RESEARCH PROFILE - What's that smell?

摘要

The sense of smell is difficult to mimic. Researchers have tried to reproduce the biological response in the lab by creating artificial olfactory sensors for applications such as noninvasive chemical monitoring and the detection of toxic substances. But many solid-state sensors exhibit poor signal reproducibility because the responses drift significantly as the sensor ages. In addition, the conventional, all-at-once approach identifies an analyte by comparing it to the reference substances in a training data set and cannot generalize the training data to identify unknowns. In a recent AC paper (2008, DOI 10.1021/ac8007048), Joshua Hertz (now at the University of Delaware), Baranidharan Raman, Kurt Benkstein, and Steve Semancik of the U.S. National Institute of Standards and Technology tackled these problems by more closely mimicking nature and identifying general classes of odors first, before refining the identification to members within the class. "The analogy we like to use is when you first walk into a coffee shop, all you smell is coffee," says Hertz. "Only after a little while do you smell that they freshly baked bread that day--the fine points of the odor." For their method, the investigators produced extensive response databases and then subdivided the olfactory recognition task in a hierarchical fashion. Their new approach can be used for any sensor array, but the authors initially tested the method with a micro-hotplate array of 16 metal oxide sensors.