{"id":{"repo_id":"stellenbosch","oai_identifier":"oai:scholar.sun.ac.za:10019.1/135922"},"canonical_url":"https://search.dev.ndltd.org/etd/stellenbosch/oai:scholar.sun.ac.za:10019.1/135922","repository":{"repo_id":"stellenbosch","name":"Stellenbosch University","base_url":"https://scholar.sun.ac.za/server/oai/request"},"display":{"title":"Constrained multi-objective Bayesian optimization of simulated moving bed chromatography","abstract":"Simulated moving bed chromatography (SMB) is a continuous counterpart to conventional batch reverse phase chromatography. SMB is a multi-column system and uses counter-current movement of the liquid and solid phases to achieve high-resolution separations. While SMB offers significant performance improvement, identifying optimal operating conditions that satisfy product specifications remains a major challenge - this is due to the presence of many decision variables (e.g. column flowrates, indexing time etc.) and hence, many degrees of freedom. To overcome the challenges of optimization, dynamic simulation is generally employed as a means of empirically estimating optimal operating conditions prior to experimentation, as a better alternative to trial-and-error approaches. However, dynamic SMB models are computationally intensive. In addition, because conventional optimization algorithms require numerous iterations to converge to solutions, the optimization of dynamic SMB models demands considerable execution time. This study proposes Bayesian optimization (BO) as a sample-efficient strategy for determining optimal SMB operating conditions. BO is a stochastic, gradient-free optimization technique and is well-suited for “expensive” black-box functions. BO applies a Gaussian Process (GP) surrogate to approximate objective function and an acquisition function (AQ) to guide sampling under specified inequality constraints. In this work BO was further extended to a multi-objective setting through Pareto Efficient Global Optimization, resulting in a constrained multi-objective Bayesian optimization (MOBO) strategy. Both the SMB model and MOBO routine were developed from first principles, validated, and applied to two case studies: (i) glucose-fructose separation and (ii) borate-hydrochloric acid (HCl) separation. In each case, the objective was to maximize component recoveries while constraining the product purity to a >99.5% threshold. In the glucose-fructose separation, the goal was to identify operating conditions that would result in 95% fructose-purity. MOBO provided operating conditions to achieve a 97.75 % fructose-recovery with fructose-purity of 96.47 %. The 96.47% fructose-purity provided a ~1.50 % safety factor to the 95% product specification while maintaining high fructose recovery. In the borate-HCl separation, MOBO successfully identified operating conditions that achieved 99.89% borate-purity with ~100% borate-recovery. Each case study was completed with a budget of 50 optimization iterations assigned to MOBO. The resulting operating conditions provided actionable recommendations for implementation on Stellenbosch University’s SMB pilot plant. Overall, the proposed framework demonstrates that MOBO can efficiently estimate optimal solutions and uncover trade-offs between competing objectives in SMB separations even within a limited number of optimization iterations.","abstract_html":"Simulated moving bed chromatography (SMB) is a continuous counterpart to conventional batch reverse phase chromatography. SMB is a multi-column system and uses counter-current movement of the liquid and solid phases to achieve high-resolution separations. While SMB offers significant performance improvement, identifying optimal operating conditions that satisfy product specifications remains a major challenge - this is due to the presence of many decision variables (e.g. column flowrates, indexing time etc.) and hence, many degrees of freedom. To overcome the challenges of optimization, dynamic simulation is generally employed as a means of empirically estimating optimal operating conditions prior to experimentation, as a better alternative to trial-and-error approaches. However, dynamic SMB models are computationally intensive. In addition, because conventional optimization algorithms require numerous iterations to converge to solutions, the optimization of dynamic SMB models demands considerable execution time. This study proposes Bayesian optimization (BO) as a sample-efficient strategy for determining optimal SMB operating conditions. BO is a stochastic, gradient-free optimization technique and is well-suited for “expensive” black-box functions. BO applies a Gaussian Process (GP) surrogate to approximate objective function and an acquisition function (AQ) to guide sampling under specified inequality constraints. In this work BO was further extended to a multi-objective setting through Pareto Efficient Global Optimization, resulting in a constrained multi-objective Bayesian optimization (MOBO) strategy. Both the SMB model and MOBO routine were developed from first principles, validated, and applied to two case studies: (i) glucose-fructose separation and (ii) borate-hydrochloric acid (HCl) separation. In each case, the objective was to maximize component recoveries while constraining the product purity to a >99.5% threshold. In the glucose-fructose separation, the goal was to identify operating conditions that would result in 95% fructose-purity. MOBO provided operating conditions to achieve a 97.75 % fructose-recovery with fructose-purity of 96.47 %. The 96.47% fructose-purity provided a ~1.50 % safety factor to the 95% product specification while maintaining high fructose recovery. In the borate-HCl separation, MOBO successfully identified operating conditions that achieved 99.89% borate-purity with ~100% borate-recovery. Each case study was completed with a budget of 50 optimization iterations assigned to MOBO. The resulting operating conditions provided actionable recommendations for implementation on Stellenbosch University’s SMB pilot plant. Overall, the proposed framework demonstrates that MOBO can efficiently estimate optimal solutions and uncover trade-offs between competing objectives in SMB separations even within a limited number of optimization iterations.","abstract_has_math":false,"creators":["Punabantu, Nawa Uriel"],"institution":"Stellenbosch : Stellenbosch University","degree_name":null,"degree_level":null,"degree_discipline":null,"degree_department":null,"school":null,"contributors":[],"advisors":["Louw, Tobias Muller","Pott, Robert William McClelland"],"committee_chairs":[],"committee_members":[],"year":2026,"date_issued":"2026-03","date_published":"2026-03","updated_at":"2026-07-24T04:40:06Z","subjects":[],"languages":["en"],"rights":[],"rights_urls":[],"identifier_entries":[]},"links":{"outbound_url":"https://scholar.sun.ac.za/handle/10019.1/135922","outbound_label":"Repository record","outbound_source":"dc:identifier.uri"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor.advisor","label":"Advisor","values":["Louw, Tobias Muller","Pott, Robert William McClelland"]},{"key":"dc:contributor.other","label":"Dc Contributor Other","values":["Stellenbosch University. Faculty of Engineering. Dept. of Chemical Engineering."]},{"key":"dc:creator","label":"Author","values":["Punabantu, Nawa Uriel"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date.accessioned","label":"Dc Date Accessioned","values":["2026-04-15T09:06:20Z"]},{"key":"dc:date.available","label":"Dc Date Available","values":["2026-04-15T09:06:20Z"]},{"key":"dc:date.issued","label":"Date","values":["2026-03"]},{"key":"dc:publisher","label":"Institution","values":["Stellenbosch : Stellenbosch University"]},{"key":"dc:type","label":"Dc Type","values":["Thesis"]}]},{"id":"language_rights","label":"Language and Rights","entries":[{"key":"dc:language.iso","label":"Language (ISO)","values":["en"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier.uri","label":"Identifier URI","values":["https://scholar.sun.ac.za/handle/10019.1/135922"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description","label":"Description","values":["Thesis (MEng)--Stellenbosch University, 2026.","Punabantu, N. U. 2026. Constrained multi-objective Bayesian optimization of simulated moving bed chromatography. Unpublished masters thesis. Stellenbosch: Stellenbosch University [online]. Available: https://scholar.sun.ac.za/items/2cbb1e0c-6520-4efa-81f3-3f9d01900d7c"]},{"key":"dc:description.abstract","label":"Abstract","values":["Simulated moving bed chromatography (SMB) is a continuous counterpart to conventional batch reverse phase chromatography. SMB is a multi-column system and uses counter-current movement of the liquid and solid phases to achieve high-resolution separations. While SMB offers significant performance improvement, identifying optimal operating conditions that satisfy product specifications remains a major challenge - this is due to the presence of many decision variables (e.g. column flowrates, indexing time etc.) and hence, many degrees of freedom. To overcome the challenges of optimization, dynamic simulation is generally employed as a means of empirically estimating optimal operating conditions prior to experimentation, as a better alternative to trial-and-error approaches. However, dynamic SMB models are computationally intensive. In addition, because conventional optimization algorithms require numerous iterations to converge to solutions, the optimization of dynamic SMB models demands considerable execution time. This study proposes Bayesian optimization (BO) as a sample-efficient strategy for determining optimal SMB operating conditions. BO is a stochastic, gradient-free optimization technique and is well-suited for “expensive” black-box functions. BO applies a Gaussian Process (GP) surrogate to approximate objective function and an acquisition function (AQ) to guide sampling under specified inequality constraints. In this work BO was further extended to a multi-objective setting through Pareto Efficient Global Optimization, resulting in a constrained multi-objective Bayesian optimization (MOBO) strategy. Both the SMB model and MOBO routine were developed from first principles, validated, and applied to two case studies: (i) glucose-fructose separation and (ii) borate-hydrochloric acid (HCl) separation. In