Walter Schaefer and Lingxi Li, “Particle characterization by analyzing light scattering signals with a machine learning approach,” Appl. Opt. 63, 7701-7707 (2024)
This work marks the foundational step in the development of the SprayQuantAI® solution for monitoring complex particles in sprays. It introduces a machine learning-based approach that enables accurate characterization of droplets using only a single light scattering signal. The authors demonstrate that key particle properties—such as size and velocity—can still be predicted with high accuracy, even when only one signal is available in the final configuration. This breakthrough paves the way for a significant reduction in hardware complexity, cost, and footprint of TSTOF-based systems. Importantly, the method also opens new opportunities for evaluating the internal composition of droplets. The ability to extract such information from light scattering signals introduces the potential to assess not just external characteristics, but also material-specific properties, such as refractive index and concentration, enabling in-situ composition monitoring of droplets within sprays.
Abstract
We present here a new instrument, to our knowledge, in combination with a machine learning approach to achieve a more cost-effective and compact measurement instrument for particle characterization in a flow based on the established measurement technique known as the time-shift-time-of-flight (TSTOF) technique. A commercial device based on TSTOF was introduced and has since been recognized under the brand name SpraySpy. In this study, we propose a machine learning model capable of using only a single signal in this device to determine the same information about particles such as particle size and particle velocity, traditionally obtained from the classical measurement device based on the TSTOF technique, where four signals have been used. To achieve this, we train a machine learning model using the four signals, but connect only a single signal to the model in the final step. The initial experimental results have been conducted, and preliminary calculations demonstrate high potential for this method. By applying this method, one light source and three detectors, along with the corresponding electronics and optics, can be eliminated from a TSTOF measurement instrument. This not only reduces hardware costs but also enables the production of a smaller measurement probe and the use of a single signal acquisition system without the need for synchronization.