Institute of Pharmaceutics and Biopharmaceutics, Heinrich Heine University, Universitaetsstrasse 1, 40225 Duesseldorf, Germany.
Institute of Pharmaceutics and Biopharmaceutics, Heinrich Heine University, Universitaetsstrasse 1, 40225 Duesseldorf, Germany.
Eur J Pharm Biopharm. 2020 Aug;153:43-51. doi: 10.1016/j.ejpb.2020.05.018. Epub 2020 May 20.
Raman spectroscopy was used as an in-line PAT tool to predict the applied coating mass of three different coating layers on caffeine cores. The different coating suspensions contained titanium dioxide in the anatase and rutile modification and iron oxide as Raman markers. Partial least squares-regression (PLSR) and multivariate curve resolution-alternating least squares (MCR-ALS) were used for multivariate analysis. The acquired Raman spectra were correlated to the applied coating mass. MCR-ALS models were built and applied offline, while PLS-regression was implemented in the coating process to enable a real-time monitoring. Inline-measurements were optimized by a higher frequency of the spectral measurements and the implementation of a moving average. By PLS-regression analysis, all three layers could be predicted with root mean square errors (RMSEP) of less than 2.3%. Inline implementation and optimization resulted in RMSEPs less than 1.9%. MCR-ALS analysis was able to predict the application of the first and the second layer with RMSEPs less than 2.9%, but failed in predicting the application of the third layer. In conclusion, a real-time monitoring of a multi-layered coating process was achieved, PLS-regression was found to be superior to MCR-ALS and smoothing by the implementation of a moving average enhanced the predictability.
拉曼光谱被用作在线过程分析技术 (PAT) 工具，以预测三种不同涂层在咖啡因核上的应用涂层质量。不同的涂层悬浮液含有锐钛矿和金红石改性的二氧化钛和氧化铁作为拉曼标记物。偏最小二乘回归（PLSR）和多变量曲线分辨交替最小二乘法（MCR-ALS）被用于多变量分析。获得的拉曼光谱与施加的涂层质量相关联。建立了 MCR-ALS 模型并离线应用，而 PLS 回归则在涂层过程中实施，以实现实时监测。通过增加光谱测量的频率和实施移动平均，对在线测量进行了优化。通过 PLS 回归分析，可以预测所有三层，其均方根误差（RMSEP）小于 2.3%。在线实施和优化导致 RMSEPs 小于 1.9%。MCR-ALS 分析能够以小于 2.9%的 RMSEP 预测第一层和第二层的应用，但无法预测第三层的应用。总之，实现了对多层涂层过程的实时监测，发现 PLS 回归优于 MCR-ALS，通过实施移动平均进行平滑处理提高了预测能力。
