基于深度学习的连续结晶形貌表征与表观生长动力学分析

李强; 张相洋

doi:10.11949/0438-1157.20260806

您当前的位置：

首页 >

文章列表页 >

基于深度学习的连续结晶形貌表征与表观生长动力学分析

分离工程 | 更新时间：2026-07-11

- 基于深度学习的连续结晶形貌表征与表观生长动力学分析
- Lightweight deep learning-based shape characterization and apparent growth kinetics analysis in continuous crystallization
- 化工学报 2026年
- 作者机构：
  
  华东理工大学化学工程与低碳技术全国重点实验室，上海 200237
- 作者简介：
  
  李强（2001—），男，博士研究生，shaun.liqiang@outlook.com
  张相洋（1981—），男，博士，教授，zxydcom@ecust.edu.cn
- 基金信息：
  
  上海市自然科学基金项目(23ZR1417100)
- DOI：10.11949/0438-1157.20260806
  中图分类号： TQ 026.5
- 收稿：2026-06-11，
  
  修回：2026-07-10，
  
  录用：2026-07-11，
- 稿件说明：
移动端阅览
李强, 张相洋. 基于深度学习的连续结晶形貌表征与表观生长动力学分析[J/OL]. 化工学报, 2026.

LI Qiang, ZHANG Xiangyang. Lightweight deep learning-based shape characterization and apparent growth kinetics analysis in continuous crystallization[J/OL]. CIESC Journal, 2026.
李强, 张相洋. 基于深度学习的连续结晶形貌表征与表观生长动力学分析[J/OL]. 化工学报, 2026. DOI： 10.11949/0438-1157.20260806.

LI Qiang, ZHANG Xiangyang. Lightweight deep learning-based shape characterization and apparent growth kinetics analysis in continuous crystallization[J/OL]. CIESC Journal, 2026. DOI： 10.11949/0438-1157.20260806.

摘要

面向连续结晶中产品形貌表征难的问题，提出基于轻量化深度学习的表观动力学分析方法。引入视觉基础大模型SAM3辅助预标注，利用轻量化网络（YOLO26m-seg）对晶体图像进行实例分割（单张耗时约17 ms），结合图像二阶矩提取形貌特征。以库埃特—泰勒连续结晶器内双氰胺结晶过程为研究对象，结果表明在稳态工况下产品

为117.0 μm，并基于双判据方法识别出约2.9%的可疑团聚体。建立二维表观动力学模型，量化双氰胺长短轴表观生长速率比（1.75），讨论了强剪切流场对传质边界层的非对称影响。为连续结晶产品质量控制提供算法支撑。

Abstract

To address the challenge of characterizing product habits in continuous crystallization

an apparent kinetic analysis method based on lightweight deep learning is proposed. A vision foundation model (SAM3) was introduced for auxiliary pre-labeling

and a lightweight network (YOLO26m-seg) was utilized for the instance segmentation of crystal images

achieving a processing speed of approximately 17 ms per frame. Morphological features were then extracted using image second-order moments. Using the continuous crystallization of dicyandiamide in a Couette-Taylor crystallizer as a case study

the median particle size (

) under steady-state conditions was determined to be 117.0 μm

and approximately 2.9% of suspicious agglomerates were identified based on a dual criterion. Furthermore

a two-dimensional apparent kinetic model was established to quantify the apparent growth rate ratio of the major to minor axes of dicyandiamide (1.75)

and the asymmetrical effect of the strong shear flow field on the mass transfer boundary layer was discussed. This study provides algorithmic support for the quality control of continuous crystallization products.

关键词

Keywords

references

Eren A , Civati F , Ma W C , et al . Continuous crystallization and its potential use in drug substance manufacture: a review [J ] . Journal of Crystal Growth , 2023 , 601 : 126958 .

Orehek J , Teslić D , Likozar B . Continuous crystallization processes in pharmaceutical manufacturing: a review [J ] . Organic Process Research & Development , 2021 , 25 ( 1 ): 16 - 42 .

Ma Y M , Wu S G , Macaringue E G J , et al . Recent progress in continuous crystallization of pharmaceutical products: precise preparation and control [J ] . Organic Process Research & Development , 2020 , 24 ( 10 ): 1785 - 1801 .

Zhang D J , Xu S J , Du S C , et al . Progress of pharmaceutical continuous crystallization [J ] . Engineering , 2017 , 3 ( 3 ): 354 - 364 .

Biri D , Rajagopalan A K , Mazzotti M . Autonomous control of the crystal size and shape in dense suspensions via imaging [J ] . Industrial & Engineering Chemistry Research , 2025 , 64 ( 19 ): 9794 - 9805 .

Kim S , et al . Codesigning tubular flow and noncontact sonication for antisolvent crystallization: rapid fouling-free production of uniform rifapentine crystals with reduced aspect ratios [J ] . Crystal Growth & Design , 2026 , 26 ( 6 ): 2249 - 2261 .

Ilett T P , Hazlehurst T A , Jiang C , et al . Reconstruction of 3D crystal growth from transmission optical microscopy images [J ] . PNAS Nexus , 2026 , 5 ( 4 ): pgag080 .

Burgdorf S J , Roddelkopf T , Thurow K . Automated crystallization monitoring in material development using computer vision and neuronal networks [J ] . Chemie Ingenieur Technik , 2024 , 96 ( 8 ): 1107 - 1115 .

Nagy Z K , Fujiwara M , Braatz R D . Monitoring and advanced control of crystallization processes [M ] // Handbook of Industrial Crystallization . Cambridge, UK : Cambridge University Press , 2019 : 313 - 345 .

