Hybrid Mechanistic and Machine Learning Approaches to Product Performance Prediction

The manufacturing process for pharmaceutical tablets involves a series of process units, designed to create a structured product from often complicated mixtures of pharmaceutical ingredients. The bulk of research in the field has concentrated on creating process models to predict tablet properties, but there's a notable absence of dependable models which depict tablet functionality, known as product performance models, or simply product models. Existing product performance models tend to be either 1) highly empirical, focusing on specific formulation properties or process conditions, or 2) complex and computationally demanding models, which provide excellent detail but are impractical to use extensively in industry. There's a growing demand for mechanistic models which can accurately describe and predict the key rate processes involved in product performance. In order to reduce computational burden, there is an additional need to hybridise these models with machine learning techniques.
Ideally, these models would facilitate the optimization of critical process parameters (CPPs) based on critical quality attributes (CQAs), such as achieving specific disintegration times. This could be accomplished by establishing a connection between the process model and the product performance model through the intermediate stage of tablet structure. Employing an appropriate inverse optimization approach would then allow for the determination of the necessary formulation and process parameters to achieve the desired outcomes. The aim of this research is to develop a coupled mechanistic and machine learning model for the performance of pharmaceutical tablets.