Non-spherical particle size and shape estimation using machine learning

Abstract

A real time measurement of particles being ingested by gas turbines would prove useful for accurately monitoring engine health and ensuring safe operations. However, typical optical methods assume spherical particles, which most ingested particles are not. We present a novel application of machine learning models that takes scattering and extinction observations as inputs and estimates non-spherical particle shape (via aspect ratio) and size. The overall method of multiple classification and regression layers, as well as the results from three test cases using simulated inputs are presented.