Extended disjoint principal-components regression analysis of SAW vapor sensor-array responses.
Zellers-ET; Pan-TS; Patrash-SJ; Han-M; Batterman-SA
Sens Actuators B Chem 1993 Apr; 12(2):123-133
Chemical vapors were monitored with extended disjoint principal components regression (EDPCR) methods using an array of polymer coated surface acoustic wave (SAW) sensors. Ten vapors were examined, including dimethylmethylphosphonate (756796) (DMMP), chloroethylene (75014), diethyl-sulfoxide (70291), toluene (108883), and 1-butanol (71363). Principal component analysis (PCA) was applied to sensor responses. The relative errors were low (less than 8%) in all cases, but the absolute error for DMMP was significantly larger than those for other vapors. Most sensor responses were linear functions of vapor concentrations, but exceptions were noted for all vapors. Deviations from linearity were especially noted for DMMP, with seven of ten regression coefficients below 0.99. Between group errors were larger than within group errors. However, for DMMP the errors obtained while classifying it into the other groups were invariably smaller than those for other vapors. EDPCR identified 36 of 45 unknowns from binary mixtures of vapors. Most misclassifications involved DMMP. A final analysis included all possible ternary mixtures, and gave correct classification rates of 67, 71, and 85% for unconstrained, partially constrained, and fully constrained conditions, respectively. Misclassifications involved DMMP. The authors conclude that the EDPCR method is an alternative to other methods for analyzing sensor array responses, and is particularly suited for SAW sensors with linear responses to individual vapors and additive responses to mixtures.
NIOSH-Grant; Grants-other; Vapor-detectors; Organic-vapors; Acoustic-signals; Statistical-analysis; Monitoring-systems; Vapor-detectors
756-79-6; 75-01-4; 70-29-1; 108-88-3; 71-36-3
Other Occupational Concerns; Grants-other
Sensors and Actuators B: Chemical
University of Michigan at Ann Arbor, Ann Arbor, Michigan