Purpose

Every successful pharmaceutical product requires a rapid and reliable scale-up process from small laboratory quantities to a large commercial production to ensure the qualities of products. Despite best efforts, a pharmaceutical company must incur losses in material costs and working-hours before achieving optimal process parameters. One possible approach to shorten this tedious process is to leverage knowledge and experience using a computational algorithm that predicts the process parameters using related data.

Method

This study aimed to demonstrate the approach of embedded successful material attributes and process parameters of wet granulation as inputs into an artificial neural network (ANN) model and support vector regression (SVR) model to predict the total number of the impeller revolution for dry mixing and wet massing to produce the granules with the same qualities.

Results

The SVR model performed better for prediction with RMSEs of 3.2552–4.0066. Using the SVR model using Gaussian kernel, which had the least RMSE, in the training stage and validation stage gave the values of MAPE of 17.12% and 12.32%, respectively.

Conclusion

The model can be implemented for the prediction of the number of impeller revolution by using the proposed parameters.

中文翻译：

---

人工神经网络和支持向量回归模型预测湿法制粒混合时间

目的

每个成功的医药产品都需要一个快速可靠的放大过程，从小实验室数量到大规模商业生产，以确保产品的质量。尽管尽了最大的努力，制药公司在实现最佳工艺参数之前必须遭受材料成本和工时的损失。缩短这一繁琐过程的一种可能方法是利用使用相关数据预测过程参数的计算算法的知识和经验。

方法

本研究旨在展示将湿法制粒的成功材料属性和工艺参数嵌入到人工神经网络 (ANN) 模型和支持向量回归 (SVR) 模型中的方法，以预测干混和叶轮的总转数。湿团聚以生产具有相同质量的颗粒。

结果

SVR 模型在 RMSE 为 3.2552–4.0066 的预测中表现更好。使用RMSE最小的高斯核SVR模型，在训练阶段和验证阶段分别给出了17.12%和12.32%的MAPE值。

结论

该模型可以通过使用所提出的参数来实现叶轮转数的预测。