Two kinds of error

Random error makes repeated runs scatter. It shows up when the same test is repeated, and it shrinks when results are averaged.

Systematic error shifts every run the same way. A sensor that reads half a degree high does so every time. Repeating does not reveal it. Checking against a reference does.

Where spread comes from

SourceTypical control
Sensor accuracyComparison check before each series
Sensor contact and positionFixed, marked, photographed positions
Ambient driftLogged, and runs rejected if it moves too far
Start temperatureRest period before each run
Sample variationMore than one sample where available
Camera emissivity settingSet and recorded for each surface
Assembly of the layer stackSame order, same compression

Repeat runs

For any result that drives a decision, the run is repeated at least three times, with the stack taken apart and rebuilt between runs. The range across repeats is the practical repeatability of that test.

Is the difference real?

If element A ends 1.5 °C warmer than element B, and repeats of the same element scatter by 2 °C, the test has not separated them. We say so. A difference is reported as real only when it clearly exceeds the repeatability.

Absolute and relative

Our methods are stronger at comparison than at absolute values. Comparing two samples in the same setup cancels most systematic errors, because both sides share them. Where an absolute figure is needed against a standard, an accredited laboratory using that standard's apparatus is the right source, and we say that too.

How figures are written

  • Temperatures to one decimal, in °C with °F in brackets.
  • Energy in watt-hours to one decimal.
  • Times as hours, minutes and seconds.
  • Every figure with its conditions and, where it matters, its repeatability.

A number without conditions is not a result. See reading a thermal report.

Put the lab to work.

Send a brief. A mutual nondisclosure agreement comes first.

Start a program