REAL TIME CONTROL FOR CONTINUOUS PROCESS
In biopharmaceutical manufacturing, the FDA has advocated the need for enhanced online monitoring and control methods to ensure consistent product quality throughout manufacturing processes (1, 2). A parallel effort is echoed by the European Medicines Agency (EMA), which has issued recommendations on real-time release testing (3). As a result, the biopharmaceutical industry is placing significant emphasis on the development of process monitoring and automated process control strategies. One notable approach, proposed by Myerson et al., involves the development of a digital twin system with relevant variables, running in parallel with operations to predict critical quality attributes in real time (4). Other works include Lu et al. presented a case study involving the shadowing of major unit operations, such as a perfusion bioreactor, packed bed chromatography separation train, and in-line dilution, using mechanistic models (5). Feidl et.al. combined mechanistic models with Raman Spectroscopy to predict the breakthrough curve and monoclonal antibody (mAb) concentration in a chromatography protein capture step (6).
In response to the increased demand for higher productivity and lower capital costs through equipment utilization optimization, continuous manufacturing has emerged as an advanced alternative to traditional batch processing (7). The FDA’s emphasis for integration of continuous processes within biopharmaceutical manufacturing has further catalyzed the industry’s adoption of this approach (8). In the context of continuous processes, implementation of robust control strategies has an essential role in ensuring process stability. Product quality consistency over time is critical as process perturbations can impact a larger product pool accumulated over time. Rooted in thermodynamics and kinetics, the mechanistic model based digital twin approach enhances the understanding of process dynamics. This advanced insight empowers the formulation of robust control strategies that effectively mitigate the impacts of process variability and disturbances originating from raw materials or upstream unit operations (9).
This paper is centered around the digital twin framework and a conceptual application of the digital twin concept to the domain of continuous biomanufacturing with the goal of enhancing product quality. It is important to note that the model presented serves as a demonstration, utilizing hypothetical data to illustrate the potential applicants.
PROCESS DIGITAL TWIN MODELING FRAMEWORK
The digital twin concept origins from the broader field of cyber-physical systems and the Internet of Things (IoT). Nowadays, “digital twin” has become a buzz word across diverse industries and functions due to its diverse array of applications, as a result, it becomes essential to provide a concise definition of the digital twin concept and its practical implementation. At Resilience, a digital twin framework has been defined, including three different levels of models (Figure 1):
Digital Model: The foundational layer of the digital twin framework is digital model. Digital model focuses on the localized phenomena in each unit operation. For instance, the distribution of hydrodynamic properties inside of a bioreactor. Digital model normally solves 2D or 3D equations to capture the spatial phenomenon. Digital model serves an array of areas such as unit operation scale-up studies, process optimization to enhance mixing and KLa, and conduct design equivalency assessments.
Digital Shadow: Digital shadow lumps the spatial characteristics and focuses on the system-level phenomenon and operations of unit operations. It envisions the impact of varied system level factors such as feeding strategy, temperature control, elution strategy, pH control and other process parameters to the unit operation’s output. Digital shadow normally solves 0D or 1D equations, which reduces the computation time which empowered its use for more complex conditions like interlinking digital shadows of multiple unit operations into an end-to-end digital shadow. The application of digital shadow covers an extensive area, such as process optimization, process characterization, root cause analysis, and predictive process control with manual adjustments.
Digital Twin: Digital twin is a cohesive system that interconnects the digital shadow with the physical plant through the strategic implementation of Process Analytical Technology (PAT). In this system, PAT captures process parameters in real time and seamlessly transmitted to the digital shadow. With the real time process data from the physical plant, the digital shadow predicts the process outputs which can be used to alert operators to make process adjustments before the product quality attributes drifts out of the control limit. Through the real tie interactions between the digital twin and the physical plant, automated real time process control can be established.
The clarification of these levels within the digital twin framework serves to provide a comprehensive explanation of the relationships and functionalities of concepts under digital twin. This paper covers a conceptual case study of digital shadow and offers a profound exploration of its potential to improve process robustness through predictive process control.
A DIGITAL SHADOW CASE STUDY
Under the concept of continuous bioprocessing, a typical process consists of cell culture, periodic-counter-current chromatography (PCCC) and filtrations to achieve uninterrupted manufacturing processes. Below is the integrated end-to-end continuous drug substance manufacturing platform developed by Resilience for monoclonal antibodies and other biological molecules (Figure 2).
