DescriptionWith the recent technological advancements in digitization and Industry 4.0 practices, the pharmaceutical industry is moving towards integrating these advancements within the drug manufacturing. This integration is termed as ‘Pharma 4.0’. A crucial component of this integration involves the development of digital twins of the manufacturing lines. Regulatory agencies are collaborating with academic and industrial partners to develop digital twins and move towards Pharma 4.0, with the aim to develop robust, flexible, and agile manufacturing lines. However, despite these efforts, there are some crucial aspects that need to be addressed to develop accurate and integrated digital twins. These include and are not limited to the accuracy of model prediction, computational time of process simulation, and handling and pre-treatment of experimental data from the manufacturing line. The work presented in this thesis focuses on these aspects, with the research goal being to develop computationally efficient multi-scale process models that incorporate detailed particulate dynamics for obtaining accurate process prediction along with statistical analysis and data pretreatment of experimentally obtained datasets. The research goal is aimed towards developing digital twin of the manufacturing lines and the respective objectives are outlined in chapter 2. The first aim of this thesis focuses on developing high-fidelity particulate simulation of pharmaceutical unit operations to understand detailed particle-level mechanics. This is demonstrated for a powder feeder unit, being located at the top of the manufacturing lines and a key contributor to the observed variability in powder flow. Chapter 3 presents the simulation of powder feeder using discrete element modeling (DEM), along with calibration of DEM parameters.
The second aim of the thesis is outlined in chapters 4 and 5, where the detailed particulate information is used to develop computationally efficient process models for digital twin applications, to improve the overall prediction of manufacturing lines. This is demonstrated for a continuous powder blender. This aim also includes a detailed review on mixing indices which are important to quantify the levels of mixing or blend uniformity within the blender system.
Lastly, the third aim of the thesis focuses on the physical component of digital twin involving data pre-treatment and analysis. This aim includes two aspects, outlined in chapter 6. First aspect involves the statistical data handling and denoising strategies for experimental residence time distribution (RTD) of manufacturing lines, given the high degree of variability in the datasets, which can significantly affect the process understanding and lead to erroneous conclusions. The second aspect of this aim focuses on optimizing the efforts required to obtain RTDs experimentally. This is investigated by identifying the important features of RTD from a quality assurance perspective, followed by directing efforts to capture them accurately. The proposed approach thus optimizes the required experimental efforts while ensuring regulation compliant manufacturing of drug products.