Bioprocess Control: Soft Sensors, PAT And Automation

A typical bioprocess consists of different unit operations where an optimal environment is required for cells to grow, divide and synthesize desired products. However, bioprocess control encounters unique challenges due to the nonlinearity, variability, and complexity of biotechnology. This article describes modern control strategies employed in bioprocessing. The article will introduce traditional control strategies (open loop, closed loop) as well as modern control schemes such as fuzzy logic, model predictive control, adaptive control, and neural network based control, and highlight their effectiveness. Furthermore, bioprocess control is not just about automation, but also includes aspects such as system architecture, software application, hardware and interfaces, all of which need to be optimized and compiled as needed. This needs to be done while maintaining process requirements, production costs, product market value, regulatory constraints and data collection requirements. This article aims to provide an overview of current best practices in bioprocess control, monitoring, and automation.

Part 1 Soft Sensor Based Control

Given the complexity of bioprocesses, modeling that captures overall features and variability remains a challenge. In bioprocessing, most process variables are evaluated off-line or near-line due to the still limited availability of reliable in-line sensors, leading to increased overall system cost and inefficient information transfer. In addition, long-term operation and calibration of online sensors in fluctuating and harsh environments, as well as manual operations of near-line sensors at different detection frequencies, still pose considerable challenges in obtaining data. This drives the development of cybernetic physical systems, where integrated physical systems are controlled by soft sensors and algorithms. Soft sensors have emerged as potential tools for evaluating and maintaining CQA in an online mode, thereby enabling QbD. Different studies have shown that soft sensors can be used to measure biomass, products, metabolites, amino acid concentrations, and other CQAs for process control. Although great progress has been made in the development of soft sensors, their real-time implementation requires further development of non-invasive analytical techniques capable of in situ or real-time monitoring; sensor devices suitable for various production systems and ensuring sensor configurations for regulatory compliance and Good Manufacturing Practice. In addition, real-time, user-friendly interfaces are required to connect and contextualize information from different sensors through digitization.

Advances in sensor miniaturization, the development of smart sensors and hardware-software integration methods (digital highways such as fieldbus/profibus or wireless) offer additional advantages for their implementation in production. Industry 4.0 requires computer algorithm-based monitoring and control of cyber-physical systems. If the connection between soft sensors and process system engineering is thoroughly and carefully studied, then it can be said that with soft sensors, Industry 4.0 can become a reality and has a bright future in bioprocessing.

Part 2 PAT-based Control Strategy

The task of the controller is to manipulate the process variable in such a way that disturbance effects are minimized and the process variable follows a specified trajectory. As mentioned earlier, traditional feedback PI controllers are widely implemented in industry, followed by cascade control strategies. So far, the industrial application of advanced control strategies such as multivariable control, model-based control, and adaptive control is still limited. Although significant progress has been made over the past two decades, more needs to be done to achieve wider acceptance among producers. Simultaneously using “all” available process information, its processing and converging to meaningful actions is a complex task requiring specialized knowledge. However, advanced control schemes can be implemented by integrating accessible process information at varying levels of complexity. Based on process understanding, advanced bioprocess control can be successfully applied by integrating expert systems and artificial intelligence. The organized use of schematic information and logical process descriptions gained from experience is the key to success. Understanding the relationship between CQA (process output) and CPP (measurable process operating variable) is critical for creating effective control schemes, performing real-time monitoring, and troubleshooting.

Recently, a PAT-based approach has encouraged the biopharmaceutical industry to shift its way of working from “quality by inspection” to “quality by design” (QbD). However, the industry has not yet seen widespread adoption mainly due to the complex regulatory environment and issues faced in implementing the technology. Furthermore, for bioprocessing, a high level of process understanding and control is required. Recently, spectral PAT tools have gained in importance because of their ability to measure multiple process variables in real-time in a non-invasive manner. Additionally, spectroscopic tools can be used to screen cell culture media as they help identify correlations between CPP and CQA. The resulting data can be processed by multivariate data analysis (MVDA).