In the context of a continuous process as described, the implementation of an end-to-end digital shadow framework can be transformative. Such framework brings in significant value across process lifecycle and offers a pathway to achieve optimal operating conditions with reduced amount of experimentation to establish and expand process knowledge. Moreover, the real-time detection of deviations in critical process parameters can trigger the use of residence time distribution analysis built into the digital shadow platform. This dynamic approach allows for the identification of the duration of product diversion in response to process perturbations, enabling the segregation of impacted products without affecting the remainder of the batch.
END-TO-END DIGITAL SHADOW DESCRIPTION
To achieve the concept of digital shadow, individual mechanistic unit operation models were constructed and subsequently interconnected in a flowsheet model using Siemens gPROMS FormulatedProducts® software. Several key unit operations are detailed below.
Perfusion Bioreactor: Within this unit operation, continuous protein production is achieved through a cell retention device such as ATF filter. Cells are recycled back to the bioreactor, while the product is directed downstream. Steady state is achieved through the introduction of a bleed stream, and the product is collected within a surge vessel which is fed to the continuous capture operation. The bioreactor is assumed to be well-mixed with monogenous properties. Mass balance is solved alongside kinetics for cell growth/cell death, and kinetics for nutrient consumption and metabolite secretion.
Chromatography Unit Operations: For mAb, Chromatography unit operations contain Protein A, Cation Exchange Chromatography (CEX) and Anion Exchange Chromatography (AEX). Employing multi-column chromatography mode, these operations achieve continuous capture and purification. The continuous capture system produces a frequent discrete production elutions, which are collected into a surge vessel to form a homogeneously mixed transient protein A elution pool that feeds to low pH inactivation step. To achieve the PCCC process, a multicolumn logic needs to be defined based on the timing of each step. In the model, the radial direction concentration gradient is neglected and the intraparticle mass transfer can be lumped into the apparent axial dispersion coefficient or kinetic adsorption isotherm parameters. Convection and axial dispersion are considered in the mass transfer equation.
Filtration Unit Operations: Filtration unit operations comprise depth filtration, viral filtration, single pass tangential flow filtration (SPTFF), and Ultrafiltration/Diafiltration. The membrane flux through the membrane is governed by Darcy’s law, where permeate pressure drives mass flow, and membrane selectivity is defined by the retention factor.
Through the connection of these unit operations from perfusion bioreactor to SPTFF, the digital shadow provides dynamic insights into the output and performance of each unit operation throughout the duration of the manufacturing process (Figure 3).
The prediction mechanism capabilities are critical in managing disturbances or deviations encountered during the continuous manufacturing process. When a disturbance occurs, the digital shadow system can be used to predict the duration and severity of the impact across subsequent unit operations. The prediction can be used to segregate the product stream and hence minimize the impact of process disturbance on product quality. Once digital twin is achieved by connecting the real time process data to with the digital shadow, an automated real-time monitoring and control strategy can be achieved. As a result, this approach contributes to the establishment of a resilient manufacturing facility, ensuring robust and consistent product quality.
DEVIATION CASE STUDY
To simulate a common deviation scenario, dimers was introduced into the surge tank before CEX unit operations. This case study demonstrated the digital shadow’s predictive capabilities in managing deviations (Figure 4). The fluctuating chromatogram peaks and surge tank concentrations of product protein and aggregates, following the process perturbation, effectively demonstrated the framework’s capability in tracing the progression of the disturbance across unit operations. In the figure below, after dimer was introduced to the process, a larger dimer peak and smaller product mAb peak was observed from the CEX chromatogram, and an increase in dimer concentration and decrease in product mAb concentration is observed in the CEX product tank. Around half a day later, the same change is observed in the next unit operation: AEX.
These predictive capabilities enable a comprehensive understanding of process capability and performance, thereby ensuring compliance with product quality acceptance criteria. Should any out-of-specification event be predicted, the digital shadow framework enables the establishment of control strategies that facilitates operation adjustments and products segregation. In this case study, a specification limit on dimer concentration can be added to the CEX and AEX product tank, and the decision and schedule for product redirection can be determined.
