Abstract
Continuous manufacturing can be seen as a promising shift in the pharmaceutical industry, offering benefits such as reduced costs and improved product quality. However, the multistage nature of continuous tablet manufacturing demands a deeper understanding of the complex interactions between process parameters, material attributes, and final product quality. This study aims to address this challenge by developing a novel, data-driven modelling framework to predict key critical quality attributes, including particle size distribution, moisture content, and tablet tensile strength across the processing stages of a pilot-scale continuous tablet manufacturing line. A sequential modelling approach was employed, integrating Random Forest and Gradient Boosting Machines to model each processing stage. These models were sequentially trained and interlinked to holistically capture process–material interactions across granulation, drying, milling, and tabletting stages. To manage error propagation between stages, Gaussian Mixture Models were incorporated for error characterisation and uncertainty reduction. The results showed that the proposed framework captured the non-linear interactions between processing parameters and the quality attributes. The incorporation of GMMs was influential in quantifying uncertainty within each process model, resulting in a final estimation of tablet tensile strength with an $$ R^2 $$ R 2 value of 0.90 using the integrated Random Forest model. This framework demonstrated considerable improvement in the predictive performance of the continuous manufacturing processes modelling through the integration of machine learning models and an uncertainty-aware strategy. The predictive tool is intended to support the Quality by Design (QbD) concept through systematic design space exploration and process understanding of the pharmaceutical continuous manufacturing.