Model construction for drug coating thickness distribution prediction based on Raman spectroscopy

In order to real-time monitor drug coating thickness distribution and overcome the drawbacks of offline measurements of average thickness and weight (which cannot satisfy the requirements of coating uniformity analysis and endpoint judgement of drug coating), a Raman spectrometer was used to collect Raman spectra of table surface in-line and real-time using selenium tablet coating as a reference. Meanwhile, the coating thickness distribution was also statistically analyzed by a microscope. Two modeling approaches including partial least squares (PLS) and convolutional neural network (CNN) were applied to build coating thickness calibration model. The correlation coefficient of determination Rp2 of PLS model was 0.923, and the Rp2 of CNN was 0.996. CNN model shows higher generation capability and accuracy than PLS model. Furthermore, coating thickness distribution predicted by CNN model is in accordance with the offline results (errors of most probable thickness and distribution width at coating time of 60 min were 0.44% and 1.24%, respectively). Therefore, drug coating thickness distribution can be accurately predicted by Raman spectroscopy combined with CNN model. © 2024 Zhejiang University. All rights reserved.