3. Conclusion

In this article, we have seen that the robustness of calibrations takes on a particular character in the highly multivariate case of continuous signals, such as spectra. The very large size of the measurement space leaves considerable scope for the expression of influence quantities, if no precautions are taken. The strategy for dealing with this problem depends on the controllability and measurability properties of the influence variable.

In cases where the value of the influence variable is available at the time the calibration model is used, three methods have been described: a priori correction, which consists in modifying the measurement to match the calibration conditions; model correction, which consists in adapting the model to the new conditions; a posteriori correction, which consists in correcting the model outputs.

When the value...