Ehecatl Antonio del Rio Chanona
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Ehecatl Antonio del Rio Chanona has a total of 191 co-authors. Here are the top co-authors by number of shared publications:
- Dongda Zhang
Department of Chemical Engineering The University of Manchester Manchester UK
46 shared publications - Klaus Hellgardt
Department of Chemical Engineering, Imperial College London, London SW7 2AZ, UK
11 shared publications Klaus Hellgardt
Department of Chemical Engineering, Imperial College London, London, UK
10 shared publications- Keju Jing
Department of Chemical and Biochemical Engineering, College of Chemistry and Chemical Engineering Xiamen University Xiamen China
8 shared publications Nilay Shah
Sargent Centre for Process Systems Engineering Roderic Hill Building South Kensington Campus London London UK
8 shared publications- King Kuok (Mimi) Hii
Department of Chemistry, Imperial College London, Molecular Sciences Research Hub, 82 Wood Lane, London W12 0BZ, UK
7 shared publications
Latest Publications
- Open Access
- 13 November 2026
- Published by Wiley in AIChE Journal
A human‐in‐the‐loop Bayesian optimization framework for constraint‐aware bioprocess development
We present Pareto Front Guided Sampling (PFGS), a human‐in‐the‐loop (HiL) Bayesian Optimization (BO) framework that reformulates surrogate‐model predictions as a multi‐objective optimization problem, allowing domain knowledge to be added intuitively to any BO problem. PFGS addresses constrained optimization by treating the posterior probability of constraint satisfaction as an explicit Pareto objective, and robust optimization via a Monte Carlo estimate of expected lower‐confidence performance under input perturbations, capturing likely degradation. The multi‐dimensional Pareto front makes trade‐offs between predicted performance, model uncertainty, constraint satisfaction, and robustness visible through pairwise two‐dimensional projections on an interactive dashboard, allowing selection criteria to be refined as the model and objectives evolve. The framework is showcased on a eight‐dimensional in‐silico fed‐batch Chinese Hamster Ovary cell culture case study. We showed improved performance over BO when the expert has perfect domain knowledge; however, PFGS is not designed to outperform BO, but rather to present a HiL acquisition function framework.
- 13 November 2026
- Published by Wiley in AIChE Journal
Open AccessA human‐in‐the‐loop Bayesian optimization framework for constraint‐aware bioprocess development
We present Pareto Front Guided Sampling (PFGS), a human‐in‐the‐loop (HiL) Bayesian Optimization (BO) framework that reformulates surrogate‐model predictions as a multi‐objective optimization problem, allowing domain knowledge to be added intuitively to any BO problem. PFGS addresses constrained optimization by treating the posterior probability of constraint satisfaction as an explicit Pareto objective, and robust optimization via a Monte Carlo estimate of expected lower‐confidence performance under input perturbations, capturing likely degradation. The multi‐dimensional Pareto front makes trade‐offs between predicted performance, model uncertainty, constraint satisfaction, and robustness visible through pairwise two‐dimensional projections on an interactive dashboard, allowing selection criteria to be refined as the model and objectives evolve. The framework is showcased on a eight‐dimensional in‐silico fed‐batch Chinese Hamster Ovary cell culture case study. We showed improved performance over BO when the expert has perfect domain knowledge; however, PFGS is not designed to outperform BO, but rather to present a HiL acquisition function framework.
- Open Access
- 27 October 2026
- Published by Elsevier in SSRN Electronic Journal
Generating Sequential Process Operation Plans using Large Language Models: A Verifier-Guided Approach
Sequential process operations—startup, shutdown, supervisory mode changes, and fault recovery—are governed by discrete operating logic: only certain actions are admissible from a given state, while disturbances may change the state independently of the controller. Large language models (LLMs) are attractive decision-support components because they can generate candidate action sequences from task descriptions and contextual information, but their outputs cannot be assumed executable. This paper proposes a verifier-guided framework in which finite-state machines (FSMs) provide executable operating logic, an LLM generates candidate plans, and a deterministic verifier checks proposed sequences before acceptance. The aim is not to replace classical graph search when the complete planning problem is formalized, but to study the reliability of language-model-generated operational sequences when a symbolic model is available for validation. Failed proposals are diagnosed and converted into feedback for repair. In particular, cut-based repair stores structural state-action exclusions, while timed failures retain their execution context in verifier feedback. The framework is evaluated on a disturbance-augmented FSM benchmark and a timed startup-style operational case study, comparing no feedback, naive verifier feedback, cut-based repair, and verifier-informed Reflexion-style feedback. GPT-4o and Qwen3-8B achieve untimed solve rates of 96.2% and 94.9%, respectively. Cut-based repair increases solve rate from 37.6% to 69.6% for LLaMA3-8B and from 26.0% to 55.6% for Mistral-7B, while the timed study shows that state- and disturbance-dependent failures require context-aware repair. The results support a hybrid decision architecture in which LLMs provide flexible candidate generation while executable symbolic verification determines admissibility and converts failures into reusable repair information.
