Testing on continuous production of mefenamic acids - design of experiment through simulation and process optimisation

In the pharmaceutical manufacturing industry, continuous production methods have been recognised as providing several benefits compared to traditional batch production. These benefits include increased flexibility, higher product output, enhanced quality assurance through better monitoring techniques, and more consistent distribution of Active Pharmaceutical Ingredients (APIs). Despite these clear advantages, there is a lack of research focused on the simultaneous optimisation of multiple sub-processes in continuous manufacturing. This study explores the optimisation processes of continuous pharmaceutical production, explicitly targeting the production of mefenamic acid using wet milling (WM) and mixed-suspension mixed-product removal (MSMPR). We employ data-driven evolutionary optimisation algorithms to address these many-objective optimisation problems (MaOPs). High-fidelity model-generated data generated via the General Process Modelling System (gPROMS) is subsequently utilised to develop simpler surrogate models based on the Radial Basis Function Neural Network (RBFNN). This enables very fast simulations, suitable for use with computationally intensive machine learning algorithms. Utilising evolutionary optimisation algorithms, these models are used for model-based process optimisation. The efficacy of the MaOP approach is evaluated using a range of numeric and visual optimisation performance indicators. Our findings underscore the viability of integrating high-fidelity and surrogate models to discern functional relationships between dependent variables (objective functions) and independent variables (decision variables), providing a robust framework for process optimisation within the pharmaceutical domain. The approximated solutions are, on average, 58% better than the solutions obtained from Latin hypercube sampling. The chosen optimal solutions can form the basis of parameter setting in upcoming experimental campaigns. The significance of this work is in the demonstration, for the first time, of a many-objective optimisation framework for continuous pharmaceuticals production using simple surrogate models derived from high fidelity simulations using Machine Learning.