| Popis: |
Odorous air samples collected from diversified sources were presented to the olfactometry panel,an electronic nose, and hydrogen sulfide (H2S) and ammonia (NH3) detectors. The measurementsof the olfactometry (odor concentrations) were used as the expected values while measurementswith other instruments as input variables. Both linear regression and artificial neural networks(ANN) are used to test the multiple hypotheses made for surrogating the olfactometrymeasurements. Principal component analysis (PCA) is utilized to reduce the dimensionality ofthe electronic nose measurements from 35 to 3 without significant loss of information. Thedimensionally reduced datasets were used to train an ANN. As an individual instrument, anelectronic nose or H2S detector alone can predict odor concentration measurements with similaraccuracy (R2=0.46 and 0.51). Although NH3 detector alone has a very poor relationship with theodor concentration measurements, the integration of H2S detector and NH3 detector can predictodor concentrations more accurately (R2=0.58) than either individual instrument. Data from theintegration of the AromaScan electronic nose, a H2S detector, and a NH3 detector produced thebest prediction of odor concentrations with a regression determination coefficient of R2=0.75.With this accuracy, odor concentration measurements can be confidently surrogated with theintegration of the AromaScan electronic nose, H2S detector, and NH3 detector. |
