Engineering & Technology · Fluid Mechanics, Flow Systems & Machines

Flow measurement, computational models and validation

A meter reading and a colourful simulation can disagree for several reasons: the meter may measure a different quantity, the numerical solution may still depend on its grid, or the physical model may omit an important effect. A useful investigation separates these possibilities.

Review force vectors, energy and power, control volumes and steady-flow thermodynamics, SI units and measurement uncertainty. Be comfortable with ratios, areas, square roots and reading graphs. Symbols are defined where used: ρ is density, μ dynamic viscosity, Q volume flow and ṁ mass flow. For the constructed liquid cases, use g = 9.81 m/s² where stated. Prior physics study of fluids is helpful but not required.

Back to the chapter

After reading, you can

Choose a flow measurand and distinguish calibration, numerical verification and physical validation.

Define what is being measured

A flow measurand might be mass flow, volume flow at specified conditions, or velocity at one location. A local velocity sensor needs a characterized profile or a suitable traverse to establish area-mean velocity. An ultrasonic, differential-pressure or other meter also has installation and calibration conditions. Pressure taps and probes must distinguish static pressure from the pressure obtained by bringing a flow toward rest.

A gravimetric reference obtains mass flow from collected mass and elapsed time, with corrections and uncertainty appropriate to its method. For constant density ρ, average volume flow is Δm/(ρΔt). Density may need temperature and pressure information. Leakage, collection timing, evaporation, scale calibration and forces on the weighing system can matter. A simplified teaching calculation does not inherit the uncertainty of a national metrology standard.

Calibration and validation need independent evidence

Calibration establishes a relation between indications and reference values under stated conditions. It does not remove all uncertainty or guarantee every fluid, range and installation. Repeated tests reveal variability; independent conditions can reveal a bias that a single fitted point hides. Record the reference, conditions, uncertainty and the correction actually used.

Computational fluid dynamics, or CFD, approximates conservation equations over a spatial grid and time steps, together with boundary conditions and physical models. Verification asks whether the equations and their numerical implementation are being solved accurately enough. It includes conservation checks, iterative convergence and grid or time-step sensitivity. A small iterative residual alone cannot demonstrate a sufficiently accurate physical prediction.

Converged calculations can still model the wrong physics

Validation compares predictions with independent observations for an intended use and range of conditions, accounting for experimental and computational uncertainty. Parameters fitted using a dataset should be tested against other appropriate evidence; fitting and validation answer different questions. A predicted flow field must also be compared using the same locations, averaging conventions and quantity as the experiment.

The diagram shows constructed predictions changing as the grid becomes finer. The smallest change between runs is not automatically the remaining numerical error. Even a stable numerical limit can differ from physical reality because of turbulence closure, geometry or boundary assumptions. A defensible report states what was checked and where uncertainty or disagreement remains.

Horizontal axis: normalized grid spacing h/href, 0 to 1.1, dimensionless. Vertical axis: predicted or reference volume flow, 0.28 to 0.38 L/s. Blue predictions are 0.360 at spacing 1, 0.330 at 0.5, and 0.315 L/s at 0.25. The dashed orange reference is 0.300 L/s; its green uncertainty band runs from 0.290 to 0.310 L/s. The finest prediction remains outside that band. Smaller changes between grids do not establish physical validity.

Constructed calibration and refinement evidence

A simplified reference collects 6.00 kg in 20.0 s at assumed constant density 1,000 kg/m³. Average mass flow is 0.300 kg/s and volume flow 0.000300 m³/s = 0.300 L/s. A meter indicating 0.312 L/s is 0.012 L/s high, or 4% of this reference value. A one-point multiplicative factor would be 0.300/0.312 ≈ 0.9615; it needs independent checks before application across a range.

For a separate hypothetical CFD comparison of that nominal flow, grid spacings normalized as 1, 0.5 and 0.25 give predictions 0.360, 0.330 and 0.315 L/s. Suppose independent reference evidence is 0.300 ± 0.010 L/s, with the stated expanded uncertainty and coverage factor k = 2. The finest prediction is still outside its 0.290–0.310 L/s interval. The result does not establish agreement or a complete discrepancy budget: remaining numerical, model and input uncertainties still need assessment.

Try the reasoning

Why might a centreline velocity multiplied by pipe area overestimate flow?
Show the answer and reasoning
Centreline velocity can exceed the area mean. Volume flow requires the area integral of normal velocity, or a justified conversion to the mean. The assumed profile must match the regime and conditions.
Does a flow prediction inside the experimental interval prove universal model validity?
Show the answer and reasoning
No. It supports a particular comparison subject to all relevant uncertainties. Error cancellation, limited conditions or a wrong measurand can hide problems; validation has a defined scope.
Why retain the coarse-grid results after obtaining a finer answer?
Show the answer and reasoning
Their differences show numerical sensitivity and help assess convergence behaviour. Keeping them makes the calculation’s evidence inspectable rather than presenting the finest value as automatically accurate.

What to carry forward

Match the measurands, preserve independent evidence, and separate meter calibration from numerical and physical model checks.

References

Teaching resources and research supporting this lesson’s explanations and examples.