Non-spherical particle size estimation using supervised machine learning

Abstract

The inverse scattering problem of non-spherical particle size estimation is solved using a series of supervised machine learning models trained on a library of light scattering data. By establishing a large library with spheres and spheroids as fundamental shapes and through optimization of model hyperparameters, the trained models are able to accurately estimate a precise equivalent volume sphere radius of particles from an external database and simulations, with root mean square errors of 2.6% and 1.9% for the external and simulated particles, respectively. It was found that classification via a k -nearest neighbor model and refinement via a trained ensemble regression model performed best for equivalent volume measurements.

Non-intrusive optical measurements of gas turbine engine inlet condensation using machine learning

“…Machine or statistical learning models use the library data to fit functions that relate the particle's scattering and extinction patterns to its characteristics. The application of supervised ML models for estimating non-spherical particle characteristics was demonstrated previously by the authors [14]. Many of the basic details regarding the method can be found in this prior publication.…”

“…More importantly, a machine learning (ML) approach for particle characterization is achieved using multi-wavelength extinction as one of its features. Previous efforts demonstrated the concept of ML models and a vast library of relevant particle data for model training and validation [14].…”

“…ML is a statistical tool that uses data to fit functions that can model complex systems [15]. Our previous efforts used a preestablished library of particle scattering data to accurately size non-spherical particles found in the literature [14]. The features used previously were scattering intensity ratios at different scattering angles and an intensity shift relative to a known reference, requiring a calibration.…”

Applications of Artificial Intelligence and Machine Learning Algorithms to Crystallization

“…k -NN has been applied to predicting the melting point using a diverse data set of organic and drug molecules, as well as to the classification of aqueous solubility . Moreover, the k-NN was also applied to solving the inverse scattering problem for nonspherical particle size and shape estimation from light scattering data …”

“…42 Moreover, the k-NN was also applied to solving the inverse scattering problem for nonspherical particle size and shape estimation from light scattering data. 43 Another commonly used classifier is the Nai ̈ve Bayes, which belongs to the broader category of the so-called generative learning algorithm and often leads to a parametric model, although nonparametric (kernel) formulations exist. This algorithm performs well in many cases, even though the nai ̈ve assumption is rarely true.…”