En
期刊文章
文章精选
当期目录
过刊浏览
高被引
高下载
特邀文章
专刊合集
待刊文章
Volume 117
您当前的位置：首页 > 期刊文章 > 过刊浏览 > Volumes 108-119 (2025) > Volume 117
An integrated image instance segmentation and multi-objective population balance framework for crystallization kinetics inversion (Open Access)
Nanchao Si a, Tuo Yao a, Sai Wang a, Renrui Xu a b, Zhenguo Gao a b c *, Junbo Gong a c
a School of Chemical Engineering and Technology, National Engineering Research Center of Industrial Crystallization Technology, Tianjin University, State Key Laboratory of Chemical Engineering and Low-Carbon Technology, Tianjin University, Tianjin, 300072, China
b Shanxi Research Institute of Huairou Laboratory, Taiyuan, 030032, China
c Haihe Laboratory of Sustainable Chemical Transformations, Tianjin, 300192, China
10.1016/j.partic.2026.07.003
Volume 117, October 2026, Pages 140-155
Received 29 March 2026, Revised 5 July 2026, Accepted 9 July 2026, Available online 15 July 2026, Version of Record 29 July 2026.
E-mail: zhenguogao@tju.edu.cn
Highlights

• YOLO11-seg links in situ microscopy to 2D PBM for taurine crystallization kinetics.

• 2D PBM with mixed nucleation and size-dependent anisotropic crystal growth.

• NSGA-II fits major and minor axis CSD simultaneously for kinetic inversion.

• Secondary nucleation generates virtually all crystals.


Abstract

The crystal size distribution is a critical quality attribute of crystalline products, yet extracting two-dimensional crystal size distributions from in situ microscopy images remains hindered by particle occlusion and computational cost. Here we couple instance segmentation with a two-dimensional population balance model to achieve kinetic parameter inversion from crystal images acquired at 100 frames per minute. The segmentation pipeline yields a mask mean average precision of 0.765 with an end-to-end throughput of 2.80 frames per second. The two-dimensional population balance model incorporates mixed primary–secondary nucleation and size-dependent anisotropic growth along both crystal axes. Seven kinetic parameters are simultaneously calibrated by minimizing the Hellinger distance between experimental and simulated major- and minor-axis crystal size distributions using multi-objective optimization. The best compromise solution achieves a combined Hellinger distance of 0.298. Over the entire crystallization window, primary nucleation contributes only a negligible initial seed population, and secondary nucleation generates virtually all crystals. The growth rate constant ratio of 2.0 yields an actual growth rate ratio of approximately 1.5 due to size-dependent decay. Bootstrap uncertainty quantification (30 replicates) indicates that secondary nucleation parameters are the most tightly constrained quantities. This end-to-end framework, from raw image acquisition to mechanistic kinetic inversion, provides a template for image-based crystallization kinetics analysis.

Graphical abstract
Keywords
Industrial crystallization; In-situ imaging; Deep learning; Instance segmentation; Population balance modeling; Multi-objective optimization
文章导读