multiphysics | Commercial | FEM | Multiphysics coupling (fluid + structural + EM + chemical), microfluidics, academic research | Commercial (modular) |
| SU2 | Open Source | FVM | Aerospace shape optimization, discrete adjoint-based design, compressible flows | Free (LGPL) |
| Code_Saturne | Open Source | FVM (unstructured) | Nuclear thermal-hydraulics, atmospheric boundary layer flows, developed by EDF (France) | Free (GPL) |
| Nek5000 / NekRS | Open Source | Spectral Element (FEM-like) | High-fidelity DNS/LES research; GPU-accelerated via NekRS | Free |
New to CFD? Start Here
Our structured learning roadmap takes you from zero CFD experience to running industrial simulations — with practical worked examples in OpenFOAM and ANSYS Fluent at each step.
View the Full Learning Roadmap