Abstract:

Crystallization is an important technology for the separation and purification of materials. Currently, the parameters of crystallization processes are mainly monitored by off-line methods, which have some disadvantages, such as time lag and labor consumption. On-line monitoring based on process analytical technology (PAT) is of great significance for the optimization and control of crystallization, which can improve its production efficiency and product quality. In situ imaging technology can intuitively reflect the crystallization state, and the quantitative information provided by the image analysis algorithm can be used to study the mechanism such as crystallization kinetics. Although the analysis of in situ images is quite challenging, the recent AI technologies represented by deep learning have significantly enhanced the analysis accuracy. Following a brief introduction of some common PAT techniques for crystallization monitoring, this review summarizes the progress on in situ imaging technique with focus on the image analysis methods together with the revealed behaviors and mechanisms of crystallization. Finally, some future directions of in situ imaging technology are prospected for the intelligent monitoring of crystallization processes.

Key words: crystallization, imaging, size distribution, algorithm, deep learning, process analytical technology

摘要：

结晶是物质分离与纯化的重要技术，目前主要采用离线分析的方法监测结晶过程参数，存在滞后性、消耗人力等弊端。基于过程分析技术（process analytical technology，PAT）的在线监测对于结晶生产工艺的优化与控制具有重要意义，可以充分提高生产效率与产品质量。原位成像技术能直观地反映结晶状态，图像分析算法提供的定量信息可用于研究结晶动力学等机理。虽然原位图像分析极具挑战性，但近些年以深度学习为代表的人工智能技术显著提升了预测精度。在简要介绍结晶领域各种常用PAT技术的基础上，重点综述了基于原位成像技术的相关研究进展，侧重于图像分析方法及其揭示的结晶行为与机理。最后，展望了原位成像技术对实现结晶过程智能监测的一些未来发展方向。

关键词: 结晶, 成像, 粒度分布, 算法, 深度学习, 过程分析技术