Part 3 Automation Of Biological Production

The beginning of bioproduction automation is a critical step towards robust process control. Digitization is the new mantra, with online sensors providing continuous data on numerous variables. The development of in-line and near-line sensors, wireless technology for sensor-to-server connection, smart sensors for data acquisition, and advances in sensor calibration and compact technology are increasingly being used to address space and logistics constraints. However, open platform communication or object linking and embedding methods can be implemented to integrate unit operations. A key challenge that remains is the integration of automated and isolated workstations into a continuous workflow without compromising process efficiency. This includes consistency between different versions of software, proprietary interfaces, and various data formats. Therefore, the first requirement for successful integration is the use of standardized communication protocols and graphical user interfaces. A key criterion for user acceptance of middleware is scalability and flexibility. Integrating the Enterprise Control System (ECS) into the Business System (BS), Manufacturing Execution System (MES) and Shop Floor Control System (SFC) is the next phase of the main challenges to be addressed when implementing a plant-wide information control system. The epitome of manufacturing intelligence addressed by industry-focused R&D activities, progress toward production integration through centralized/distributed hardware/software automation architectures is growing. For example, a Supervisory Control and Data Acquisition (SCADA) platform is implemented for an end-to-end integrated process, where the system integrates and analyzes different unit operations, and collects and stores data for monitoring and control. Additionally, the study demonstrates the digital twin concept for enabling control for a small end-to-end monoclonal antibody production platform. However, the full potential of automation has yet to be realized in industrial-scale end-to-end bioproduction processes.

Another big challenge revolves around data analysis. Bioprocess analysis is considered a key technical barrier to the acceptance of automation in bioproduction. Research laboratories are generating more and more data, both in volume and complexity, requiring further analysis. Additionally, regulators focus on the issue of data integrity. Therefore, data management requires operational and industry-standard software systems. In addition, biological production systems must be designed such that they can handle highly complex data with sufficient agility and flexibility.

Part 4 Summary

Over the past two decades, enormous improvements in process productivity have been achieved. Advances in process control and monitoring help reduce development costs and increase affordability. The production of products such as mAbs poses a significant challenge to traditional manufacturing practices given the associated complexities and inconsistencies. Inspired by recent regulatory guidelines within the QbD framework, stochastic and mechanistic model-based controllers are now becoming popular choices for bioprocess control. In contrast to traditional experimental approaches, it can be observed that the use of simulations and advanced statistics reduces costs and time. This has driven the development of process control strategies, from cascade to adaptive to hybrid, and the introduction of neural network (NN) based controllers. Recent work has conceptualized controllers based on holistic process models to provide end-to-end digital “replicas” of biological processes. It involves the integration of single unit operations with monitoring and control, leading to a deeper understanding of the impact of complex associations between CPP and CQA on coupled unit operations. Currently, model-based controllers can contribute to root cause analysis, molecular interaction, and refinement of unit operation models. However, they usually have a high computational burden. Neural network-based control strategies have made significant progress in recent years, but their use in process modeling requires large datasets. Therefore, comprehensive approaches that combine statistical models with detailed theoretical models are needed to avoid exhaustive experiments and gain a deeper understanding. Control strategies need to be designed to achieve high levels of precision, accuracy and robustness.

Furthermore, the automation of production will provide substantial opportunities to overcome many of the challenges to commercialization success. However, significant technical and business strategy challenges are encountered when creating scalable and automated bioproduction solutions. Biotherapeutics developers must therefore think in terms of large-scale production and incorporate automation concepts from the earliest stages of process development. Insufficient standardization of software, hardware, and design specifications complicates attempts at automation. To overcome these barriers, manufacturers must continue to improve process understanding and use this understanding to develop a simplified and efficient bioproduction process that includes validated process testing and controls. Last but not least, stakeholders and technology solution providers should try to bridge innovation gaps in bioproduction and jointly design and develop automation solutions with biotherapeutic developers.