Methods of Classifying Aerosol Types by Using Scattering of Light

This project demonstrates that different aerosol types can be precisely discriminated. This discrimination is done by using Machine Learning Algorithms which are trained via scattering light levels at different wavelengths. Paraffin, DEHS and PAO aerosols were applied to the optical measurement cell that contains blue and infrared laser diodes. The wavelengths of photodiodes as the light sources for creating different scattering characteristics were selected due to their commercial availability. The wavelengths were 450nm and 980nm. Five photodiodes were placed to measure forward, side and backward scattering of light. To create a dataset, each aerosol was tested many times at ten different concentrations. Models with more features were compared to the models with less features to determine the minimum requirements for the related discrimination. Neural network classifier, support vector machine and random forest were used as machine learning algorithms in this study. Despite similar physical characteristics, PAO, DEHS and paraffin aerosols were classified with high accuracy between 82% to 96%. The models which were proposed can easily be adapted to new aerosol types and offer a significant discrimination capability.