Short answer. Computational fluid dynamics (CFD) predicts how a gas or liquid flows through or around a design by solving the equations of fluid motion with a computer. The space the fluid occupies is divided into many small cells, often millions, and in each one the solver balances mass, momentum and, when temperature matters, energy. The result is a map of velocity, pressure and temperature, and the numbers engineers decide with: pressure drop, flow split, peak velocity.
What CFD actually solves
The motion of a viscous fluid such as air or exhaust gas is described by the Navier–Stokes equations: conservation of mass, momentum and energy, relating the fluid’s velocity, pressure, temperature and density. For real engineering geometry they are far too difficult to solve by hand, so CFD solves approximations of them numerically. The flow domain is divided into a mesh of cells, the equations are written for each cell, and the solver iterates until every cell agrees with its neighbours and with the conditions set at the boundaries.
Most engineering flows are turbulent. Resolving every eddy would be far too expensive for design work, so most industrial CFD solves the time-averaged (Reynolds-averaged, or RANS) equations and represents the effect of turbulence with a turbulence model. That is efficient and widely used for pressure drop and flow distribution, but the turbulence model is also one of the main sources of modeling uncertainty, and a time-averaged result contains no unsteady fluctuations. That is why a steady CFD result can’t be read directly as a noise prediction, although its velocities support an estimate of flow-generated noise (CFD vs. acoustic simulation).
How a CFD study is built
- Define the question and the operating points: the flow rates, temperatures and limits the answer will be judged against.
- Prepare the geometry: simplify the CAD, but keep the passages, edges and openings that control the flow.
- Mesh the flow domain, with finer cells where the flow changes quickly, such as near walls and through restrictions.
- Choose the physics and set the boundary conditions: steady or unsteady, the turbulence model, the inlet flow and outlet pressure, and sub-models for parts too fine to mesh directly, such as perforated plates, filters or packing.
- Solve, and confirm the solution has converged.
- Check it: does the answer change when the mesh is refined, and does it agree with hand calculations, supplier data or test results?
- Turn the results into the numbers and pictures the decision needs, with the model’s limitations stated.
What CFD can tell you
- Pressure drop between any two points, and how it divides between components
- Flow distribution: how evenly flow splits between parallel passages, across a filter or radiator face, or across an outlet
- Velocity: where the flow speeds up, separates or recirculates
- Temperature, where heat is part of the problem, for example how hot exhaust gas raises velocity and pressure drop
- How each of those changes between design variants, before anything is built
It can’t tell you more than its inputs and models allow. Geometry simplified too far, an operating point that doesn’t match reality, or a sub-model that misrepresents a perforated plate will all produce a confident-looking answer that is wrong. That is what the checking step is for.
How accurate is CFD?
There is no single number. NASA’s CFD verification and validation guidance, which follows the AIAA guide on the subject, separates the differences between a simulation and reality into errors that can be identified and estimated, and uncertainties that come from incomplete knowledge. In practice, the main ones are:
- Discretization error: the mesh and time step are finite. A mesh-refinement (grid convergence) study runs the model on successively finer meshes to estimate it.
- Iterative error: the solver stopped before the solution fully settled. Monitoring residuals and the quantities of interest controls it.
- Physical modeling uncertainty: turbulence models and sub-models are approximations. Comparison with test data (validation) is how it is judged.
- Input uncertainty: simplified geometry, boundary conditions and material data.
- Setup error: the wrong model or the wrong inputs. Experience and independent checks catch it.
Two practical consequences follow. Comparisons between design variants run on the same model are often more reliable than a single absolute number, because many modeling errors affect every variant alike. And for silencer pressure drop, accuracy depends heavily on how perforated tubes and porous or absorptive material are represented, so those parts of a model are checked against data or hand calculations before the results are trusted.
Pressure drop: CFD or a hand calculation?
For a straight pipe run with standard fittings, a hand calculation is often enough. Friction loss in a pipe and the loss through each fitting both scale with the dynamic pressure of the flow, ½ρV², multiplied by a friction term or by a loss coefficient from published data. CFD earns its cost when the geometry has no published coefficient (a custom silencer, an unusual outlet), when components sit close enough together to affect each other’s flow, or when the question is how the flow is distributed rather than only how much pressure is lost. The two are complementary: hand calculations are a standard check on a CFD model, and CFD fills the gaps the tables can’t.
CFD or physical testing?
A test measures the real thing, including effects a model leaves out. A simulation is available before hardware exists and shows the whole flow field, not only the points where sensors fit. The strongest programs use both: simulation to compare options and decide what to build and where to measure, testing to confirm the chosen design and to validate the model. When a prototype or test rig already exists and the question is one number, a test may be the faster answer.
CFD vs. FEA
Both divide a problem into many small cells or elements and solve it numerically. In everyday engineering use, CFD means fluid flow, while FEA usually means finite-element analysis of stresses and deflections in solid parts. The finite-element method itself is also used for other physics, including acoustics: silencer and enclosure acoustic models are often finite-element models.
When to bring in a specialist
When the decision is expensive to reverse, when nobody in-house runs CFD regularly, or when an in-house model needs an independent check. CFD consulting at SH Consulting centres on airflow, internal flow and pressure drop: engine intake and exhaust systems, silencers, generator-set enclosures and generator rooms, and equipment airflow.
Related services
Read next
- CFD ConsultingHow a CFD consulting study is scoped, checked and reported.
- CFD vs. Acoustic Simulation: Which One Answers Your Question?What a steady CFD result can and can’t say about noise.
- Generator Exhaust Back Pressure, ExplainedA worked view of pressure drop in an engine exhaust.
Sources
- NASA Glenn Research Center, Navier-Stokes Equations
- NPARC Alliance, NASA Glenn Research Center, CFD Verification and Validation tutorial: the CFD analysis process
- NPARC Alliance, NASA Glenn Research Center, Uncertainty and error in CFD simulations
- NPARC Alliance, NASA Glenn Research Center, Examining spatial (grid) convergence
- J. M. Cimbala, Penn State University, Fluid Mechanics Lesson 08F: major and minor losses in pipe flow