Differentiation of chemical components in a binary solvent vapor mixture using carbon/polymer composite-based chemiresistors

We demonstrate a "universal solvent sensor" constructed from a small array of carbon/polymer composite chemiresistors that respond to solvents spanning a wide range of Hildebrand solubility parameters. Conductive carbon particles provide electrical continuity in these composite films. When the polymer matrix absorbs solvent vapors, the composite film swells, the average separation between carbon particles increases, and an increase in film resistance results, as some of the conduction pathways are broken. The adverse effects of contact resistance at high solvent concentrations are reported. Solvent vapors including isooctane, ethanol, diisopropylmethylphosphonate (DIMP), and water are correctly identified ("classified") using three chemiresistors, their composite coatings chosen to span the full range of solubility parameters. With the same three sensors, binary mixtures of solvent vapor and water vapor are correctly classified; following classification, two sensors suffice to determine the concentrations of both vapor components. Poly(ethylenevinyl acetate) and poly(vinyl alcohol) (PVA) are two such polymers that are used to classify binary mixtures of DIMP with water vapor; the PVA/carbon particle composite films are sensitive to less than 0.25% relative humidity. The Sandia-developed visual-empirical region of influence (VERI) technique is used as a method of pattern recognition to classify the solvents and mixtures and to distinguish them from water vapor. In many cases, the response of a given composite sensing film to a binary mixture deviates significantly from the sum of the responses to the isolated vapor components at the same concentrations. While these nonlinearities pose significant difficulty for (primarily) linear methods such as principal component analysis, VERI handles both linear and nonlinear data with equal ease. In the present study, the maximum speciation accuracy is achieved by an array containing three or four sensor elements, with the addition of more sensors resulting in a measurable accuracy decrease.