- 27 October 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessGenerating Sequential Process Operation Plans using Large Language Models: A Verifier-Guided Approach
Sequential process operations—startup, shutdown, supervisory mode changes, and fault recovery—are governed by discrete operating logic: only certain actions are admissible from a given state, while disturbances may change the state independently of the controller. Large language models (LLMs) are attractive decision-support components because they can generate candidate action sequences from task descriptions and contextual information, but their outputs cannot be assumed executable. This paper proposes a verifier-guided framework in which finite-state machines (FSMs) provide executable operating logic, an LLM generates candidate plans, and a deterministic verifier checks proposed sequences before acceptance. The aim is not to replace classical graph search when the complete planning problem is formalized, but to study the reliability of language-model-generated operational sequences when a symbolic model is available for validation. Failed proposals are diagnosed and converted into feedback for repair. In particular, cut-based repair stores structural state-action exclusions, while timed failures retain their execution context in verifier feedback. The framework is evaluated on a disturbance-augmented FSM benchmark and a timed startup-style operational case study, comparing no feedback, naive verifier feedback, cut-based repair, and verifier-informed Reflexion-style feedback. GPT-4o and Qwen3-8B achieve untimed solve rates of 96.2% and 94.9%, respectively. Cut-based repair increases solve rate from 37.6% to 69.6% for LLaMA3-8B and from 26.0% to 55.6% for Mistral-7B, while the timed study shows that state- and disturbance-dependent failures require context-aware repair. The results support a hybrid decision architecture in which LLMs provide flexible candidate generation while executable symbolic verification determines admissibility and converts failures into reusable repair information.
- Open Access
- 17 October 2026
- Published by Elsevier in Computers & Chemical Engineering
- Vol. 216
Physics-informed symbolic regression for data-efficient kinetic model discovery
- 17 October 2026
- Published by Elsevier in Computers & Chemical Engineering
- Vol. 216
Open AccessPhysics-informed symbolic regression for data-efficient kinetic model discovery
- Open Access
- 27 July 2026
- Published by Elsevier in SSRN Electronic Journal
Physics Informed Symbolic Regression for Data-Efficient Kinetic Model Discovery
The industrialization of catalytic processes hinges on the availability of reliable kinetic models for design, optimization, and control. Traditional mechanistic models demand extensive domain expertise, while many data-driven approaches often lack interpretability and fail to enforce physical consistency. To overcome these limitations, we propose the Physics-Informed Automated Discovery of Kinetics (PI-ADoK) framework. By integrating physical constraints directly into a symbolic regression approach, PI-ADoK narrows the search space and substantially reduces the number of experiments required for model convergence. Additionally, the framework incorporates a robust uncertainty quantification strategy via the Metropolis-Hastings algorithm, which propagates parameter uncertainty to yield credible prediction intervals. Benchmarking our method against conventional approaches across several catalytic case studies demonstrates that PI-ADoK not only enhances model fidelity but also lowers the experimental burden, highlighting its potential for efficient and reliable kinetic model discovery in chemical reaction engineering.
- 27 July 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessPhysics Informed Symbolic Regression for Data-Efficient Kinetic Model Discovery
The industrialization of catalytic processes hinges on the availability of reliable kinetic models for design, optimization, and control. Traditional mechanistic models demand extensive domain expertise, while many data-driven approaches often lack interpretability and fail to enforce physical consistency. To overcome these limitations, we propose the Physics-Informed Automated Discovery of Kinetics (PI-ADoK) framework. By integrating physical constraints directly into a symbolic regression approach, PI-ADoK narrows the search space and substantially reduces the number of experiments required for model convergence. Additionally, the framework incorporates a robust uncertainty quantification strategy via the Metropolis-Hastings algorithm, which propagates parameter uncertainty to yield credible prediction intervals. Benchmarking our method against conventional approaches across several catalytic case studies demonstrates that PI-ADoK not only enhances model fidelity but also lowers the experimental burden, highlighting its potential for efficient and reliable kinetic model discovery in chemical reaction engineering.
