Sniffing Out Danger

It has long been known that the presence of certain gasses can be responsible for causing health problems, and in the worst case, even death, in humans. Being generally colorless, and often odorless, toxic fumes are very difficult to detect before they have a chance to do harm. Well over 100 years ago, before the modern age of technology, miners sought a way to detect the presence of carbon monoxide in underground mines. Lacking technological solutions, they brought canaries into the mines with them to serve as early-warning signals, as they were noted to be more sensitive to carbon monoxide poisoning than humans.

Today we have much better solutions, like artificial noses, that can reliably detect even minute traces of toxic gasses and alert us to their presence. There are still some areas where further advancements are needed to keep us safe, however. Consider nitrogen dioxide, for example. This gas, released by generators, burners, water boilers, and other fossil fuel combustion systems causes severe respiratory problems like asthma and chronic obstructive pulmonary disease when inhaled. Existing technologies that detect nitrogen dioxide lack sensitivity, exhibit inconsistent performance, and are otherwise bulky and impractical for real-world use.

A group headed up by researchers at the University of Virginia has taken on the challenge of building a more accurate and practical nitrogen dioxide sensing system. Their solution involved the development of novel sensing hardware as well as machine learning algorithms to assist in interpreting the sensor data. Testing of this system showed that it is capable of accurately detecting nitrogen dioxide, and of pinpointing the precise location of leaks.

The hardware for the gas monitoring system employs an artificial olfactory receptor inspired by biological olfactory mechanisms. The receptor uses an AlGaN/GaN high-electron mobility transistor (HEMT) with a two-dimensional electron gas (2DEG) channel, which allows for highly sensitive current modulation in response to environmental changes. This HEMT structure is paired with palladium (Pd) nano-islands deposited on a graphene gate electrode. The Pd nano-islands catalytically interact with nitrogen dioxide molecules, breaking them into charged ions that temporarily bond to the graphene surface, effectively altering the electric field and modulating the current in the HEMT's 2DEG channel. This design enables the receptor to exhibit high responsivity at room temperature.

The team then built a gas monitoring system by combining multiple artificial olfactory receptors with a machine learning model based on an artificial neural network (ANN). This network was trained to analyze time-dependent data from the receptors to pinpoint the locations of nitrogen dioxide leaks within a given space. To enhance sensor placement for accurate monitoring, the researchers employed a high-dimensional optimization algorithm called TuRBO (trust-region Bayesian optimization). This algorithm efficiently identified optimal sensor configurations by dividing the search space into smaller subsets, allowing parallel optimization. The optimal setup was determined based on minimizing the distance between actual and predicted leak locations.

Once optimized, the system was deployed with the ANN running on near-sensor microprocessors for efficient, localized processing. It was shown that the system could perform precise, real-time tracking of gas concentrations without relying on large computing resources or cloud access. This integrated hardware-software approach provides a reliable and energy-efficient tool for gas leak detection, beneficial for safety in both industrial and residential environments.