Scientists from the University of Electronic Science and Technology of China, Zhejiang University and Kunming University of Science and Technology have developed a compact “electronic nose”, a system that can detect low concentrations of harmful gases in the air with high accuracy. The device is considered an inexpensive and efficient tool for monitoring air pollution at industrial facilities, in urban environments and indoors.
Precise and prompt detection of volatile organic compounds, such as ethanol, ammonia and toluene, remains a critical challenge. Even in low concentrations, these substances pose a threat to human health and environment. While conventional analysis methods, including gas chromatography, provide high accuracy, they require expensive equipment and complex preparation, and cannot be used for continuous monitoring outside of laboratories. The Chinese researchers have proposed a more practical approach: combining several readily available gas sensors with machine learning algorithms to create a system that can detect characteristic signatures of individual gases and their mixtures.
The electronic nose was built from six sensors: four commercially available tin dioxide (SnO₂) sensors and two laboratory-scale sensors based on cobalt oxide (Co₃O₄) and its modified version containing manganese (Mn-Co₃O₄). These materials react differently to the presence of target gases, changing their electrical resistance. This diversity of responses creates a collective olfactory effect, allowing the system to obtain a much more detailed picture than a single sensor would.
The experiments were conducted in a sealed chamber, into which ethanol, ammonia and toluene were fed at concentrations ranging from 2 to 10 parts per million, a range typical for real industrial emissions. Every sensor recorded changes in resistance over time, forming an individual response curve. Dozens of parameters were automatically extracted from these curves, including reaction speed, recovery time, maximum signal values and area under curve. As a result, each measurement was described by 180 parameters making up a detailed digital portrait of the gas environment.
To analyze these data, the researchers tested four machine learning algorithms: the k-nearest neighbors algorithm, support vector machines, logistic regression and random forest. Although all of them successfully identified individual gases, the k-nearest neighbors algorithm performed best, achieving 100% accuracy. The identification of binary mixtures, when two gases are simultaneously present in the air in different proportions, proved particularly challenging. In this case, the system using data from all six sensors demonstrated 97.2% accuracy, identifying both the mixture composition and the concentration of each component almost without error.
Currently, the researchers plan to test the system’s operation under real-world conditions. If the tests are successful, these devices will be implemented at industrial facilities and in urban air-quality monitoring systems.
It should be noted that technologies for identifying complex chemical environments based on sensor arrays and intelligent data processing are finding application in a wide variety of industries. Last year, scientists from Skoltech and Kazakhstan’s Eurasian National University proposed using an electronic nose to analyze the component composition of oil and assess the impacts of oil spills.
