MULTIVARIATE INSTRUMENT STANDARDIZATION

Three closely related problems associated with using multivariate calibration methods in spectroscopy have surfaced recently. The first problem involves the desire to transport a calibration model developed on one instrument to a second or even multiple instruments. Differences between the primary and secondary instruments which can occur for a variety of reasons can lead to erroneous results, thereby prohibiting transferring the calibration model and necessitating the transport of the calibration samples. The second problem occurs when instruments change over time (e.g. wavelength shift) for any reason. Again, using a calibration model for analysis when the instrument responses are altered after the time calibration was performed is problematic. The third problem is caused by the variation between samples from different production batches. The calibration model built from one batch might not be applicable to another batch. Using the mathematics of multivariate calibration, four different approaches to solving these two problems have been derived and tested with computer simulation. The four standardization methods proceed by acquiring the spectra of a well-chosen subset of the calibration samples and then either correcting the primary calibration model for use on secondary instruments or correcting the spectra acquired on the secondary instrument to account for the response differences. While standardization does not outperform using the entire calibration set for recalibration, only a 1.2-1.6 times larger error is obtained by standardization in the simulations and a study of a near-infrared data set.