Use of classical least squares/partial least squares (CLS/PLS) hybrid algorithm for calibration and calibration maintenance of surface acoustic wave (SAW) devices

Many data analysis algorithms that are currently employed in SAW sensors lack the ability to easily maintain calibration models in the presence of unmodeled interferents or sensor drift. The classical least squares/partial least squares (CLS/PLS) hybrid algorithm is tested in this study for its ability to update calibration models for unmodeled interferents and sensor drift with information from only a single recalibration standard. Use of the CLS/PLS hybrid algorithm for calibration and calibration maintenance of surface acoustic wave (SAW) devices was investigated for synthetic mixtures of iso-octane-methanol-water and with synthetic mixtures of nerve agent analogs, di-iso-propyl methyl phosphonate (DIMP)-kerosene-water along with a true ternary mixture of dimethyl methyl phosphonate (DMMP)-kerosene-water. Calibration statistics using the hybrid algorithm were found to be as good as those obtained from a standard partial least squares (PLS) analysis. In prediction, the hybrid algorithm models were found to perform equivalently to PLS models in the absence of unmodeled interferents or sensor drift, with an accuracy of 5-10% of the reference values and a high degree of precision. In the case of prediction in the presence of unmodeled interferents and/or sensor drift, PLS models and prediction augmented CLS/PLS (PACLS/PLS) hybrid models were compared using a single standard sample to update each model for prediction. For the cases studied, PACLS/PLS hybrid models were comparable to or outperformed updated PLS models that used subset recalibration or piece-wise direct standardization. (C) 2004 Published by Elsevier B.V.