Neural network-enhanced dynamic light scattering for accurate sizing of large microparticles

Abstract

Dynamic light scattering (DLS) is a well-established technique for measuring particle sizes in colloidal suspensions, but its accuracy significantly diminishes for microparticles due to the dominance of multiple scattering, especially in high-volume samples. In this study, we introduce a data-driven approach using deep neural networks (DNNs) to enhance the performance of DLS for characterizing micron-sized particles. Silica microparticles of defined size ranges were prepared using a sieve tower and suspended in a water-glycerin mixture to ensure stability during measurements. Raw time-resolved scattering signals were recorded without pre-processing and used to train both classification and regression DNNs. Our results demonstrate high accuracy and repeatability in particle sizing for a wide range of sizes. Furthermore, by employing a moving sampling window across the scattering signal, we analyzed the effect of signal segments on classification performance and found that data corresponding to early-time single-scattering regions are crucial for model accuracy. This method opens new possibilities for in situ and real-time particle sizing in complex industrial environments where traditional DLS fails. © 2025 Elsevier Ltd