Analysis of continuous manufacturing process of oral solid dosage using neural network

Abstract

In this study, a neural network (NN) was applied to a continuous manufacturing process of tablets to predict the tablet physical properties based on a twin-screw granulation operating conditions and tableting pressure. The hyperparameters in the NN model were optimized to accurately predict the tablet physical properties. The constructed NN model successfully demonstrated the predictive capability with the R2 of ca. 0.9 in both training and validation. The effects of the granulation operating conditions and the tableting pressure on the tablet physical properties were investigated. It was found that the tableting pressure was the most dominant factor for the tablet hardness and disintegration time. Among the granulation operating conditions, liquid solid ratio had the strongest impact on the tablet physical properties. Focusing on the tableting pressure and liquid solid ratio, the contour maps for the relationships between the operating conditions and the tablet physical properties were obtained by the NN model. It was suggested that the obtained contour maps can be helpful to predict the continuous manufacturing of tablets with the desired tablet physical properties.