Abstract l -ascorbic acid is synthetized in large industrial scale from glucose and marketed as an immune system strengthening agent and anti-oxidant ingredient. The overall yield of conversion of the precursor glucose to l -ascorbic acid is limited, therefore the crystallization is a critically important step of the l -ascorbic acid production from economic point of view. It is widely accepted that the crystal size distribution (CSD) influences numerous relevant macroscopic properties of the final crystalline product and it also significantly affects the downstream operations. The present paper discusses the chord length distribution (CLD, which is directly related to the CSD) control, during the crystallization of l -ascorbic acid from aqueous solution. Batch crystallization process is employed, which is the classical, and still dominant, operation in fine chemical and pharmaceutical industries. A comparative experimental study of two state-of-the-art Quality-by-Control (QbC) based crystallization design approaches are presented: (1) a model-free QbC based on direct nucleation control (DNC) and (2) a model-based QbC using a novel nonlinear model predicative control (NMPC) framework. In the first investigation, the DNC, a process analytical technology based state-of-the-art model free control strategy, is applied. Although, DNC requires minimal preliminary system information and often provides robust process control, due to the unusual crystallization behavior of l -ascorbic acid, it leads to long batch times and oscillatory operation. In a second study the benefits of model-based QbC approach are demonstrated, based on using a NMPC approach. A population balance based crystallization process model is built and calibrated by estimating the nucleation and growth kinetics from concentration and CLD measurements. A projection based CSD to CLD forward transformation is used in the estimation of nucleation and growth kinetics. For robustness and adaptive behavior, the NMPC is coupled with a growing horizon state estimator, which is aimed to continuously improve the model by re-adjusting the kinetic constants. The study demonstrates that the model-based QbC framework can lead to rapid and robust crystallization process development with the NMPC system presenting good control behavior under significant plant model mismatch (PMM) conditions.

中文翻译：

---

使用基于机械 PBM 的非线性模型预测控制的质量控制 (QbC) 框架的实验实施，包括弦长分布测量，用于 L-抗坏血酸的间歇冷却结晶

摘要 l-抗坏血酸是从葡萄糖大规模工业化合成的，并作为免疫系统增强剂和抗氧化成分销售。前体葡萄糖转化为 l-抗坏血酸的总产率是有限的，因此从经济的角度来看，结晶是 l-抗坏血酸生产中至关重要的步骤。人们普遍认为，晶体尺寸分布 (CSD) 会影响最终结晶产品的许多相关宏观特性，并且还会显着影响下游操作。本文讨论了从水溶液中结晶 l-抗坏血酸过程中弦长分布 (CLD，它与 CSD 直接相关) 的控制。采用分批结晶工艺，这是经典的，仍然占主导地位，精细化工和医药行业。介绍了两种最先进的基于质量控制 (QbC) 的结晶设计方法的比较实验研究：(1) 基于直接成核控制 (DNC) 的无模型 QbC 和 (2) 模型使用新型非线性模型预测控制 (NMPC) 框架的基于 QbC。在第一项调查中，应用了 DNC，一种基于最先进模型自由控制策略的过程分析技术。尽管 DNC 需要最少的初步系统信息并且通常提供稳健的过程控制，但由于 l-抗坏血酸的异常结晶行为，它会导致较长的批次时间和振荡操作。在第二项研究中，基于使用 NMPC 方法的基于模型的 QbC 方法的优势得到了证明。通过从浓度和 CLD 测量估计成核和生长动力学，建立和校准基于群体平衡的结晶过程模型。基于投影的 CSD 到 CLD 正向转换用于估计成核和生长动力学。对于鲁棒性和自适应行为，NMPC 与不断增长的水平状态估计器相结合，旨在通过重新调整动力学常数来不断改进模型。该研究表明，基于模型的 QbC 框架可以导致快速和稳健的结晶过程开发，NMPC 系统在显着的工厂模型失配 (PMM) 条件下表现出良好的控制行为。基于投影的 CSD 到 CLD 正向转换用于估计成核和生长动力学。对于鲁棒性和自适应行为，NMPC 与不断增长的水平状态估计器相结合，旨在通过重新调整动力学常数来不断改进模型。该研究表明，基于模型的 QbC 框架可以导致快速和稳健的结晶过程开发，NMPC 系统在显着的工厂模型失配 (PMM) 条件下表现出良好的控制行为。基于投影的 CSD 到 CLD 正向转换用于估计成核和生长动力学。对于鲁棒性和自适应行为，NMPC 与不断增长的水平状态估计器相结合，旨在通过重新调整动力学常数来不断改进模型。该研究表明，基于模型的 QbC 框架可以导致快速和稳健的结晶过程开发，NMPC 系统在显着的工厂模型失配 (PMM) 条件下表现出良好的控制行为。