Acevedo D , Wu W L , Yang X C , et al . Evaluation of focused beam reflectance measurement (FBRM) for monitoring and predicting the crystal size of carbamazepine in crystallization processes [J ] . CrystEngComm , 2021 , 23 ( 4 ): 972 - 985 .

Li H Y , Kawajiri Y , Grover M A , et al . Application of an empirical FBRM model to estimate crystal size distributions in batch crystallization [J ] . Crystal Growth & Design , 2014 , 14 ( 2 ): 607 - 616 .

Yu L X , Lionberger R A , Raw A S , et al . Applications of process analytical technology to crystallization processes [J ] . Advanced Drug Delivery Reviews , 2004 , 56 ( 3 ): 349 - 369 .

Li W L , Chu T F , Cheng Y Y , et al . An improved method for crystal morphology extraction using in situ imaging [J ] . Chemical Engineering & Technology , 2025 , 48 ( 12 ): e70138 .

Odziomek K , Ushizima D , Oberbek P , et al . Scanning electron microscopy image representativeness: morphological data on nanoparticles [J ] . Journal of Microscopy , 2017 , 265 ( 1 ): 34 - 50 .

Zong S L , Zhou G Z , Li M , et al . Deep learning-based on-line image analysis for continuous industrial crystallization processes [J ] . Particuology , 2023 , 74 : 173 - 183 .

Bharati P , Pramanik A . Deep learning techniques: R-CNN to mask R-CNN: a survey [C ] // Computational Intelligence in Pattern Recognition . Singapore : Springer , 2020 : 657 - 668 .

Kang Y L , Duan Z Y , Tong T L , et al . An enhanced deep learning-based pharmaceutical crystal detection with regional filtering [J ] . Crystals , 2024 , 14 ( 8 ): 709 .

Yao T , Liu J , Wan X X , et al . Deep-learning based in situ image monitoring crystal polymorph and size distribution: Modeling and validation [J ] . AIChE Journal , 2024 , 70 ( 2 ): e18279 .

Wang X , Duan T S , Ouyang G X , et al . Comprehensive review of segment anything model across multiple domains [J ] . Digital Signal Processing , 2025 , 167 : 105459 .

Zhang C H , Liu L , Cui Y W , et al . A comprehensive survey on segment anything model for vision and beyond [EB/OL ] . 2023 : arXiv : 2305 . 08196 . https://arxiv.org/abs/2305.08196 https://arxiv.org/abs/2305.08196

Bochkovskiy A , Wang C Y , Liao H M . YOLOv4: optimal speed and accuracy of object detection [EB/OL ] . 2020 : arXiv : 2004 . 10934 . https://arxiv.org/abs/2004.10934 https://arxiv.org/abs/2004.10934

Redmon J , Divvala S , Girshick R , et al . You only look once: unified, real-time object detection [C ] // 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) . June 27 - 30 , 2016 . Las Vegas, NV, USA . IEEE , 2016: 779 - 788 .

Rath S . YOLOv8 Ultralytics: state-of-the-art yolo models [EB/OL ] . LearnOpenCV , 2023 . https://learnopencv.com/ultralytics-yolov8 https://learnopencv.com/ultralytics-yolov8

Manee V , Zhu W , Romagnoli J A . A deep learning image-based sensor for real-time crystal size distribution characterization [J ] . Industrial & Engineering Chemistry Research , 2019 , 58 ( 51 ): 23175 - 23186 .

Salami H , McDonald M A , Bommarius A S , et al . In situ imaging combined with deep learning for crystallization process monitoring: application to cephalexin production [J ] . Organic Process Research & Development , 2021 , 25 ( 7 ): 1670 - 1679 .

Moenck K , Thieu D T , Koch J , et al . Industrial language-image dataset (ILID): adapting vision foundation models for industrial settings [J ] . Procedia CIRP , 2024 , 130 : 250 - 263 .

Carion N , Hu Y T , Debnath S , et al . SAM 3: Segment anything with concepts [EB/OL ] . 2025 : arXiv : 2511 . 16719 . https://doi.org/10.48550/arXiv.2511.16719 https://doi.org/10.48550/arXiv.2511.16719

Nguyen A T , Yu T , Kim W S . Couette-Taylor crystallizer: Effective control of crystal size distribution and recovery of L-lysine in cooling crystallization [J ] . Journal of Crystal Growth , 2017 , 469 : 65 - 77 .

Nguyen A T , Joo Y L , Kim W S . Multiple feeding strategy for phase transformation of GMP in continuous couette–Taylor crystallizer [J ] . Crystal Growth & Design , 2012 , 12 ( 6 ): 2780 - 2788 .

Wang W . X-AnyLabeling: advanced auto labeling solution with added features [CP/OL ] . GitHub , 2023 . https://github.com/CVHub520/X-AnyLabeling https://github.com/CVHub520/X-AnyLabeling

Jocher G , Qiu J , Liu M Y , et al . Ultralytics YOLO26: unified real-time end-to-end vision models [EB/OL ] . 2026 : arXiv : 2606 . 03748 . https://arxiv.org/abs/2606.03748 https://arxiv.org/abs/2606.03748

浏览量

下载量

CSCD

文章被引用时，请邮件提醒。

提交

工具集

关联资源

基于双阶段双通道深度学习的化工故障分类模型

基于机器学习的改型歧管式微通道热沉流动传热特性多目标优化

基于Micro-CT和深度学习的盐溶液冰离心分离微观结构研究

融合自适应证据的深度学习框架预测PFAS环境迁移性质

基于化学语言的分子物化性质深度学习通用建模框架