Listening For Trouble

In a petrochemical plant, a pump or compressor that suddenly goes on the fritz can rack up huge downtime costs--$1 million a day or more if it shuts down the plant. To preclude such failures, Domain Dynamics Ltd. (DDL), a startup on the Swindon campus of Britain's Cranfield University, has developed technology to eavesdrop on machinery and spot subtle changes that signal impending problems--before the equipment crashes.

One key is neural-network software. Neural nets mimic the brain's circuitry and can detect minute variations in the noise that equipment makes. Thus, neural-net sensors can perform like "golden ear" technicians who, after years of experience, can tell whether a machine is functioning normally just by listening to it.

DDL's software goes further, to TESPAR, short for time-encoded signal processing and recognition. This technology transforms sounds into digital models based solely on the shape of the analog signal. As a result, says DDL director Ian M. Taylor, TESPAR can spot changes that would elude conventional analysis of a signal's frequency or duration.