Machine learning based tool for the efficient estimation of geometric features of aggregated aerosol particles

While significant progress has been made in developing models for the formation and transport of aerosol aggregates, there is still a need for a simple, versatile tool capable of estimating intrinsic properties of aggregated particles. Scalar friction factor is an important parameter used extensively in the field of aerosol science. The scalar friction factor for non-spherical particles can be computed with the information on two geometric parameters, hydrodynamic radius (R_h) and orientationally averaged projected area (PA), depending on the momentum transfer regime. Although the existing methods for the estimation of these descriptors are efficient, many applications involve frequent estimation of these geometric descriptors, which can be time-consuming. We propose a Machine Learning (ML) based tool that can predict these descriptors using Fractal Dimension, pre-exponential factor, number of monomers and anisotropy factors as the input. An extensive database comprising fractal parameters, anisotropy factors, R_h, and PA is developed for testing and training the ML models. Five ML methods were assessed, with random forest (RF) identified as the most effective. The RF model demonstrated high accuracy in the testing phase, with R-squared value of 0.9875 for R_h and 0.9979 for PA, and average errors of 3.17% and 1.21% for R_h and PA, respectively. The predicted R_h and PA values were then used to estimate other relevant 3-dimensional properties such as mobility diameter, shape factor, and aerodynamic diameter, with the results indicating high accuracy of the prediction tool. Python-based tool offers ease of use, and can be easily integrated with other numerical codes.