Milanesi Andrea, Rizzuto Francesco, Rinaldi Maurizio, Foglio Bonda Andrea, Segale Lorena, Giovannelli Lorella
Department of Pharmaceutical Sciences, Università del Piemonte Orientale, 28100 Novara, Italy.
Department of Mechanical and Aerospace Engineering, University of Strathclyde, Glasgow G1 1XQ, UK.
Pharmaceutics. 2022 Jan 27;14(2):296. doi: 10.3390/pharmaceutics14020296.
The use of design space (DS) is a key milestone in the quality by design (QbD) of pharmaceutical processes. It should be considered from early laboratory development to industrial production, in order to support scientists with making decisions at each step of the product's development life. Presently, there are no available data or methodologies for developing models for the implementation of design space (DS) on laboratory-scale spray dryers. Therefore, in this work, a comparison between two different modeling approaches, thermodynamics and computational fluid dynamics (CFD), to a laboratory spray dryer model have been evaluated. The models computed the outlet temperature (Tout) of the process with a new modeling strategy that includes machine learning to improve the model prediction. The model metrics calculated indicate how the thermodynamic model fits Tout data better than CFD; indeed, the error of the CFD model increases towards higher values of Tout and feed rate (FR), with a final mean absolute error of 10.43 K, compared to the 1.74 K error of the thermodynamic model. Successively, a DS of the studied spray dryer equipment has been implemented, showing how Tout is strongly affected by FR variation, which accounts for about 40 times more than the gas flow rate (Gin) in the DS. The thermodynamic model, combined with the machine learning approach here proposed, could be used as a valid tool in the QbD development of spray-dried pharmaceutical products, starting from their early laboratory stages, replacing traditional trial-and-error methodologies, preventing process errors, and helping scientists with the following scale-up.
设计空间（DS）的应用是制药工艺质量源于设计（QbD）的关键里程碑。从早期实验室研发到工业生产都应予以考虑，以便在产品开发生命周期的每个阶段为科学家提供决策支持。目前，尚无用于在实验室规模的喷雾干燥器上开发设计空间（DS）实施模型的数据或方法。因此，在本研究中，评估了两种不同建模方法——热力学和计算流体动力学（CFD）——对实验室喷雾干燥器模型的比较。这些模型采用一种新的建模策略计算该工艺的出口温度（Tout），该策略包括机器学习以改善模型预测。计算得出的模型指标表明，热力学模型比CFD模型更能拟合Tout数据；实际上，CFD模型的误差随着Tout和进料速率（FR）值的升高而增大，最终平均绝对误差为10.43 K，而热力学模型的误差为1.74 K。随后，实施了所研究喷雾干燥器设备的设计空间（DS），结果表明Tout受FR变化的强烈影响，在设计空间（DS）中，FR变化的影响比气体流速（Gin）大40倍左右。结合本文提出的机器学习方法的热力学模型，可作为喷雾干燥药品QbD开发的有效工具，从早期实验室阶段开始，取代传统的试错方法，防止工艺误差，并帮助科学家进行后续的放大生产。
