Begin with a decision and a measurement
A simulation should answer a specific question. “What will happen?” is too broad. “Will the liquid reach sensor A within 12 seconds under these starting conditions?” defines a measurable result and a success criterion.
Useful outputs can include arrival time, displacement, flow rate, pressure, temperature, interface position, trapped-gas volume, or sensor response. A visualization can help explain those values, but it still needs a result that can be measured.
Verification and validation answer different questions
Verification asks whether the equations and numerical methods were implemented and solved correctly enough for the intended calculation. Validation asks how well the model represents physical observations for the intended use.
NIST’s review of verification, validation, and uncertainty quantification distinguishes code verification, solution verification, and validation against the physical world. NASA’s modeling-and-simulation standard also requires attention to inputs, verification, validation, uncertainty, sensitivity, limits, and reporting.
A model can solve its chosen equations accurately and still omit important physics. It can also represent the right physics but produce unreliable results because of numerical error. Both problems must be examined.
Reserve data for testing
If every experimental record is used to tune a model, agreement with those same records says little about how it will perform on a new case. A stronger check reserves some data for testing.
The held-out cases should be chosen before results are inspected when possible. They should include meaningful changes in geometry, initial condition, material property, or environment. A failure is informative if it exposes a missing mechanism or a narrow operating range.
Ground Blue will keep tuning records separate from test records so the comparison can be repeated.
Report uncertainty and sensitivity
Inputs are rarely exact. Material properties vary. Contact angles can depend on preparation and history. Sensors have error. Initial conditions may be imperfectly known. Numerical resolution also contributes uncertainty.
A useful result should include a range when the available data supports one, along with the inputs used to produce it. Sensitivity analysis shows which variable changes the answer most and can guide the next design change or measurement.
Uncertainty is not a decorative confidence score. It should be connected to identified sources of error and the decision being made.
Separate interpolation from extrapolation
Interpolation estimates behavior within a range covered by reference data. Extrapolation reaches beyond that range. Geometry, materials, and the balance of physical forces all affect where the boundary lies.
For example, a model checked on several capillary geometries at low Bond number may fail for a larger container where gravity becomes important. The software should warn when the materials, geometry, or operating conditions fall outside the tested range.
A focused model with visible limits is more useful than a broad simulator that hides where its results become unreliable.
Make the result repeatable
Another person should be able to reproduce the result from the model version, inputs, numerical settings, and output definition. Changes to the model or tuning data should create a new version rather than silently replacing earlier results.
NASA’s modeling guidance covers documentation, input sources, numerical error, robustness, verification, and comparison with physical tests. Those records show how a result was produced and where it should be used.
What every result should include
Each Ground Blue simulation should include:
- a structured design and environment definition;
- the model and version used;
- omitted effects and simplified inputs;
- a measurable output and acceptance criterion;
- numerical and input error ranges;
- the tested operating range;
- calibration records kept separate from held-out tests;
- the measured result when a physical test exists;
- the error and any model update made afterward.
This gives another engineer enough information to reproduce the calculation and design a useful test.