Abstract
Oil-contaminated soils are a major environmental problem for Kazakhstan. Oil spills or leaks lead to profound changes in the physical and agrochemical properties of the soil and the accumulation of hazardous substances. Whilst there are many remote sensing techniques and complex laboratory methods for oil spill detection, developing simple, reliable, and inexpensive tools for detecting the presence of pollutants in the soil is a relevant research task. The study aims to research the possibilities of an electronic nose combining a chemical sensor array with pattern recognition techniques to distinguish volatile organic compounds from several types of hydrocarbon soil pollutants. An electronic nose system was assembled in our laboratory. It includes eight gas metal oxide sensors, a humidity and temperature sensor, an analog-digital processing unit, and a data communication unit. We measured changes in the electrical conductivity of sensors in the presence of volatile organic compounds released from oil and petroleum products and samples of contaminated and uncontaminated soils. The list of experimental samples includes six types of soils corresponding to different soil zones of Kazakhstan, crude oil from three oil fields in Kazakhstan, and five types of locally produced fuel oil (including gasoline, kerosene, diesel fuel, engine oil, and used engine oil). We used principal component analysis to statistically process multidimensional sensor data, feature extraction, and collect the volatile fingerprint dataset. Pattern recognition using machine learning algorithms made it possible to classify digital fingerprints of samples with an average accuracy of about 92%. The study results show that electronic nose sensors are sensitive to soil hydrocarbon content. The proposed approach based on machine olfaction is a fast, accurate, and inexpensive method for detecting oil spills and leaks, and it can complement remote sensing methods based on computer vision.