Novel Smart Gas Sensor Is Smaller, Better Than Its Rivals — Thanks to Machine Learning

Researchers at the King Abdullah University of Science and Technology (KAUST) have come up with a smarter gas sensor, using machine learning to augment a relatively simple temperature sensor and sniff out specific gas types in the air.

"Unlike traditional gas sensors," project lead Mohammad Younis explains, "our sensor does not require any special coating, which enhances the chemical stability of the device and also makes it scalable. You can scale the device down to the nano-regime without affecting its performance since it does need a big surface for the coating."

The team's sensor is, in effect, a temperature sensor — constructed from a heated strip of silicon forming a microbeam resonator, clamped at both ends and bent to just before its buckling point. "When operated near buckling point," first author Usman Yaqoob explains,

"the heated microbeam shows significant sensitivity to different gases when they have a heat conductivity lower or higher than air. The shift in resonance frequency is detected using a microsystem analyzer vibrometer."

Data from the sensor was fed through a machine learning system which could map frequency changes to their corresponding gases. "Data processing and machine learning algorithms are used to generate unique signature markers for each tested gas to develop an accurate and selective gas classification model," Yaqoob says — with experimental testing showing the system able to identify helium, argon, and carbon dioxide with 100 percent accuracy from an unknown dataset.

The team is hoping the sensor system will see broad deployment, thanks to its scalability down to extremely small sizes for embedded use and its relatively simple production process plus high resistance to incorrect identification through cross-sensitivity compared with traditional gas sensors.

The paper detailing the researchers' work has been published in the IEEE Sensors Journal, under closed-access terms; an open-access copy is available directly from KAUST as a PDF download.