each case, the objective was to maximize component recoveries while constraining the product purity to a >99.5% threshold. In the glucose-fructose separation, the goal was to identify operating conditions that would result in 95% fructose-purity. MOBO provided operating conditions to achieve a 97.75 % fructose-recovery with fructose-purity of 96.47 %. The 96.47% fructose-purity provided a ~1.50 % safety factor to the 95% product specification while maintaining high fructose recovery. In the borate-HCl separation, MOBO successfully identified operating conditions that achieved 99.89% borate-purity with ~100% borate-recovery. Each case study was completed with a budget of 50 optimization iterations assigned to MOBO. The resulting operating conditions provided actionable recommendations for implementation on Stellenbosch University’s SMB pilot plant. Overall, the proposed framework demonstrates that MOBO can efficiently estimate optimal solutions and uncover trade-offs between competing objectives in SMB separations even within a limited number of optimization iterations."]},{"key":"dc:title","label":"Title","values":["Constrained multi-objective Bayesian optimization of simulated moving bed chromatography"]}]}],"canonical_facts":{"dc:contributor.advisor":["Louw, Tobias Muller","Pott, Robert William McClelland"],"dc:contributor.other":["Stellenbosch University. Faculty of Engineering. Dept. of Chemical Engineering."],"dc:creator":["Punabantu, Nawa Uriel"],"dc:date.accessioned":["2026-04-15T09:06:20Z"],"dc:date.available":["2026-04-15T09:06:20Z"],"dc:date.issued":["2026-03"],"dc:description":["Thesis (MEng)--Stellenbosch University, 2026.","Punabantu, N. U. 2026. Constrained multi-objective Bayesian optimization of simulated moving bed chromatography. Unpublished masters thesis. Stellenbosch: Stellenbosch University [online]. Available: https://scholar.sun.ac.za/items/2cbb1e0c-6520-4efa-81f3-3f9d01900d7c"],"dc:description.abstract":["Simulated moving bed chromatography (SMB) is a continuous counterpart to conventional batch reverse phase chromatography. SMB is a multi-column system and uses counter-current movement of the liquid and solid phases to achieve high-resolution separations. While SMB offers significant performance improvement, identifying optimal operating conditions that satisfy product specifications remains a major challenge - this is due to the presence of many decision variables (e.g. column flowrates, indexing time etc.) and hence, many degrees of freedom. To overcome the challenges of optimization, dynamic simulation is generally employed as a means of empirically estimating optimal operating conditions prior to experimentation, as a better alternative to trial-and-error approaches. However, dynamic SMB models are computationally intensive. In addition, because conventional optimization algorithms require numerous iterations to converge to solutions, the optimization of dynamic SMB models demands considerable execution time. This study proposes Bayesian optimization (BO) as a sample-efficient strategy for determining optimal SMB operating conditions. BO is a stochastic, gradient-free optimization technique and is well-suited for “expensive” black-box functions. BO applies a Gaussian Process (GP) surrogate to approximate objective function and an acquisition function (AQ) to guide sampling under specified inequality constraints. In this work BO was further extended to a multi-objective setting through Pareto Efficient Global Optimization, resulting in a constrained multi-objective Bayesian optimization (MOBO) strategy. Both the SMB model and MOBO routine were developed from first principles, validated, and applied to two case studies: (i) glucose-fructose separation and (ii) borate-hydrochloric acid (HCl) separation. In each case, the objective was to maximize component recoveries while constraining the product purity to a >99.5% threshold. In the glucose-fructose separation, the goal was to identify operating conditions that would result in 95% fructose-purity. MOBO provided operating conditions to achieve a 97.75 % fructose-recovery with fructose-purity of 96.47 %. The 96.47% fructose-purity provided a ~1.50 % safety factor to the 95% product specification while maintaining high fructose recovery. In the borate-HCl separation, MOBO successfully identified operating conditions that achieved 99.89% borate-purity with ~100% borate-recovery. Each case study was completed with a budget of 50 optimization iterations assigned to MOBO. The resulting operating conditions provided actionable recommendations for implementation on Stellenbosch University’s SMB pilot plant. Overall, the proposed framework demonstrates that MOBO can efficiently estimate optimal solutions and uncover trade-offs between competing objectives in SMB separations even within a limited number of optimization iterations."],"dc:identifier.uri":["https://scholar.sun.ac.za/handle/10019.1/135922"],"dc:language.iso":["en"],"dc:publisher":["Stellenbosch : Stellenbosch University"],"dc:title":["Constrained multi-objective Bayesian optimization of simulated moving bed chromatography"],"dc:type":["Thesis"]},"updated_at":"2026-07-24T04:40:06Z"}