Constrained Multi-Objective Bayesian Optimization of Simulated Moving Bed Chromatography

Simulated Moving Bed (SMB) chromatography is a continuous separation process widely adopted in pharmaceutical and sugar industries for high-resolution purification of binary and even ternary mixtures. Different SMB configurations must be investigated to optimise product grade and recovery but modelling such systems involves solving complex systems of partial differential equations, making optimization computationally intensive. The performance of the Constrained Multi-Objective Bayesian Optimization (CMOBO) algorithm was evaluated with limits on computational resources. A sparse Pareto front was generated using 20 iterations which closely followed the trend established by a more exhaustive 100-iteration optimisation run – preserving predominant trade-offs and insight into potential operating modes. The 100-iterations required ~74 CPU hours, while the 20-iterations required ~9 CPU hours on the institutional computing cluster. This comparison highlights the CMOBO’s strong sampling ability to maximize SMB performance parameters, even under a limited number of allowable function evaluations, enabling rapid prototyping of SMB configurations.