- 19 June 2026
- Published by The CACHE Foundation, Inc.
- Vol. 6, 2008-2014
Optimization of Large-Scale Lycopene Production from Tomato Waste: A Comparative Study of Different Processing Technologies
For process simulation, Python and Aspen Plus can be combined to leverage the advantages of both. This work utilizes the integration of Python and Aspen Plus for the design and optimization of a lycopene production process from tomato waste. Three production pathways are studied: acetone and hexane as solvents, enzymes with ethyl acetate, and supercritical CO2 with ethanol. This allows for the scaling of laboratory-scale studies into industrial-scale analyses. Genetic algorithms are used for optimization, enabling the determination of the optimal process design, costs, and operating conditions, while minimizing the total annual cost. The process with acetone and hexane yields a final production of 0.21 kg/h, the process with enzymes and ethyl acetate, 5.13 kg/h, and the process with supercritical CO2 and ethanol, 0.13 kg/h. It is shown that the process with ethyl acetate has a higher production and the process with supercritical CO2 and ethanol results in lower production and higher costs.
- 19 June 2026
- Published by The CACHE Foundation, Inc.
- Vol. 6, 2008-2014
Optimization of Large-Scale Lycopene Production from Tomato Waste: A Comparative Study of Different Processing Technologies
For process simulation, Python and Aspen Plus can be combined to leverage the advantages of both. This work utilizes the integration of Python and Aspen Plus for the design and optimization of a lycopene production process from tomato waste. Three production pathways are studied: acetone and hexane as solvents, enzymes with ethyl acetate, and supercritical CO2 with ethanol. This allows for the scaling of laboratory-scale studies into industrial-scale analyses. Genetic algorithms are used for optimization, enabling the determination of the optimal process design, costs, and operating conditions, while minimizing the total annual cost. The process with acetone and hexane yields a final production of 0.21 kg/h, the process with enzymes and ethyl acetate, 5.13 kg/h, and the process with supercritical CO2 and ethanol, 0.13 kg/h. It is shown that the process with ethyl acetate has a higher production and the process with supercritical CO2 and ethanol results in lower production and higher costs.
- 19 June 2026
- Published by The CACHE Foundation, Inc.
- Vol. 6, 1558-1564
Control-Guided Reinforcement Learning for Cooperative Energy Management
Addressing the urgent transition to low-carbon energy systems requires microgrids capable of locally coordinating electricity generation, storage, and flexible consumption. Their efficient integration calls for control strategies that are scalable, privacy-preserving, and robust to uncertainty. To address such a challenging control problem, this work proposes a decentralised Multi-Agent Reinforcement Learning (MARL) approach based on the Cross-Entropy Method (CEM) for the coordination of prosumers, equipped with renewable generation and vehicle-to-grid capabilities. To improve sample efficiency and robustness, the policy is warm-started using Behaviour Cloning (BC) from a classical Proportional-Integral-Derivative (PID) controller, resulting in a hybrid BC–CEM framework. The proposed method is evaluated in a realistic microgrid simulation with stochastic demand and real weather and generation profiles. Results show that BC–CEM accelerates convergence and achieves lower energy costs compared to both PID control and randomly initialized CEM, without sacrificing comfort or mobility requirements. The findings highlight the effectiveness of combining derivative-free optimization with imitation learning in complex MARL tasks, such as energy flexibility coordination.
- 19 June 2026
- Published by The CACHE Foundation, Inc.
- Vol. 6, 1558-1564
Control-Guided Reinforcement Learning for Cooperative Energy Management
Addressing the urgent transition to low-carbon energy systems requires microgrids capable of locally coordinating electricity generation, storage, and flexible consumption. Their efficient integration calls for control strategies that are scalable, privacy-preserving, and robust to uncertainty. To address such a challenging control problem, this work proposes a decentralised Multi-Agent Reinforcement Learning (MARL) approach based on the Cross-Entropy Method (CEM) for the coordination of prosumers, equipped with renewable generation and vehicle-to-grid capabilities. To improve sample efficiency and robustness, the policy is warm-started using Behaviour Cloning (BC) from a classical Proportional-Integral-Derivative (PID) controller, resulting in a hybrid BC–CEM framework. The proposed method is evaluated in a realistic microgrid simulation with stochastic demand and real weather and generation profiles. Results show that BC–CEM accelerates convergence and achieves lower energy costs compared to both PID control and randomly initialized CEM, without sacrificing comfort or mobility requirements. The findings highlight the effectiveness of combining derivative-free optimization with imitation learning in complex MARL tasks, such as energy flexibility coordination.
