A method for determining the optimal set of polymer sensor coatings to include in a surface acoustic wave (SAW) sensor array for the analysis of organic vapors is described. The method combines an extended disjoint principal components regression (EDPCR) pattern recognition analysis with Monte Carlo simulations of sensor responses to rank the various possible coating selections and to estimate the ability of the sensor array to identify any set of vapor analytes. A data base consisting of the calibrated responses of 10 polymer-coated SAW sensors to each of six organic solvent vapors from three chemical classes was generated to demonstrate the method. Responses to the individual vapors were linear over the concentration ranges examined, and coatings were stable over several months of operation. Responses to binary mixtures were additive functions of the individual component responses, even for vapors capable of strong hydrogen bonding. The EDPCR-Monte Carlo method was used to select the four-sensor array that provided the least error in identifying the six vapors, whether present individually or in binary mixtures. The predicted rate of vapor identification (87%) was experimentally verified, and the vapor concentrations were estimated within 10% of experimental values in most cases. The majority of errors in identification occurred when an individual vapor could not be differentiated from a mixture of the same vapor with a much lower concentration of a second component. The selection of optimal coating sets for several ternary vapor mixtures is also examined. Results demonstrate the capabilities of polymer-coated SAW sensor arrays for analyzing of solvent vapor mixtures and the advantages of the EDPCR-Monte Carlo method for predicting and optimizing performance.

The influences of temperature and atmospheric humidity on the performance of an array of eight polymer-coated 158-MHz surface acoustic wave vapor sensors were investigated. Sensitivities to the seven organic vapors examined all exhibited negative Arrhenius temperature dependencies, with responses increasing by factors of 1.5-4.4 on going from 38 to 18 degrees C. The magnitudes of the temperature effects, while generally similar, differed sufficiently among certain sensor-vapor combinations to cause marked changes in vapor response patterns. In addition, it was found that operating identically coated sensors at different temperatures could provide a means for discriminating certain vapors. The changes in sensor responses with temperature agreed reasonably well with those expected assuming ideal vapor sorption behavior and indicated that changes in the moduli of the sensor coatings were not important mediating factors. Responses to relative humidity (RH) from 0 to 85% RH were important even for the nonpolar sensor coatings. Significant changes in the sensitivities to the organic vapors were observed as a function of atmospheric humidity for several sensor-vapor combinations, which, in turn, affected the patterns of responses obtained from the sensor array. Results indicate that small changes in temperature or humidity have a larger effect on baseline stabilities than on the responses to the vapors. Monte Carlo simulations of sensor responses show that the ability to discriminate vapors in binary and ternary mixtures using a four-sensor array remains high regardless of the operating temperature and ambient humidity, provided that temperature-or humidity-induced changes in the response patterns are taken into account.

The application of a disjoint principal-components regression method to the analysis of sensor-array response patterns is demonstrated using published data from ten polymer-coated surface-acoustic-wave (SAW) sensors exposed to each of nine vapors. Use of the method for the identification and quantitation of the components of vapor mixtures is shown by simulating the 36 possible binary mixtures and 84 possible ternary mixtures under the assumption of additive responses. Retaining information on vapor concentrations in the classification models allows vapors to be accurately identified, while facilitating prediction of the concentrations of individual vapors and the vapors comprising the mixtures. The effects on the rates of correct classification of placing constraints on the maximum and minimum vapor concentrations and superimposing error on the sensor responses are investigated.

No abstract

Abstract