AAAR 37th Annual Conference October 14 - October 18, 2019 Oregon Convention Center Portland, Oregon, USA
Abstract View
A Data-driven Approach for Detection of Toxic Metallic Particulate Matters Using Spark Emission Spectroscopy and Machine Learning Algorithms
SEYYED ALI DAVARI, Anthony S. Wexler, University of California, Davis
Abstract Number: 842 Working Group: Air Quality Sensors: Low-cost != Low Complexity
Abstract Toxic metal particulate matter (PM) has been associated with serious health issues. Conventional techniques for measuring their atmospheric concentration are limited by cost, cumbersome sample preparation and poor time resolution. Spark emission spectroscopy is introduced as a portable, modular and yet affordable instrument to quantify these compounds in real time. Each analysis results in a high-dimensional spectrum, which can be difficult to interpret specially for poor resolution spectrometers. To detect useful features from the spectrum and improve the detection limits and reliability, an unsupervised learning algorithm was combined with two supervised learning algorithms. Specifically, a K-Means clustering algorithm was employed to address shot-to-shot variations as well as inherent faults due to low-cost components. Then, partial least square (PLS) regression and least absolute shrinkage and selection operator (LASSO) were applied for quantitative predictions. Various hazardous metal elements such as Cr, Pb and Ni were analyzed. The technique provides an affordable, real-time approach compared to conventional techniques such as ICP-OES and XRF.