Technical Notes
Recurring questions in CFD and thermal analysis, method notes and validation practice.
How do you demonstrate grid independence?
Running three meshes and saying 'the results are close' is not a demonstration. The method, and the numerical criterion.
Which turbulence model should you choose?
k-ε or k-ω SST? The decision follows not from the geometry but from whether the flow separates.
When is CFD necessary, and when is it not?
Not every problem needs CFD. Commissioning an analysis where a hand calculation suffices wastes time and money.
Convergence: why falling residuals are not enough
Residuals dropping below a threshold does not mean the solution has converged. Convergence has three separate criteria, and all three belong in the report.
Boundary conditions: pressure or flow rate?
A wrong boundary condition renders the best mesh and the most appropriate model meaningless. What you specify at inlet and outlet follows from the physics.
Steady state or transient?
A transient solution costs an order of magnitude more. When is it genuinely required, and when does steady state suffice?
When do symmetry and periodicity assumptions mislead?
Halving the model halves the cost; sometimes it also invalidates the result entirely. Where is the line?
Cell type: hexa, tetra or polyhedral?
Element count alone is not a measure of quality. Cell type, alignment with the flow and boundary layer prisms matter more.
Conjugate heat transfer: when must you solve the solid?
Applying a heat transfer coefficient to a wall is easy. It is also, in most problems, assuming the answer you were looking for.
When can radiation not be neglected?
Radiation scales with the fourth power of temperature. The decision to neglect it is not a matter of feel but of a short comparison.
Rotating machinery: frozen rotor or sliding mesh?
A moving reference frame is cheap and sufficient at most operating points. What matters is knowing where it stops being sufficient.
Multiphase flow: which approach?
Free surface, droplets or bubbles? The choice follows from the volume fraction of the phases and the scale of the interface.
How many digits should a CFD result be quoted to?
The solver prints eight digits. How many of them mean anything is for the engineer to say, not the solver.
Why is natural convection harder than forced convection?
No fan does not make the problem easier — it makes it harder. The velocity field is born from the temperature field, and the two feed each other.
What does the Mach 0.3 rule actually say?
The familiar rule is about density change, not about the flow being incompressible in every respect. Where do you cross the line?
Moving boundaries: should the mesh deform or slide?
A valve opens, a piston travels, a flap folds. There are several ways to put motion into the model, and the choice depends on how much the geometry changes.
Solution will not converge: what to check, and in what order
Lowering the relaxation factors should not be the first reflex. Failure to converge is usually a setup or a physics problem, not a settings problem.
Verification and validation are not the same thing
One asks whether you are solving the equations right, the other whether you are solving the right equations. Confuse them and agreement with measurement becomes a coincidence.
Where does cooling actually fail: data centre or enclosure?
The problem is usually not cooling capacity but where the air goes. Bypass and recirculation consume capacity in systems that look adequate on paper.
Pump cavitation: why an NPSH margin of 1.1 is not enough
Satisfying NPSHa > NPSHr does not mean there is no cavitation. The definition of NPSHr is already a 3% loss of performance.
The operating point belongs to the system, not the fan
The flow rate in the catalogue is what that fan delivers against a particular resistance. You cannot select a fan without knowing the system; producing the system curve is what CFD is for here.
Heat exchangers: where does ε-NTU end and CFD begin?
The established method gives overall performance in minutes. Calling on CFD is worthwhile where that method's assumptions break down.
Thermal management and runaway in a battery pack
Heat generation is not a single number; it has reversible and irreversible parts that behave differently. And the acceptance criterion is not peak temperature but cell-to-cell spread.
Critical velocity in tunnel fires: how is a jet fan sized?
Critical velocity is the lowest ventilation velocity that stops smoke spreading back against the flow. Too much is as dangerous as too little.
What actually governs efficiency in a cyclone separator?
Cut size alone is not a sufficient measure. The shape of the efficiency curve, the particle size distribution and the loading together decide the outcome.
