Thesis : “PROCESS ANALYTICAL TECHNOLOGY (PAT) APPLICATIONS IN HARVEST UNIT OPERATIONS” as per the following schedule:
Date : 20th December 2024
Venue : Block II Room 276 (Chemical Engineering Seminar Room)
Time : 1100 am
Abstract:
In today's technological era, smart factories within manufacturing industries are not a distant prospect. Industrial sectors are shifting toward digitalization and automation as a precondition for Industry 4.0. In spite of its remarkable rise, which saw it reach a $188 billion sales value in 2017, the biopharmaceutical industry clearly trails behind in this shift. Data analytics and digitalisation have been recognized as crucial strategic initiatives for enabling continuous manufacturing and faster and cheaper biotherapeutic production. Initiatives such as Quality-
by-design (QbD) and Process Analytical Technology (PAT) are considered white knights of the biopharmaceutical industry for ensuring consistent product quality in a complex manufacturing process and the key towards process automation. This thesis is centred around the application of PAT in the Chinese hamster ovary (CHO) cell harvest bioprocessing area using CAS, Pall Lifesciences, USA. The study involves rapid monitoring of key process parameters (KPPs) using online turbidity sensors and employing mechanistic, empirical and hybrid modelling approaches to achieve real-time monitoring, control, optimization and integration of the unit operation in continuous biopharmaceutical manufacturing. The first work (Chapter 3) involves KPP identification in CAS, which affects the two primary process outputs: cell separation efficiency (CSE) and protein recovery (PR). A mechanistic model based on the population balance model (PBM) along with an acoustic agglomeration kernel was developed to predict cell size distribution as a function of time, resulting in CSE estimation. While an empirical model using the design of experiments (DoE) approach was used to screen out KPP affecting PR. Results show that > 90% CSE was achieved for feed flowrate (FF) 12.5 W while > 90% PR was obtained for FF >15 mL min-1. Both the models combined formed a hybrid model, and experimental results validate its accuracy in predicting the CHO agglomeration process. Furthermore, a generalized correlation has been developed using the Buckingham Pi theorem between the dimensionless numbers composed of relevant processes and geometrical variables to predict the agglomeration efficiency. To our knowledge, this is the first such study to model the agglomeration dynamics of CHO cells in acoustic standing waves.
The next study (Chapter 4) proposes a Digital twin (DT) and a cyber-physical system (CPS) for CAS. While the CPS - DT serves as the core, a real-time model based on the developed hybrid models has been implemented in tandem to present a highly reliable hybrid model to improve the model's adaptability. The framework employs a distributed control system (DCS), the heart of the CPS architecture, that updates the parameters of the physical system in real-time, thereby simplifying process control. To evaluate the hybrid-model predictive control strategy in real time, three case studies were conducted, which involved introducing abrupt turbidity pulses of high, low, and multi values. The sudden deviation in the turbidity of the feed sample was controlled by adjusting the AP of the chamber (the control variable). The use of the proposed controller resulted in a 5 % mean deviation from the setpoint in the first chamber, while a > 33 % mean deviation resulted in the absence of the controller. The study's outcome demonstrates the high effectiveness of hybrid model-based CPS - DT control in consistently achieving high CSE (> 90 %) and integrating the unit operation in a continuous processing train to establish a smart manufacturing prototype for the biopharmaceutical industry. Following development of CPS architecture, the next study (Chapter 5) introduces a methodology for developing a holistic smart manufacturing framework comprising 4 interactive modules. These modules deal with process control, prognosis, and diagnosis and act as an external communication platform using different methodologies. The architecture, real-time monitoring, controller experiments, and detailed discussions are presented to implement the automatic process control (APC) algorithm on CAS. The CPS framework was designed to address the aforementioned objectives and was enhanced with deep learning-based neural networks at various levels. The proposed CPS controller resulted in a 5% mean deviation from the setpoint in the first chamber, while more than 33% resulted in the controller's absence. In addition, the time taken by the framework to monitor and implement control action is 10 s, leading to robust process regulation even in case of a minor deviation. In the absence of the CPS framework, the monitoring becomes purely manual and is prone to significant process deviation. If a disturbance occurs, the error can propagate downstream due to delayed implementation of corrective action. It is envisaged that the proposed CPS framework powered by ML models can provide solutions for a compendious ‘smart’ approach for real-time process monitoring and control of all unit operations in the manufacturing industry.
Finally, after the development of a comprehensive framework for monitoring, multi-objective optimization was performed on the acoustic clarification using CAS (Chapter 6). Three objectives, namely CSE, PR and residence time (RT) of harvest, were evaluated using a hybrid model approach. Next, the non-dominated sorting genetic algorithm (NSGA-II) was successfully employed for tri-objective optimization using KPPs, namely FF, CF and load cell density (LCD) as the decision variables. The optimization study was performed for three different configurations of the acoustic chambers: series, series–parallel, and parallel to obtain the optimal design and set of solutions. The results of the NSGA-II algorithm revealed the relationships between CSE, PR, and RT and evaluated a tradeoff and an optimal balance between them was achieved. The series–parallel configuration was evaluated as the most optimal design, giving CSE: 80.20 %, PR: 82.61 % and RT: 10.42 mins at an LCD of 30 × 106 cells mL−1. An application study was conducted to obtain the optimal solution set at different LCDs in a single run. Further, controller studies were also performed to apply corrective actions in case of feed turbidity disturbances in order to maintain a steady state. Overall, this is the first attempt to develop a multi-objective optimization study for CHO cell clarification, and this may be utilized as a tool for simulation, design, optimization, and control of any cell harvest unit operation.