How to compare simulation and experimental results

To compare simulation and experimental results across engineering papers, build one record per comparison. Capture the response quantity, unit and normalization basis, geometry, boundary conditions, measurement uncertainty, numerical uncertainty, and validation metric. Compare values only after those fields match closely enough for the review question.

Agreement is meaningful only after the two results answer the same physical question.

Build a comparison record before reading the conclusion

Start with the methods, figures, appendices, and data tables. The conclusion may call a model accurate without stating which response was validated, how uncertainty was handled, or whether the experiment was independent of model calibration. A structured record forces those details into view.

1. Define the physical quantity precisely

A label such as “temperature,” “pressure drop,” or “efficiency” is not precise enough. Record the measurand: the quantity intended to be measured. For temperature, that may be a point value at a sensor, an area average, a volume average, a time average, or the maximum value over a cycle. Those quantities can differ even when they share the same unit.

The BIPM guides in metrology point to the international Guide to the Expression of Uncertainty in Measurement and its treatment of a well-defined measurand. For a review, the practical rule is simple: write the quantity definition beside every extracted value. If the paper does not define it, mark the field unclear instead of guessing from a figure label.

2. Separate verification from validation

Verification and validation answer different questions. ASME describes verification as checking whether the computational model fits its mathematical description. Validation asks whether the model represents the real application. Grid convergence, solver tolerance, and code checks support verification. Agreement with a physical experiment supports validation.

Do not merge both into one “validated” column. Record numerical verification, experimental comparison, and uncertainty quantification separately. A paper may have a careful mesh study but a weak experiment. Another may match one test curve without reporting numerical error. Those are different evidence profiles.

3. Normalize units and the reporting basis

Unit conversion is the easy part. The reporting basis is where many comparisons fail. Heat transfer can be reported as total heat rate, heat flux, or a coefficient normalized by area and temperature difference. Battery performance may be normalized by cell, module, mass, volume, or stored energy. A percentage error may use the experiment, simulation, or an average as its denominator.

Keep the published value unchanged in a source column. Add a second normalized column with the conversion equation, original unit, target unit, and basis. This preserves the evidence while making the comparison auditable. Never overwrite the source value with a converted number.

4. Reconstruct the operating point

Two results at the same nominal load may use different boundary conditions. Extract the inputs that control the response: geometry, material properties, inlet conditions, ambient state, load history, initial condition, control logic, and measurement location. Add discipline-specific fields when they change the interpretation.

For a battery cold plate, flow rate alone is insufficient. Coolant composition, inlet temperature, channel geometry, contact resistance, heat generation model, discharge profile, and sensor placement can all alter the reported maximum temperature. If critical conditions differ, compare trends or mechanisms rather than treating the final values as replications.

5. Read the uncertainty statement, not just the error bar

An interval needs a definition. Record whether the paper reports a standard deviation, standard error, confidence interval, combined standard uncertainty, or expanded uncertainty. Also record the coverage factor or confidence level when given. These quantities are not interchangeable.

NIST Technical Note 1297 specifies that expanded uncertainty should be reported with its coverage factor, or that combined standard uncertainty should be reported. It also calls for the uncertainty components and their evaluation methods to be described. In a review, preserve those details rather than reducing every interval to “± error.”

For simulations, look for model-form uncertainty, input uncertainty, and numerical uncertainty. ASME’s VVUQ 10.2 overview identifies these sources and explains that validation metrics incorporate uncertainty from simulations and experiments. Do not interpret a prediction gap without asking what uncertainty surrounds both values.

6. Work through one engineering comparison

Suppose three papers evaluate cooling methods for a lithium-ion battery module. Paper A reports the simulated maximum cell temperature. Paper B reports the highest thermocouple reading. Paper C reports a volume-averaged temperature. All three use degrees Celsius, but only A and B might answer a comparable question.

First align the discharge rate, inlet temperature, geometry, coolant flow, contact assumptions, and temperature location. Then preserve Paper B’s sensor uncertainty and Paper A’s mesh and input sensitivity. If A predicts 42.1 °C and B reports 43.0 ± 1.2 °C at the matched location and operating point, the observed gap is 0.9 °C. That arithmetic does not prove the model is valid. It creates a transparent comparison that can be judged alongside the validation design and uncertainty statements.

Paper C still contributes evidence. Use it for temperature distribution or energy-balance trends if those match the review question. Do not force its average into a table of maxima. A useful review preserves non-comparable results and explains why they were synthesized differently.

7. Write the synthesis from the comparison fields

A strong synthesis names the condition and the evidence boundary. Write which responses agreed, at which operating points, under which uncertainty treatment, and where comparison was not possible. Avoid ranking models by the smallest reported percentage error when the papers use different quantities or denominators.

Use a sentence pattern that keeps the qualification attached:

At the matched 3C discharge condition, the simulation and point measurement differed by 0.9 °C; the experimental interval was ±1.2 °C, while the paper reported mesh sensitivity separately. The comparison supports agreement at that sensor location, not validation of the full temperature field.

In Rivul, keep the paper and exact source passage beside each comparison note, then use the evidence matrix method to group studies by response quantity and operating condition. The researcher still decides whether the evidence is comparable and whether the validation claim is justified.

Final comparison checklist

• The simulation and experiment use the same physical quantity and location.
• Units, normalization basis, and error denominator are explicit.
• Geometry, material properties, boundary conditions, and load history are aligned.
• Experimental and numerical uncertainty are recorded without collapsing unlike intervals.
• Verification evidence is separate from physical validation evidence.
• The synthesis states where comparison is valid and where it is not.