The limits of the Merkel method in cooling towers
A hundred-year-old method is still the backbone of the design. But the three assumptions it rests on do not always hold in modern towers.
Why air change rate is not enough for a cleanroom
Classification is defined by particle concentration, not by flow rate. Two rooms with the same air change rate can give very different results.
Why scale effect is a separate job in ship hydrodynamics
Model tests are run at Froude similarity while friction depends on Reynolds number. The two cannot be satisfied at once, and a procedure bridges the gap.
High-speed heat loads: why the stagnation point is critical
In the hypersonic regime the heat flux varies inversely with the square root of the nose radius. A sharp nose is aerodynamically good and thermally catastrophic.
Surrogate models: how many runs do you really need?
If every run is expensive, trial and error is the costliest method of all. A good surrogate gives the same information with an order of magnitude fewer runs.
Why does a first-order scheme blur the result?
Numerical diffusion behaves like real diffusion: it smooths gradients, damps vortices and smears the solution in a way that is not physical.
Pressure-velocity coupling: segregated or coupled solver?
In incompressible flow pressure has no equation of its own; it is derived from continuity. How that derivation is done sets the convergence behaviour.
Gradients and limiters: what happens on a poor mesh
In a cell-centred solver every flux rests on a gradient estimate. This is where mesh quality first enters the solution.
Does parallel partitioning change the result?
It should not. If it does, that is not a parallelisation bug but a sign that the solution has not converged.
RANS, URANS, DES, LES: which one, when?
The choice between modelling eddies and resolving them changes the cost by orders of magnitude. It follows from which scale of turbulence the answer depends on.
How do you read a y+ map?
A single average y+ says almost nothing. What matters is the range the distribution falls in, and in which region.
Legitimate ways to speed up convergence
Cutting the relaxation factors does not speed things up, it slows them down. Real acceleration comes from the initial condition, ramping and multigrid.
Porous media models: where do the coefficients come from?
Honeycombs, filters, packings, heat exchanger coils — all are represented as porous regions. The value of the model rests on where the coefficients came from.
Boundary layer mesh: how is the first cell height computed?
The first cell height is not guessed, it is calculated. The calculation is a few lines and sets the run up correctly from the start.
Numerical error or model error?
When the two act in opposite directions the result lands close to the measurement and the study looks successful. The only way to separate them is to work in sequence.
Heat transfer coefficient: which reference temperature?
h is not a material property but a definition. Unless the temperature it is defined against is stated, the number cannot be compared with anything.
Mixed convection: there is a fan, but buoyancy is there too
The range where forced and natural convection act together is harder than either. One dimensionless number tells you which regime you are in.
Thermal storage with phase change materials
Latent heat provides a large buffer, but as the melt front advances the resistance grows. The critical quantity is not stored energy but how fast it can be taken out.
Surface condensation: film or dropwise?
Dropwise condensation gives an order of magnitude better heat transfer than filmwise. What persists in the field, however, is usually film.
Boiling regimes and critical heat flux
Raising the heat flux improves heat transfer up to a point. Beyond it the surface is blanketed with vapour and the temperature jumps.
Thermal contact resistance: the one not in the catalogue
When two surfaces touch, heat does not pass without interruption. In assembly stacks this is often the governing term in the total resistance.
CFD to structural analysis: mapping temperature and pressure
The input to a thermal stress calculation is the temperature field. What mapping usually loses is the gradient — and the gradient sets the stress.
How is a heat pipe modelled?
Resolving the interior is rarely necessary, but an equivalent conductivity is not unlimited either. A heat pipe has a transport capacity, and beyond it performance collapses.
Insulation and thermal bridges: small area, large loss
Heat takes the easiest path. A metal fastening covering one percent of the surface can carry a far larger share of the total loss.
What changes in cryogenic systems?
As temperature falls, material properties stop being constant, radiation becomes unexpectedly important, and a small heat leak is a large loss.
Separation and reattachment: why is prediction so hard?