- 19 June 2026
- Published by The CACHE Foundation, Inc.
- Vol. 6, 692-701
Pareto Front Guided Sampling for Efficient Bioprocess Experimentation
This work presents Pareto Front Guided Sampling (PFGS), a model-guided Design of Experiments (DoE) strategy for bioprocess development that makes the exploration–exploitation trade-off explicit and integrates human expertise into experiment selection. Starting from an initial experimental design, PFGS fits a probabilistic surrogate and then proposes new experiments by solving a multi-objective design problem that simultaneously rewards (i) high predicted performance (posterior mean) and (ii) high information gain (posterior uncertainty). Rather than collapsing this trade-off into a single acquisition value, PFGS generates a Pareto set of candidate experiments, that reflect different balances between improvement-seeking and learning. To prevent wasted runs, an automated screening step is performed to remove candidates in (i) low predicted-mean regions unlikely to yield near-optimal performance and (ii) low-uncertainty regions already well explained by the surrogate, concentrating effort on promising yet under-explored areas and encouraging coverage of multiple near-optimal basins. For industrial application, PFGS was extended to batch selection by choosing a diverse subset of Pareto candidates to support parallel experimentation without substantial loss in sample efficiency, and a termination rule to stop experimentation once user-defined accuracy thresholds are met. PFGS is benchmarked against Latin Hypercube Sampling (LHS) and Bayesian Optimisation (BO) on two representative problems: (i) a bioprocess case study with shallow performance gradients, where naive exploitation is unreliable and non-adaptive designs are inefficient, and (ii) a multi-modal landscape, where methods that focus on a single basin can miss alternative optima. Across these settings, PFGS allocates experimental resources more effectively than non-adaptive baselines and, in the multi-modal case, uniquely identifies and characterises all stationary regions while maintaining strong overall performance. By combining Pareto-transparent decision support, automated filtering, batch practicality, and principled stopping, PFGS provides an interpretable and resource-efficient DoE methodology for biopharmaceutical process optimisation.
- 19 June 2026
- Published by The CACHE Foundation, Inc.
- Vol. 6, 692-701
Pareto Front Guided Sampling for Efficient Bioprocess Experimentation
This work presents Pareto Front Guided Sampling (PFGS), a model-guided Design of Experiments (DoE) strategy for bioprocess development that makes the exploration–exploitation trade-off explicit and integrates human expertise into experiment selection. Starting from an initial experimental design, PFGS fits a probabilistic surrogate and then proposes new experiments by solving a multi-objective design problem that simultaneously rewards (i) high predicted performance (posterior mean) and (ii) high information gain (posterior uncertainty). Rather than collapsing this trade-off into a single acquisition value, PFGS generates a Pareto set of candidate experiments, that reflect different balances between improvement-seeking and learning. To prevent wasted runs, an automated screening step is performed to remove candidates in (i) low predicted-mean regions unlikely to yield near-optimal performance and (ii) low-uncertainty regions already well explained by the surrogate, concentrating effort on promising yet under-explored areas and encouraging coverage of multiple near-optimal basins. For industrial application, PFGS was extended to batch selection by choosing a diverse subset of Pareto candidates to support parallel experimentation without substantial loss in sample efficiency, and a termination rule to stop experimentation once user-defined accuracy thresholds are met. PFGS is benchmarked against Latin Hypercube Sampling (LHS) and Bayesian Optimisation (BO) on two representative problems: (i) a bioprocess case study with shallow performance gradients, where naive exploitation is unreliable and non-adaptive designs are inefficient, and (ii) a multi-modal landscape, where methods that focus on a single basin can miss alternative optima. Across these settings, PFGS allocates experimental resources more effectively than non-adaptive baselines and, in the multi-modal case, uniquely identifies and characterises all stationary regions while maintaining strong overall performance. By combining Pareto-transparent decision support, automated filtering, batch practicality, and principled stopping, PFGS provides an interpretable and resource-efficient DoE methodology for biopharmaceutical process optimisation.
- 19 June 2026
- Published by The CACHE Foundation, Inc.