When the detected deviation can be transferred to the digital shadow in real time and achieve a digital twin framework, it can seamlessly integrate real-time insights with predictive ability, improving the reliability, control, and product quality assurance for the continuous biomanufacturing process (Figure 5).
FROM DIGITAL SHADOW TO DIGITAL TWIN
An opportunity to benefit from industrial automation systems and cloud infrastructure
Development of the digital shadow and evolving it to digital twin relies on the availability of data acquisition and sensing capabilities. On-going modeling uses built-in sensor data available but an untapped opportunity here is to identify the limitations of the built-in sensor data. Then the model fidelity can be enhanced by identifying and integrating important additional sensors (high value peripheral sensors). The sensor data pipelines would be integrated with Supervisory Control and Data Acquisition (SCADA) such as Ignition and connect to the cloud services through Ignition or Internet of things (IoT) Greengrass platform. Within the cloud services environment, models can be built and deployed as infrastructure-as-code (IaC). Such IaC style design, development, and deployment will allow model scalability and enhancements for varying unit operations (e.g., chromatography model vs. bioreactor model, etc.) and vendors (e.g., Sartorius system vs. Cytiva system, etc.). Key research questions that should be answered are as follows:
How do we get continual input from the physical systems into the Digital Twin (DT)? Is there a low cost and automated data flow using public cloud service?
Can a machine learning model be used/developed to produce “digital signatures” and inferences? If an adversary exploits any zero-day vulnerabilities in the algorithm, how can one protect the algorithm against cyberattacks?
How can one scale the models and pertinent prediction algorithms for other biomanufacturing systems with little or minor modifications?
How can one verify the performance of the digital twin through correlation between its predictions and the physical system events for given initial and progressing conditions? In other words, if a combination of synthetic and real data is used, what are the controllable parameters to calibrate the digital twin models and the corresponding machine learning models to use their predictions and drive actions on real-world systems?
Building a robust sensor data pipeline with communications to the digital shadow could lead to resilient and scalable digital twin. An illustrative infrastructure is shown in the figure below to depicts the relationship between the physical and digital systems and pertinent dataflows. The green box with the digital twin or digital shadow and machine learning process boxes could be hosted on cloud infrastructure (e.g., AWS as shown below. However, the model components may also be tested on GCP or Azure, etc.). The physical system would be a manufacturing system with electrical and mechanical subsystems, built-in and peripheral sensors. A real-time feedback loop would be in place between the digital and physical systems to ensure that the model behavior is optimized, and the predictions can be used in correlation with the real-system. The machine learning process would focus on generating synthetic data, digital signatures, and inferences by consuming the real data and the digital twin model’s physics-based processes.
Exploring to the realm of Biomanufacturing Cybersecurity and Cyber-resiliency
Cybersecurity is vital to ensure safe and reliable manufacturing processes. The importance of security and resiliency are the emphasis of the Executive Order (EO) 14081 (10), Advancing Biotechnology and Biomanufacturing Innovation for a Sustainable, Safe, and Secure Bioeconomy, and digital twins have the potential to meet several aspects of the EO. The cybersecurity and cyber-resiliency benefits offered by digital twins is often discussed in the literature as well. For instance, Mylrea et al., (11) discusses the combination of consequence-driven cyber-informed engineering with digital twins to facilitate cyber-secure operations of a biomanufacturing process. Testing for cyber security implications on live manufacturing networks and processes comes with significant safety and business risks. Digital twin offer a promising platform to test a biomanufacturing process for security implications. For instance, accurate replica of a physical manufacturing process will allow to test for implications due to parametric changes to the network and the systems. In addition, one can test for feasibility of such manipulations within the system and network and the implications on the manufactured end-product. Such studies will shed light on manufacturing network and system-level augmentations to be performed to ensure secure and safe operations.
CONCLUSIONS
In biopharmaceutical manufacturing, the digital shadow platform enables the prediction of process performance and product quality, facilitating virtual exploration of the process design space. This data-driven, science-bases, and risk-aware approach serves as a foundation for the integrated continuous biomanufacturing process control, they are key elements for biomanufacturing to reach digital plant maturity model (DPMM) (12), please see appendix for details. With the conceptual demonstration of an application of digital shadow in a continuous manufacturing environmental, we aim to underscore the substantial benefit of leveraging digital twin technology in the biopharmaceutical industry for the process’s resilience to disturbance and uncertainties and an improved product quality. Future work includes developing a digital twin platform by generating the connections between the virtual model with the physical manufacturing process to achieve fully automated real-time process control.