At a sharp edge the separation point is fixed and easy to predict. On a smooth surface the turbulence model decides it — and that is where the error begins.
Vortex shedding and the Strouhal number
The shedding frequency is given by a dimensionless number that stays nearly constant over a wide Reynolds range. That makes it the most useful check you have.
Secondary flow in bends: why the loss beats the catalogue
A bend does not only produce a local loss; it leaves a swirling flow at its exit, and that changes the performance of everything downstream.
Fluid-structure interaction: when is one-way enough?
If the structure's deflection does not change the flow, one-way mapping is sufficient and far cheaper. The criterion is the deflection relative to the flow geometry.
Vortex-induced vibration: avoiding resonance is not enough
Keeping the shedding frequency away from the natural frequency is the classical design rule. Lock-in shows that it leaves a narrower margin than you think.
Non-Newtonian fluids: viscosity is no longer a number
Paint, paste, slurry, polymers, food products and blood — in all of them viscosity depends on shear rate. An analysis run with a single value is wrong from the start.
Impinging jets: highest heat transfer, hardest prediction
An impinging jet gives the highest heat transfer per unit area. For the same reason, its local distribution is a classic weak point of turbulence models.
Sloshing: why is a half-full tank harder than a full one?
A free surface has its own natural frequency. If the excitation approaches it, a small motion produces a large load.
Ejectors: no moving parts, but not easy to design
An ejector works stably up to a certain back pressure; beyond that threshold the suction collapses abruptly. Finding that threshold is the real design work.
Diffusers: open it wider, you still recover no more
Pressure recovery rises as the section grows — until separation begins. Past that point, more angle means less recovery.
Combustion: is mixing the limit, or chemistry?
In most industrial burners the combustion rate is set by turbulent mixing, not by chemistry. That distinction tells you directly which model is appropriate.
Why is NOx prediction a separate job?
NOx depends exponentially on temperature and forms in quantities too small to affect the main flow. Those two properties make it a separate and far more sensitive calculation.
Flame stability in burners: why does the flame stay put?
What holds a flame in place is a recirculation zone that brings hot products back to the fresh mixture. That zone is the real subject of the design.
Film cooling: why does the hole angle matter so much?
The coolant has to form a blanket attached to the surface. Blow too hard and the jet lifts off, producing the opposite of what you wanted.
Heat distribution in boilers and reactors
The overall heat balance closes easily. What is hard is knowing how that heat is distributed over the surfaces — and that is what sets the life.
Wet steam and droplet erosion
Condensation begins in the last stages of a steam turbine. The droplets that form mean both an efficiency loss and blade erosion.
Thermoacoustic instability: why does a combustor hum?
If heat release and the pressure wave are in phase they feed each other. The result is an oscillation that can damage hardware within seconds.
What changes with hydrogen fuel?
In a system designed for natural gas, hydrogen is not simply another fuel. Its flame speed, temperature and density break several design assumptions at once.
Indoor comfort: temperature alone is not enough
Comfort is the joint result of six variables. A space at the right temperature but the wrong air velocity will keep generating complaints.
Hoods: increasing the flow rate does not increase capture
Capture depends far more on geometry and on surrounding air movement than on exhaust flow. A door in the wrong place defeats the strongest fan.
The hardest particle for a filter is not the smallest one
The efficiency curve has a minimum, and classification is defined at exactly that point — which is not where most people assume it is.
Wind and pedestrian comfort: corner acceleration
A tall building does not merely block the wind; it accelerates it. The velocity at ground level can exceed the free-stream value.
Pollutant dispersion: why is stack height not enough?
If the stack is not tall enough, the plume is captured by the recirculation zone behind the building and brought down. What matters is not absolute height but position relative to that zone.
Smoke extraction in atria and car parks
The aim is not to remove the smoke but to preserve a clear layer for the duration of the evacuation. That is about stratification more than about exhaust flow.
Heat pump outdoor units: breathing your own exhaust
The unit's performance is not the catalogue figure but the figure at the temperature of the air it actually draws. Siting sets that temperature.