- Vol. 6, 1551-1557
Optimisation of Synthetic Natural Gas Production via Direct Air Capture and Utilisation using Reduced Models under a Novel Trust-Region Funnel Method
In this study, we propose a novel trust-region funnel (TRF) optimisation framework for process systems that integrate external black-box models, such as rigorous models, within equation-oriented (EO) formulations. The framework is applied to optimise a synthetic natural gas production process combining direct air capture and catalytic CO2 conversion using dual-function material (DFM) technology, with the objective of minimising the total annualised cost. The problem is formulated in Pyomo and solved using IPOPT, treating the DFM reactor as an external black-box model. The TRF method achieves substantial improvements compared to published mixed-integer nonlinear programming and direct nonlinear programming approaches, reducing capture cost from 460 USD to 426 USD per tonne of CO2. Key design improvements include reducing the number of DFM units per train by one-third and achieving a 22% reduction in DFM capital costs. These results highlight the TRF framework’s ability to overcome numerical challenges and unlock economically superior designs for next-generation carbon capture and utilisation systems.
- 19 June 2026
- Published by The CACHE Foundation, Inc.
- Vol. 6, 1551-1557
Optimisation of Synthetic Natural Gas Production via Direct Air Capture and Utilisation using Reduced Models under a Novel Trust-Region Funnel Method
In this study, we propose a novel trust-region funnel (TRF) optimisation framework for process systems that integrate external black-box models, such as rigorous models, within equation-oriented (EO) formulations. The framework is applied to optimise a synthetic natural gas production process combining direct air capture and catalytic CO2 conversion using dual-function material (DFM) technology, with the objective of minimising the total annualised cost. The problem is formulated in Pyomo and solved using IPOPT, treating the DFM reactor as an external black-box model. The TRF method achieves substantial improvements compared to published mixed-integer nonlinear programming and direct nonlinear programming approaches, reducing capture cost from 460 USD to 426 USD per tonne of CO2. Key design improvements include reducing the number of DFM units per train by one-third and achieving a 22% reduction in DFM capital costs. These results highlight the TRF framework’s ability to overcome numerical challenges and unlock economically superior designs for next-generation carbon capture and utilisation systems.
- Open Access
- 4 May 2026
- Published by Elsevier in SSRN Electronic Journal
Physics Informed Symbolic Regression for Data-Efficient Kinetic Model Discovery
The industrialization of catalytic processes hinges on the availability of reliable kinetic models for design, optimization, and control. Traditional mechanistic models demand extensive domain expertise, while many data-driven approaches often lack interpretability and fail to enforce physical consistency. To overcome these limitations, we propose the Physics-Informed Automated Discovery of Kinetics (PI-ADoK) framework. By integrating physical constraints directly into a symbolic regression approach, PI-ADoK narrows the search space and substantially reduces the number of experiments required for model convergence. Additionally, the framework incorporates a robust uncertainty quantification strategy via the Metropolis-Hastings algorithm, which propagates parameter uncertainty to yield credible prediction intervals. Benchmarking our method against conventional approaches across several catalytic case studies demonstrates that PI-ADoK not only enhances model fidelity but also lowers the experimental burden, highlighting its potential for efficient and reliable kinetic model discovery in chemical reaction engineering.
- 4 May 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessPhysics Informed Symbolic Regression for Data-Efficient Kinetic Model Discovery
The industrialization of catalytic processes hinges on the availability of reliable kinetic models for design, optimization, and control. Traditional mechanistic models demand extensive domain expertise, while many data-driven approaches often lack interpretability and fail to enforce physical consistency. To overcome these limitations, we propose the Physics-Informed Automated Discovery of Kinetics (PI-ADoK) framework. By integrating physical constraints directly into a symbolic regression approach, PI-ADoK narrows the search space and substantially reduces the number of experiments required for model convergence. Additionally, the framework incorporates a robust uncertainty quantification strategy via the Metropolis-Hastings algorithm, which propagates parameter uncertainty to yield credible prediction intervals. Benchmarking our method against conventional approaches across several catalytic case studies demonstrates that PI-ADoK not only enhances model fidelity but also lowers the experimental burden, highlighting its potential for efficient and reliable kinetic model discovery in chemical reaction engineering.
- 20 March 2026
- Published by Wiley in Biotechnology and Bioengineering
- Vol. 123 (4)
Front Cover Image, Volume 123, Number 4, April 2026
- 20 March 2026
- Published by Wiley in Biotechnology and Bioengineering
- Vol. 123 (4)
Front Cover Image, Volume 123, Number 4, April 2026