ACKNOWLEDGEMENT
The authors would like to acknowledge Sarah Fadda (Siemens), Edward Close (Siemens), Nehal Patel (Siemens), Christopher Polster (Siemens), Tom Ransohoff (RTI Consulting, LLC.), Michael Mylrea (Cyber Team 7, LLC.), Blair Okita (University of Virginia), Behnam Partopour (Sartorius), Tanveer Ahmed (National Resilience, Inc.) for all the help and inspirations.
Figure 1. Process Digital Twin Framework at Resilience.
Figure 2. Integrated end-to-end continuous drug substance manufacturing platform developed by Resilience
Figure 3. Unit Operation outputs at all time point from In-Silico Model
Figure 4. Model output for an aggregation deviation case study.
Figure 5. Illustration of digital twin deployment infrastructure.
Table 1. Where digital twin can map into DPMM.
REFERENCES AND NOTES
- FDA, “Guidance for Industry: PAT—A Framework for Innovative Pharmaceutical Development, Manufacturing, and Quality Assurance.,” U.S. Department of Health and Human Services, Food and Drug Administration., 2004.
- J. Glassey, K. V. Gernaey, C. Clemens, T. W. Schulz, R. Oliveira, G. Striedner and C. F. Mandenius, “Process analytical technology (PAT) for biopharmaceuticals,” Biotechnology Journal, pp. 6(4), 369, 2011.
- EMA, “Guideline on Real Time Release Testing (formerly Guideline on Parametric Release).,” Europeans Medicines Agency, Committee for Medicinal Products for Human Use, 2012.
- A. S. Myerson, M. Krumme, M. Nasr, H. Thomas and R. D. Braatz, “Control systems engineering in continuous pharmaceutical manufacturing,” Journal of Pharmaceutical Sciences, pp. 104, 832–839, 2015.
- A. E. Lu, J. A. Paulson, N. J. Mozdzierz, A. Stockdale, A. N. Ford Versypt, K. R. Love, J. Christopher Love and R. D. Braatz, “Control Systems Technology in the Advanced Manufacturing of Biologic Drugs,” in IEEE Conference on Control Applications (CCA), Sydney, 2015.
- F. Feidl, S. Garbellini, M. Luna, S. Vogg, J. Souquet, H. Broly, M. Morbidelli and A. Butté, “Combining mechanistic modeling and Raman spectroscopy for monitoring antibody chromatographic purification,” Processes, pp. 7(10), 683, 2019.
- H. Mahal, H. Branton and S. S. Farid , “End‐to‐end continuous bioprocessing: Impact on facility design, cost of goods, and cost of development for monoclonal antibodies,” Biotechnology and Bioengineering, p. (118) 3468–3485, 2021.
- S. Chatterjee, “FDA Perspective on Continuous Manufacturing,” in IFPAC Annual Meeting, 2012.
- K. Ahmed, S. Chatterjee, C. Garvin, K. Ghosh, M. Ierapetritou, M. Jeffers, D. L. Pla, S. L. Lee, D. Lovett, O. Lyngberg, J. Mack, E. McManus, S. Romero-Torres, C. Undey and Venkatasubramanian, “Moving APC Forward in the Pharmaceutical Industry,” in AIChE PD2M Advanced Process Control (APC) Workshop, Washington DC, 2019.
- J. R. Biden Jr., “Executive Order on Advancing Biotechnology and Biomanufacturing Innovation for a Sustainable, Safe, and Secure American Bioeconomy,” The White House, 2022.
- M. Mylrea, C. Fracchia, H. Grimes, B. Reid, N. Case, W. E. Austad and G. E. Shannon, BioSecure Digital Twin: Manufacturing Innovation and Cybersecurity Resilience, 2011.
- BioPhorum Operations Group, “Digital plant maturity model,” [Online]. Available: https://www.biophorum.com/resource/digital-plant-maturity-model/.
- APPENDIX
- Tools and frameworks such as the Digital Plant Maturity Model (DPMM) (BioPhorum Operations Group, n.d.) assists biomanufacturing facilities to track their digital transformation journey. Digital twin is a crucial tool in such journey and can be directly mapped to most controls in DPMM levels 3 to 5 (Table 1).
