Towards a digital twin: Digitization and model-based optimization of the innovative high-gradient magnetic separator

Downstream processing in biotechnology relies on multiple unit operations to achieve high product purity, driving up costs, time, and yield losses. High-Gradient Magnetic Separation (HGMS) offers a promising alternative by consolidating steps and enabling direct target capture from complex media. However, its industrial adoption is hindered by suboptimal performance, limited scalability, and insufficient automation for reproducibility. Furthermore, process efficiency is often not fully realized due to the reliance on fixed operational recipes. This study presents a digital twin framework for a pilot-scale HGMS system, integrating real-time monitoring, automated control, advanced mechanistic models, and multi-objective optimization using Bayesian algorithms. The framework was validated for robustness, scalable data handling, and predictive control. Key contributions include the development of soft sensors, automated control strategies for improved reproducibility, in-silico optimization of a human Immunoglobulin G (hIgG) capture process — a monoclonal antibody broadly relevant in biopharmaceutical applications — with real-time pH adjustment, and a sensitivity analysis of objective weights, revealing trade-offs between yield, resource use, and processing time. Optimization results indicated a theoretical 4% productivity gain and a 3 percentage point yield improvement, while exposing critical design constraints in the HGMS chamber. These findings underscore the potential of digital twins to accelerate process optimization and reduce development costs through in-silico experimentation. Future work will focus on refining identified design limitations and extending the framework to optimize process conditions for diverse bioproducts, enhancing scalability and efficiency. • Innovative pilot-scale magnetic separation for downstream bioprocessing. • Digital twin with real-time control enhances HGMS process optimization. • Bayesian multi-objective optimization improves yield and productivity. • Real-time pH adaptation ensures robust and reproducible process performance. • In-silico modeling identifies chamber constraints and reduces development costs.