ML-based analysis of 3D colloidal self-assembly induced by rotating magnetic fields

Abstract

Machine learning is increasingly applied in the post-processing of physical experiments, where it significantly accelerates inference compared to traditional human-based analysis. In this work, we present a methodology for analyzing the three-dimensional self-assembly of colloidal particles induced by rotating magnetic fields. The work focuses on the influence of controlled parameters – system density, coil current, anisotropy, and precession angle – on both cluster configurations and the diffusion of individual particles.

Cluster segmentation was performed using a zero-shot segmentation model for high-density systems and the Density-Based Spatial Clustering of Applications with Noise (DBSCAN) algorithm for low-density systems. Cluster size was estimated under the assumption of uniform hexagonal tiling.

For the reconstruction of colloid coordinates in 3D space, we developed a method combining k-means clustering of scattering-profile geometric parameters with calibration experiments. Specifically, we recovered the power-law relationship between the z-coordinate and the scattering profile area by employing a dual-camera setup: the primary camera with a laser plane in the xy plane and an auxiliary camera with a laser plane of a different wavelength in the xz plane.

Particle tracking was performed using the Crocker-Grier algorithm. The proposed post-processing pipeline accurately identifies characteristic features of colloidal scattering profiles and extracts their spatial arrangement.
In addition to the experimental studies, molecular dynamics simulations were performed for a system relevant to the experimental conditions. Based on the clustering results, statistical characteristics of the cluster size distribution - including mean, maximum, and median particle number per cluster - were computed as functions of the simulation parameters. Furthermore, analysis of the trajectories yielded mean squared displacement (MSD) curves and diffusion coefficients as functions of the controlled parameters.

The suggested approach enables high-resolution spatiotemporal monitoring of colloidal self-assembly in three dimensions and holds promise for applications in advanced materials development, 3D bioprinting, and micro- and nanofabrication.

Speaker

Anton I. Shvetsov
Bauman Moscow State Technical University, Moscow, Russia
Russia

Discussion

Ask question