ANYL 273 |
| Christine M Wehlburg1, Robert Guenard2, Randy Pell3, David K. Melgaard, and David M. Haaland4. (1) Chemical and Biological Sensing, Imaging and Analysis, Sandia National Laboratories, 1515 Eubank SE, Albuquerque, NM 87123, (2) Molecular Structure, The Dow Chemical Company, 2301 N. Brazosport Blvd., B-1219, Freeport, TX 77541-3257, (3) Modeling and Automation, The Dow Chemical Company, 1897s, Midland, MI 48667, (4) IDEA Department, Sandia National Laboratories, MS0886, PO Box 5800, Albuquerque, NM 87185-0886 |
| There are various experimental and algorithmic approaches for facilitating the transfer of calibration between two spectrometers. Many groups have reported successful efforts by collecting a subset of sample spectra on both instruments and utilizing piecewise direct standardization (PDS) to transfer the calibration. As a result, PDS has emerged as the benchmark transfer of calibration method. Our group has developed augmented classical least squares/partial least squares (CLS/PLS) algorithms that outperformed PDS in both inter- and intravendor transfer of calibration studies. Together with these results, we will present our findings regarding the various variables that affected each method including subset selection, model assessment and outlier detection criteria, and instrument drift. PDS is limited in its ability to transfer a calibration in the presence of instrument drift. The CLS/PLS algorithm can correct for multiple variations such as instrument differences and instrument drift and should be considered the next benchmark method for calibration transfer. |
|
Process Analytical Chemistry
8:30 AM-11:55 AM, Thursday, March 27, 2003 Convention Center -- Room 384, Oral
Division of Analytical Chemistry |