ONLINE DETECTION AND IDENTIFICATION OF INTERFERENCES IN MULTIVARIATE PREDICTIONS OF ORGANIC GASES USING FT-IR SPECTROSCOPY

One of the most serious problems that can occur when a multivariate mo del is used for the compositional analysis of an unknown mixture is th e presence of an unexpected constituent, not modeled in the calibratio n phase, The interferent will almost certainly influence the predicted concentrations of the modeled constituents, which leads to erroneous and, more seriously, misleading results. Usually, a recalibration-buil ding a new calibration model in which the interferent is included-will be necessary. However, in many applications of multivariate calibrati on, recalibration will be possible only if an unambiguous identificati on of the interferent can be made, In this paper, we describe how spec tral residuals resulting from a multivariate prediction can be used to detect and identify unknown interferents. The identification of the i nterferent is performed by matching the residual spectrum with a libra ry of residual spectra, This library was built by processing the membe rs of a regular spectral library by the calibration model and storing the resulting residual spectra. After successful identification, a str aightforward procedure can be used to correct the concentrations of th e modeled constituents, without the need for a recalibration, The meth ods are demonstrated using a relatively simple principal component cal ibration model to predict the concentrations of organic vapors and gas es in ambient air with FT-W spectroscopy. In addition, the influence o f different interferents on the predicted concentrations of the modele d constituents is described.