Abstract The aim of this work is to follow the agglomeration mechanism of an in situ Fluid Bed Melt Granulation (FBMG) by means of machine learning with Artificial Neural Network (ANN) modelling. Scanning Electron Microscopy (SEM) was used as a complementary tool with particle size analysis to evaluate the effect of the material properties on the quality attributes of the final granules and provide further insight into the growth mechanisms. The experiments were performed using lactose monohydrate as model filler and two grades of polyethylene glycol (PEG 2000 and 6000) as meltable binders in different contents and in different size fractions, respectively. A multilayer perceptron neural network was developed using MATLAB neural network toolbox and the evaluated quality attributes of the final granules were used as a database for the development and selection of the optimal architecture of the ANN model. The Garson equation was used to quantify the relative importance of each independent variable and thus it was established that the particle size had the highest impact on the granule properties. The distribution and immersion granule growth mechanisms were determined to occur for low and high binder particle size, respectively, as confirmed by the SEM pictures, and the response surfaces helped determine the optimal design space of the process and the optimal FBMG conditions.

中文翻译：

---

使用机器学习工具阐明和表征原位流化床熔体造粒的生长机制

摘要 这项工作的目的是通过机器学习和人工神经网络 (ANN) 建模来跟踪原位流化床熔体造粒 (FBMG) 的凝聚机制。扫描电子显微镜 (SEM) 被用作粒度分析的补充工具，以评估材料特性对最终颗粒质量属性的影响，并提供对生长机制的进一步了解。分别使用乳糖一水合物作为模型填充剂和两种等级的聚乙二醇（PEG 2000 和 6000）作为不同含量和不同尺寸分数的可熔粘合剂进行实验。使用 MATLAB 神经网络工具箱开发了多层感知器神经网络，并将最终颗粒的评估质量属性用作开发和选择 ANN 模型最佳架构的数据库。Garson 方程用于量化每个独立变量的相对重要性，因此确定粒度对颗粒特性的影响最大。SEM图片证实，分布和浸入式颗粒生长机制分别在低和高粘合剂粒径下发生，响应面有助于确定工艺的最佳设计空间和最佳FBMG条件。Garson 方程用于量化每个独立变量的相对重要性，因此确定粒度对颗粒特性的影响最大。SEM图片证实，分布和浸入式颗粒生长机制分别在低和高粘合剂粒径下发生，响应面有助于确定工艺的最佳设计空间和最佳FBMG条件。Garson 方程用于量化每个独立变量的相对重要性，因此确定粒度对颗粒特性的影响最大。SEM图片证实，分布和浸入式颗粒生长机制分别在低和高粘合剂粒径下发生，响应面有助于确定工艺的最佳设计空间和最佳FBMG条件。