Design and Optimization of Electronic Nose Sensor Array for Real-Time and Rapid Detection of Vehicle Exhaust Pollutants

Abstract

Traditional vehicle exhaust pollutant detection methods, such as bench test and remote sensing detection, have problems such as large volume, high cost, complex process, long waiting time, etc. In this paper, according to the main components of vehicle exhaust pollutants, an electronic nose with 12 gas sensors was designed independently for real-time and rapid detection of vehicle exhaust pollutants. In order to verify that the designed electronic nose based on machine learning classification method can accurately identify the exhaust pollutants from different engines or different concentration levels from the same engine. After feature extraction of the collected data, Random Forest (RF) was used as the classifier, and the average classification accuracy reached 99.92%. This result proved that the designed electronic nose combined with RF method can accurately and sensitively judge the concentration level of vehicle exhaust pollutants.. Then, in order to enable the electronic nose to be vehicle-mounted and to achieve real-time and rapid detection of vehicle exhaust pollutants. We used Recursive Feature Elimination with Cross Validation (RFECV), Random Forest Feature Selector (RFFS) and Principal Component Analysis (PCA) to optimize the sensor array. The results showed that these methods can effectively simplify the sensor array while ensuring the RF classifier’s classification recognition rate. After using RFECV and RFFS to optimize the sensor array, the RF classifier’s classification recognition rate of the optimized sensor arrays for vehicle exhaust pollutants reached 99.77% and 99.44%, respectively. The numbers of sensors in the optimized sensor arrays were six and eight respectively, which achieved the miniaturization and low-cost of the electronic nose. With the limitation of six sensors, RFECV is the best sensor array optimization method among the three methods.