Drying: is the limit the surface or internal diffusion?
In the first stage of drying the air sets the rate; in the second, the product does. Increasing the air flow in the second stage achieves nothing.
Air distribution inside a refrigerator
The average cabinet temperature can be right while the spread between shelves sets the product life. And a full cabinet is not the same cabinet as an empty one.
Why is even baking hard in an oven?
Heat arrives by three mechanisms at once and each dominates somewhere different. Even baking means balancing all three distributions together.
Duct noise and pressure loss are two faces of the same thing
Turbulence generates noise, and it generates pressure loss. Geometric improvements that reduce the loss usually reduce the noise as well.
Car park ventilation in normal operation
The fire scenario and daily operation are two different design questions on the same system. The criterion in daily operation is exhaust gas concentration.
Compressor surge: what happens left of the peak?
Pressure rises as flow falls — up to a point. Left of the peak the system is unstable and the flow can reverse.
Turbochargers: the gap between the map and real operation
Compressor and turbine maps are measured in steady flow. An engine, however, sends pulsating flow to the turbine, and the difference is not negligible.
Cooling an electric motor: put the losses in the right place
A motor's heat does not come from one place. Winding, iron and magnet losses arise in different places and proportions at different operating points.
Oil flow and the hot spot in a power transformer
A transformer's life is set not by the mean oil temperature but by the hottest spot in the winding. The difference depends on how the oil flows.
Vehicle aerodynamics: where should the drag be assigned?
A total drag coefficient is one number and gives no guidance for improvement. Without a breakdown you do not know where to intervene.
Brake and exhaust heat: protecting the neighbours
What the hot component does to the parts around it matters as much as its own temperature. And when the vehicle stops, the cooling stops too.
Tunnel entry pressure wave in rail systems
A train entering a tunnel generates a pressure wave. The wave travels along the tunnel, produces a bang at the far portal and disturbs passenger comfort.
Propeller and rotor noise: two different sources
Tonal noise comes from blade passing, broadband noise from turbulence. They have different causes and call for different remedies.
Flow distribution plates in fuel cells
In a stack of hundreds of cells the weakest cell sets the performance. The problem is usually not the chemistry but the distribution of reactant and water.
Electric vehicles: one circuit, many consumers
Battery, motor and cabin want different temperatures. Once they share a circuit, the design becomes a compromise rather than an optimisation.
Mixing time and dead zones in a stirred tank
Increasing the impeller power does not shorten the mixing time proportionally. If there is a dead zone, it does not shorten it at all.
Water hammer: how fast can you close a valve?
The pressure rise reaches its maximum if the closing time is shorter than the wave's round trip. That time is set by the length of the pipe.
Valve characteristics: installed curve is not inherent
The flow characteristic in the catalogue is how the valve behaves alone. Installed in a system the curve changes, and control quality suffers.
In a spray, droplet size decides everything
Evaporation, penetration and coating quality all follow from the size distribution. Working with a mean diameter erases that information.
Fluidised beds: are bubbles unavoidable?
The bed mixes well because of the bubbles. But those same bubbles let gas escape without contacting the solids.
Dust explosion venting: where will the pressure go?
If combustible dust ignites in an enclosed volume the pressure rises in under a second. The design has to keep that pressure below the equipment's strength.
Water distribution in washing machines and dishwashers
Cleaning performance is set not by total water or pressure but by the mechanical action reaching the surfaces. Any surface left in shadow stays dirty.
Geometry preparation: what to remove and what to keep
Simplification takes the largest share of the time and produces the most silent errors. The criterion is the detail's effect on the flow or the heat path.
Sensitivity analysis: which input should I know better?
There is no budget to measure every input more precisely. Sensitivity analysis tells you which one is worth measuring.
Can an analysis be repeated six months later?
What has to be archived is not the report but the whole chain that produced the result. An analysis that cannot be repeated cannot be defended.