Mesh Quality and Why More Cells Are Not Always Better

The computational domain must be divided into a mesh before the governing equations can be solved numerically.

Mesh resolution affects the ability of the model to capture important physical features.

Areas that may require additional resolution include:

• boundary layers
• narrow passages
• sharp geometric changes
• wakes
• separated flow
• regions with strong temperature gradients

However, increasing mesh size indefinitely is not a substitute for good modelling.

The important question is whether the predicted engineering quantity has become sufficiently insensitive to further mesh refinement.

For example, a mesh study might produce:

Mesh	Cells	Predicted pressure drop
Coarse	0.8 million	91 Pa
Medium	1.7 million	87 Pa
Fine	3.5 million	86 Pa
Very fine	6.8 million	85.7 Pa

The useful information is not that the last mesh contains the most cells.

The useful information is that the predicted pressure drop is becoming relatively insensitive to further refinement.

For the pressure-drop quantity of interest, the change from the fine mesh to the very fine mesh is only:

(86.0 − 85.7) / 86.0 × 100 ≈ 0.35%

That provides evidence that further refinement is having a relatively small effect on the selected engineering result.

Mesh independence does not mean that every aspect of the flow is perfectly resolved. It means that the chosen quantity of interest has become sufficiently stable for the purpose of the analysis.

Turbulence Modelling Requires Engineering Judgement

Most practical engineering flows are turbulent or contain regions of transitional behaviour.

Resolving every turbulent scale directly is generally computationally impractical for industrial design problems. Instead, turbulence models are used to represent the effects of unresolved turbulent motion.

Common approaches include:

• Reynolds-averaged Navier–Stokes, or RANS
• k–ε models
• k–ω models
• SST formulations
• transitional models
• Large Eddy Simulation, or LES

There is no universally correct turbulence model.

The appropriate approach depends on the physics being investigated, the geometry, the expected flow regime and the required level of fidelity.

For many industrial applications, RANS-based methods provide a practical balance between computational cost and predictive capability.

For highly unsteady flows, strong separation or other complex phenomena, a more advanced modelling approach may be justified.

The important point is that model selection should follow the engineering problem, not the other way around.

A Converged Solution Is Not Automatically a Correct Solution

This distinction is essential.

A CFD solver can reach convergence while the model still contains unrealistic assumptions.

There are three different questions:

Has the numerical solution converged?

The calculated quantities and residuals have stabilised sufficiently.

Has the numerical problem been solved correctly?

The discretisation, mesh and numerical method behave as intended.

Does the model represent the physical system adequately?

The model predictions agree sufficiently with physical behaviour for the intended purpose.

The last question is validation.

Where suitable experimental information is available, CFD results should be compared with measurements such as:

• pressure
• temperature
• flow rate
• velocity
• system performance

The purpose is not to make the simulation “look right”.

It is to establish confidence in its ability to predict the behaviour that matters.

This distinction between verification and validation is important because numerical convergence alone cannot demonstrate physical accuracy.

Understanding the Results

A CFD project can generate a large amount of data.

Not all of it is equally useful.

Velocity

Velocity contours can identify acceleration, low-flow regions and potential flow restrictions.

But the important engineering question is why the velocity is high or low and whether that behaviour affects system performance.

Pressure

Pressure distribution can reveal where energy is being lost and help identify inefficient geometry.

Streamlines

Streamlines are particularly useful for understanding dominant flow paths, recirculation and bypass flow.

They should, however, be interpreted correctly. A streamline is a visual representation of the calculated velocity field, not a physical photograph of individual air particles.

Temperature

Temperature distributions show whether heat is being removed effectively and where thermal margins may be reduced.

Pressure drop

Pressure drop is often a key system-level metric because it directly influences required fan performance.

A useful simulation therefore focuses on the quantities that matter to the engineering decision rather than generating every possible contour available in the solver.

Sensitivity Analysis: Which Assumptions Actually Matter?

A single simulation represents one combination of inputs.

Real products do not operate at one perfectly fixed condition.

Ambient temperature changes. Fan speed changes. Heat generation changes. Manufacturing tolerances alter geometry. Filters become less permeable. Operating conditions vary.

