Image: Mark Hooper

In 1982, Krishna Persaud and George Dodd at the University of Warwick, in Coventry, England, put together the first sensor array for odor ­recognition—that is, a collection of electronic sensors, each of which responded in different ways to a range of volatile chemicals. Persaud and Dodd used oscillating semiconducting transducers that changed frequency when they detected certain compounds.

In 1988, Julian Gardner, also at the University of Warwick, dubbed this approach the electronic nose. Rather than cataloging the chemical compounds of an odor, an e-nose identifies complex odors by using pattern-recognition strategies similar to those of the human olfactory system, albeit with different sensing mechanisms. Given a glass of wine, the average person knows from the smell alone what the liquid is. But only a serious oenophile might be able to break the odor down into its constituent parts: alcohols, acids, and esters. The human nose has hundreds of different types of odor sensors, whose response patterns are processed by the brain, which then searches its memory for matches to stored response patterns. An electronic nose uses far fewer sensors; commercial ­systems have around 10 to 50 sensing elements.

Your typical e-nose consists of a sampling system, a gas-sensor array, and a signal processor coupled to a pattern-recognition system of some sort. The sampling system brings vapor-laden air into the sensor array; in a laboratory setup it might have a fan that blows air across the array in an action reminiscent of human sniffing. The nose might allow a vial of air to be released inside it; perfume makers capture samples this way. Or it might work passively, simply because the array is exposed, like the sensor in a smoke detector.

In the sensor array, each of the sensors responds to a broad range of gases, with much duplication; multi­ple sensors will respond to the same gas, but not in the same way, and not to all the same gases. To identify specific odors requires the signal processor to analyze the array response with pattern-recognition algorithms; in today's expensive electronic noses, a microprocessor uses a large set of stored algorithms to sort through patterns. In the future, however, single-purpose noses looking for a simple change—food gone bad, for example—could use application-specific integrated circuits for analysis.

Much like that of the human nose, this type of odor recognition is more flexible and more powerful than what is possible with a lock-and-key sensor, which can detect only a single compound, say, carbon monoxide. Such a sensor would have a hard time telling the difference between Grandma's apple pie and Mom's. But it may be possible to train an array-type e-nose to discriminate between them and all other apple pies.

And unlike systems based on lock-and-key sensors, electronic noses can be enormously flexible. Rather than developing one nose for wine monitoring and a different one to detect bad fish, the same piece of hardware could be trained separately for different tasks. Imagine an electronic-nose system shipped with standard pattern-recognition libraries. Load up one for the refrigerator and the system will sniff for spoiling foodstuffs; load up a different one for the garden and the system searches instead for the telltale odors of snails and other pests. And what if you want the e-nose to learn the difference between Grandma's apple pie and Mom's? Well, chances are the manufacturers will have never met Grandma or Mom or sampled the output of their ovens. But they may have included software for generating new pattern-recognition libraries. If so, you would hook up the nose to the training system, introduce it to one apple pie at a time, and find out if the pies generate distinguishable responses in the array. If they do, then generate a new library, load it up, and you've got a personalized apple- pie connoisseur.