By Raphael Mawrence
In an attempt to expand the current spatial resolution of and capacity for detecting nitrogen dioxide (NO2), a calibration model is developed for Alphasense’s NO2-A43F and NO2-AE electrochemical sensors while on board a DJI Matrice 100 quadcopter. This is done in a third calibration phase after calibrating the sensors with traditional procedures using outdoor and indoor methods. The outdoor procedure (Phase 1) places the sensors next to another already-calibrated sensor on a rooftop at a measuring station in Amsterdam Vondelpark, and the indoor method (Phase 2) places the sensors in a controlled environment, where known concentrations of different gasses, including NO2, are pumped into a chamber that houses the sensors. Simple multivariable regression, polynomial regression, and logarithmic regression calculations are used to calibrate sensor readings for each sensor by using the raw output readings, temperature, and relative humidity of each sensor as independent variables throughout each of these phases, which further builds upon previous experiments by adding an additional variable in the third phase that accounts for interferences that might be coming from the UAV: rotor speed. After executing Phase 1 and Phase 2, the best performing model of the NO2-A43F and NO2-AE sensors were both during Phase 2 and have a predictive r-squared of 0.8346 and 0.9744, respectively, indicating that the outdoor calibration method is less suitable for calibrating these sensors as compared to the indoor calibration method. Coefficients derived from this calibration phase are used in order to obtain ground truth* readings for Phase 3, which calibrate the sensors while on board an UAV. This research found the statistical significance of adding rotor speed to the calibration model in order to derive more accurate NO2 concentrations in ppm. During the UAV calibration phase where this is implemented, the best performing model for the NO2-A43F has a predictive r-squared of 0.4991, and 0.7961 for the NO2-AE sensor. Upon integrating these calibrated sensors on an UAV in an additional experiment to validate these results, NO2 is detected in patterns that reflect the performance of these models. Future research should improve this study by using machine learning algorithms and other improvements throughout the methodological approach in order to derive more reliable measurements of detecting NO2 with electrochemical sensors on board an UAV, thereby further enhancing the spatial resolution of and capacity for detecting toxic gases.
Keywords: Air quality monitoring network; Alphasense; electrochemical sensor; nitrogen dioxide; spatial resolution; UAV; unmanned aerial vehicle