This work evaluates the usefulness of two linear chemometric algorithms, principle component regression and partial least-squares analysis, for modeling the responses of an evanescent fiber optic chemical sensor to aqueous mixtures of organicanalyses with individual concentrations ranging from 50 to 200 ppm. Two data sets were examined. One contained trichloroethylene, 1,1,2 trichloroethane, toluene, and chloroform. The second set contained these four analyses as well as tetrachloroethene.Both chemometric algorithms performed comparably on a given data set with cross-validated root mean squared errors of prediction (RMSEP) for trichloroethylene, 1,1,2 trichloroethane, toluene, and chloroform of approximately 6, 9, 6, and 16 ppm from thefirst set and 7, 11, 13, and 31 ppm from the second set with tetrachioroethene RMSEP of 31 ppm. The decrease in the quantitative performance of the algorithm for modeling toluene and chloroform upon addition of tetrachloroethene to the sample solutions is due to increased intensity of cladding absorption features in the spectral response matrix. These features overlap with the analyze absorption features of toluene and chloroform and reveal one of the limitations with this type of sensing format.
展开▼
