Limits of recognition for binary and ternary vapor mixtures determined with multitransducer arrays

The discrimination of simple vapor mixtures from their components with polymer-coated multitransducer (MT) arrays as a function of the absolute and relative concentrations of those components is explored. The data set consists of calibrated responses to 11 organic vapors from arrays of 5 or 8 microsensors culled from a group of 5 cantilever, 5 capacitor, and 5 calorimeter transducers coated with 1 of 5 different sorptive-polymer films. Monte Carlo methods are applied to simulate error-enhanced composite responses to all possible binary and ternary mixtures of the 11 vapors, and principal component regression models are established for estimating expected rates of recognition as a function of mixture composition. The limit of recognition (LOR), defined as the maximum recognizable mixture composition range, is used as the metric of performance. With the optimal 8-sensor NIT array, 19 binary and 3 ternary mixtures could be identified (i.e., discriminated from their components) with <5% error. The binary-mixture LORs are shown to decrease with increases in the baseline noise levels and random sensitivity variations of the sensors, as well as the similarity of the vapors. Importantly, most of the binary LOR contours are significantly asymmetric with respect to composition, and none of the mixtures could be recognized with <5% error at component relative concentration ratios exceeding 20:1. Discrimination of ternary mixtures from their components and binary subcomponent mixtures is possible only if the relative concentration ratio between any two of the components is <5:1. In comparing binary LORs for the best five-sensor single-transducer (ST) array to those of the best five-sensor NIT array, the latter were larger in nearly all cases. The implications of these results are considered in the context of using such arrays as detectors in microanalytical systems with upstream chromatographic modules.