Intelligent PM 2.5 mass concentration analyzer using deep learning algorithm and improved density measurement chip for high-accuracy airborne particle sensor network. (January 2023)

• Record Type: Journal Article
• Authors: Lee, Seung-Soo; Song, Woo-Young; Kim, Yong-Jun
• Is Part Of: Journal of aerosol science. Volume 167(2022)
• Publication Date: 2023-01
• Subjects: PM 2.5 mass concentration -- Deep learning algorithm -- Nanoparticle size distribution -- Effective density -- Real-time measurement -- Airborne particle sensor network -- Air quality monitoring; Aerosols -- Periodicals; Aerosols -- Periodicals; Aérosols -- Périodiques; 541.34515
• DOI: 10.1016/j.jaerosci.2022.106097
• Languages: English
• ISSNs: 0021-8502

Abstract

Abstract: The concentration of ultrafine airborne particles is continuously increasing, and research has shown that it has adverse effects when inhaled into the human body. Accordingly, there is a growing demand for a measurement network utilizing sensors to evaluate an individual's exposure to these airborne particles. However, the current low-cost sensors have the limitation of low accuracy. To solve this, we devised a method to calibrate a low-cost mass concentration sensor accurately in real time. In a previous study, we developed an analyzer that could measure the effective density and nanoparticles that cause the low accuracy of the mass concentration sensor. However, it had a hardware stability problem when used to monitor the outside air for a long period. This has been improved by modifying the shape of the Micro-electromechanical system (MEMS)-based chip integrated with the electrical/inertial analysis technology. Hence, the device can now operate with sufficient stability in the outdoor air. In addition, the retrieval algorithm used to convert the measured current values into the effective density and nanoparticle size distribution was prone to errors. It was modified to a deep learning-based physical parameter conversion algorithm to minimize the errors. Thus, we developed a standalone analyzer that integrates the improved nanoparticle and effective density analyzer, temperature and humidity sensor, and low-cost mass concentration sensor. In addition, we developed a technique to calibrate the mass concentration sensor data accurately in real time based on the data measured by each sensor and analyzer and a deep learning-based mass concentration calibration algorithm. The calibration precision was confirmed through comparative evaluation of the analyzer with the results of a beta attenuation mass monitor. In the future, this analyzer can be utilized to build a sensor network that precisely monitors the mass concentration in a large area through multi-point deployment. Alternatively, it can be used for monitoring the mass concentration in an indoor private space, where it is difficult to place expensive and large equipment. Highlights: A low-cost, high precision, portable PM 2.5 analyzer was developed in this study. A deep learning algorithm is proposed for converting nanoparticle size distribution and effective density. Another deep learning algorithm is developed for mass concentration calibration. The proposed analyzer enables precise mass concentration analysis in real time. The device can be conveniently deployed in indoor/large-scale outdoor environments.