CFD MINDSET
CFD is not just running software
CFD means converting a physical problem into a mathematical model, then a numerical model, and finally a defensible result. Colourful contours without checks on physics, mesh, boundary conditions and validation have little research value.
A good CFD project must explain what was simplified, why it was acceptable, and how the numerical result was verified against physics, literature or experiment.
GOVERNING EQUATIONS
Three core principles: mass, momentum and energy
Conservation of mass in the flow domain.
Newton’s second law applied to a fluid element.
Thermal transport through convection, diffusion and source terms.
Density is treated as constant; common for low-speed liquid flows.
Density changes become important; pressure waves and Mach number matter.
Mean solution does not change with time.
Time history is required; essential for vortex shedding and VIV.
DISCRETIZATION
Finite Volume Method: from continuum domain to control volumes
In the finite volume method, the domain is divided into small control volumes. The integral form of conservation laws is applied to each cell, and fluxes are balanced across cell faces.
Conservation
FVM is strong for conservation of mass, momentum and energy.
Schemes
Upwind, central and higher-order schemes change numerical diffusion and stability.
Pressure–velocity coupling
Algorithms such as SIMPLE and PISO coordinate pressure and velocity fields.
MESH QUALITY
A good mesh is the foundation of a reliable result
High skewness can destabilize the solution and reduce accuracy.
Non-orthogonal cells increase discretization error and correction cost.
Large aspect ratios may be acceptable in boundary layers but must be controlled.
Near-wall resolution must match the turbulence model and wall treatment.
Mesh independence logic
Do not claim a CFD result is reliable before checking whether the main outputs change significantly with mesh refinement.
BOUNDARY CONDITIONS
Boundary conditions must represent the physics
Use when incoming velocity or mass flow is known.
Use when downstream static pressure is known or assumed.
Velocity at solid wall equals wall velocity; thermal condition must also be defined.
Useful only when the domain and physics justify it.
Changing boundary conditions until the solution “looks good” is not CFD. Every boundary condition must have a physical justification.
TURBULENCE
Turbulence model selection is a research decision
Practical engineering average
Cost-effective for many industrial flows; model choice matters.
Unsteady averaged flow
Useful for periodic or transient mean-flow behaviour.
Large eddy simulation
Resolves large turbulent structures; much more expensive.
Direct simulation
Solves all scales; mainly research-level and extremely expensive.
Typical model questions
- Is the flow attached or separated?
- Is near-wall behaviour important?
- Is the problem steady or inherently unsteady?
- Is the computational budget compatible with the chosen model?
CONVERGENCE
Convergence is not only low residuals
Residuals should decrease and stabilize at an acceptable level.
Inlet and outlet mass flow must be physically consistent.
Forces, pressure drop, heat transfer or other key outputs should stabilize.
Contours, vectors and integral values must make physical sense.
VALIDATION + REPORTING
From software output to publishable result
Did you solve the equations correctly? Check mesh, timestep, residuals and numerical settings.
Did you solve the right physical problem? Compare with experiment, benchmark or reliable literature.
Report assumptions, sensitivity, mesh dependence and limitations.
Document geometry, mesh, boundary conditions, solver settings and post-processing method.
OPENFOAM MINI WORKFLOW
case → constant → system → solver → postProcess
For an academic portfolio, a small validated OpenFOAM case is stronger than only listing OpenFOAM as a skill.
CFD knowledge check
0 / 51. What makes a CFD result research-grade?
2. What does FVM balance across control-volume faces?
3. What does y+ mainly relate to?
4. Which item is not enough alone to prove convergence?
5. Validation means:
How this learning page supports my research profile
This CFD education section is designed to show not only that I use CFD software, but that I understand the numerical workflow behind publishable CFD research: modelling assumptions, discretization, mesh control, convergence, validation and reproducible reporting.