Sensitivity analysis helps determine which parameters have the greatest effect on performance.

For an electronics cooling system, for example, the engineer might investigate changes in:

• fan flow rate
• ambient temperature
• component heat generation
• vent size
• vent location
• internal obstruction
• component placement

Suppose the maximum component temperature changes as follows:

Parameter	Lower condition	Baseline	Higher condition
Fan flow rate	80%	100%	120%
Maximum component temperature	78 °C	68 °C	62 °C

This simple comparison suggests that fan flow rate has a material influence on thermal performance.

A different parameter may have very little influence.

The objective is not simply to produce more simulations.

It is to identify which variables actually control the system.

That information can influence design margins, component specifications and operating limits.

Airflow Simulation Is Not a Replacement for Testing

Simulation and physical testing serve different purposes.

Testing tells us what a physical system does under specific conditions.

Simulation helps us understand why it behaves that way and allows alternative designs to be investigated without physically manufacturing every variation.

A strong engineering development process can therefore look like:

Concept → Simulation → Design optimisation → Prototype → Test → Correlation → Refinement

Where suitable test data are available, the simulation can be compared against reality.

For example, measured pressure loss might be compared with the CFD prediction:

CFD prediction: 86 Pa

Measured value: 89 Pa

The relative difference would be:

(89 − 86) / 89 × 100 ≈ 3.4%

The significance of that difference depends on the application, the measurement uncertainty and the intended use of the model.

The purpose of validation is therefore not to chase an arbitrary percentage difference. It is to understand whether the model is sufficiently predictive for the engineering decision being made.

When CFD Adds the Most Value

Airflow simulation is particularly valuable when:

• internal flow paths are difficult to visualise
• multiple competing flow paths exist
• thermal performance depends on local airflow
• pressure loss affects component selection
• geometry is complex
• experimental testing is expensive
• several design options need to be compared
• operating conditions vary
• the cost of finding a design problem late in development is high

It is less useful to perform an elaborate simulation simply because the software can do it.

The model should be proportionate to the engineering question.

A simple problem should not automatically become a million-cell CFD project.

Conversely, a complex three-dimensional flow problem should not be reduced to a single hand calculation merely because it is quicker.

From CFD Results to Engineering Decisions

The most valuable output of an airflow simulation is usually not the simulation file.

It is the decision that follows from it.

A useful CFD study might conclude that:

• an outlet is undersized
• a component is located in a recirculation region
• a fan is operating away from its intended point
• a manifold produces uneven distribution
• a pressure loss is concentrated in an avoidable restriction
• natural convection is insufficient at the required ambient temperature
• a small geometry change can significantly improve cooling

These are engineering conclusions.

They provide a direct link between computational analysis and product design.

How Twelvium Approaches Airflow Simulation

At Twelvium, airflow simulation is approached as an engineering problem rather than simply a CFD exercise.

The process begins with the question the design needs to answer.

From there, the relevant physics, boundary conditions, geometry and modelling approach are established. The computational model is then developed with appropriate mesh resolution and numerical methods, followed by verification of the solution and analysis of the quantities that matter to the design.

Where appropriate, sensitivity studies can be used to understand the effect of operating conditions and design variables.

The final objective is straightforward:

turn airflow behaviour into useful engineering information that can improve the design.

That may mean reducing pressure loss, improving cooling, increasing flow uniformity, selecting an appropriate fan, changing geometry or identifying a design limitation before it becomes a physical prototype problem.

Conclusion

Airflow simulation is most valuable when it answers a question that is difficult, expensive or uncertain to answer through intuition alone.

The engineering challenge is not simply to calculate velocity or temperature. It is to understand the physical mechanisms behind the result and determine what they mean for the design.

A credible CFD analysis therefore depends on more than a powerful solver.

It depends on the correct physical assumptions, appropriate boundary conditions, suitable turbulence and numerical models, adequate mesh resolution, verification, and where required, comparison against physical measurements.

When these elements are brought together, CFD becomes much more than a visualisation tool.

It becomes a method for reducing uncertainty, testing design decisions and understanding how a product or system will behave before the final hardware is built.