Development of artificial intelligence systems for signal processing and signal enhancement in particulate matter sensors

Abstract

The World Health Organization (WHO) claims that air pollution will be a significant environmental concern in the coming years, leading to increased emphasis on actions focused on reducing pollutant levels in the air, from which particulate matter (PM) is the most harmful to health, with a large percentage of the population exposed to levels exceeding WHO standards. In this sense, air quality monitoring is essential to know pollutant levels in the air. However, traditional measuring devices are expensive and not automatic, making necessary to develop solutions based on new paradigms such as the Internet of Things (IoT) that are able to reduce costs. However, IoT devices present significant biases in measurement processes, especially concerning particles, due to the influence of external factors such as humidity. Therefore, this thesis explores solutions to improve signal in particle sensors using a bottom-up methodology. Firstly, a dryer system based on silicone columns has been developed, which has been evaluated with respect to reference methods. Next, an edge computing layer has been deployed allowing data processing using machine learning models to make short-term particle concentration predictions. Later, hyperlocal data from devices is used to calibrate the CHIMERE-WRF chemical transport model. The results demonstrate that the developed drying system improves PM2.5 accuracy compared to existing sensors, with a coefficient of determination (R2) equal to 0.83. Additionally, these devices can be easily integrated into an IoT-Edge-Cloud architecture for risk assessment in work environments, using machine learning to predict exceedances with 80 % accuracy. On the other hand, calibration of the CHIMERE-WRF model allows for a 63 % improvement in system correlation for NO2 and a 25 % improvement for O3. Lastly, a new proposal for air quality zoning for the Murcia Region, using an automated methodology, allows for the identification of new monitoring points. In conclusion, the combination of hyperlocal air quality measurements and simulations with chemical transport models allows for improved particle measurement and the generation of products to assess the impact on sustainability of different policies, known as sustainability impact assessment. This knowledge is crucial in the Smart Cities sector. The future challenge lies in applying large amounts of data generated by IoT in more complex scenarios and digital solutions for cities.