Disturbance Variables, Drift and Experiment Repeatability

The quality of the model heavily depends on the quality of the measuring data. During the measurement, e.g. of an engine, disturbance variables such as engine, oil and charge air temperature in addition to a time drift of the measuring arrangement itself can negatively affect the measurement.

For this reason, repetition points should be measured for every measurement series and, in particular, at defined operating points and in such a way that they cover the entire time range of a measurement (start, middle, and end of measurement).

On the one hand, these measurements can be used to determine the experiment reproducibility (RMSE) – no model can be more precise without improving the measurement technique.

On the other hand, it allows identifying a disturbance variable-based drift that can be identified by means of the measuring time or measuring point number. In that case, drift effects could be corrected by inserting the disturbance variable in the model. But a requirement for the disturbance variable correction is that the disturbance variable does not feature any correlation to other model parameters (no "sorted" parameters).