{"id":{"repo_id":"tamu","oai_identifier":"oai:oaktrust.library.tamu.edu:1969.1/1599339"},"canonical_url":"https://search.dev.ndltd.org/etd/tamu/oai:oaktrust.library.tamu.edu:1969.1/1599339","repository":{"repo_id":"tamu","name":"Texas A&M University","base_url":"https://oaktrust.library.tamu.edu/server/oai/request"},"display":{"title":"Advances in Hybrid Modeling: Parameter Estimation and Control for Chemical and Biochemical Processes","abstract":"Hybrid modeling has emerged as a promising paradigm to combine the interpretability of first-principles models with the adaptability of data-driven techniques. This dissertation develops a unified framework for hybrid modeling, estimation, and control in complex chemical and biochemical processes. At its core, this work addresses a fundamental challenge: the inability of traditional mechanistic models to capture latent, time-varying phenomena, and the tendency of purely data-driven models to lack robustness and physical fidelity. By embedding neural networks within mechanistic models to estimate hidden parameters or transport coefficients, we demonstrate enhanced predictive accuracy, numerical stability, and real-time control capability across a diverse set of applications. The first part of this work focuses on a hybrid model for aerobic fermentation systems, where temporal variation in oxygen transfer and microbial activity is significant. By learning latent dynamics that cannot be explicitly modeled from first principles, the hybrid model significantly improves predictive performance compared to standalone mechanistic or data-driven models. This approach serves as a foundation for the development of digital twins and dynamic optimization in industrial bioreactors. The second component extends hybrid modeling to systems governed by reaction-diffusion dynamics, represented by partial differential equations. Here, the spatiotemporally varying diffusivity is modeled as a function of space and time, allowing the model to adapt to local variations and nonlinear effects of cell density. Since hybrid training in such contexts can suffer from numerical instability and solution multiplicity, the next phase introduces regularization strategies to improve robustness. Building upon these modeling foundations, the next component of this dissertation develops a hybrid model-based observer and model predictive control (MPC) strategy. The observer reconstructs unmeasured states using partial measurements and parameters estimated by the neural network, while the MPC uses the hybrid model to optimize process trajectories under uncertainty. A user-friendly graphical interface is developed to support deployment and tuning, emphasizing the framework's readiness for industrial integration. The final chapter extends the hybrid modeling framework to two-timescale systems, with a focus on crystallization processes. Recognizing the stiffness and computational burden of complex mechanistic models, we develop a hybrid two-timescale model that treats fast dynamics (nucleation, growth) using method-of-moments equations and models slow, stiff aggregation terms using neural networks. This decoupled approach enables stable simulations, rapid control optimization, and precise regulation of crystal size distributions. Together, these contributions establish a generalizable methodology for constructing, training, and deploying hybrid models across diverse process systems. By bridging mechanistic insight with the flexibility of data-driven models, the framework offers a scalable path toward intelligent monitoring and control of real-world chemical and biochemical operations.","abstract_html":"Hybrid modeling has emerged as a promising paradigm to combine the interpretability of first-principles models with the adaptability of data-driven techniques. This dissertation develops a unified framework for hybrid modeling, estimation, and control in complex chemical and biochemical processes. At its core, this work addresses a fundamental challenge: the inability of traditional mechanistic models to capture latent, time-varying phenomena, and the tendency of purely data-driven models to lack robustness and physical fidelity. By embedding neural networks within mechanistic models to estimate hidden parameters or transport coefficients, we demonstrate enhanced predictive accuracy, numerical stability, and real-time control capability across a diverse set of applications. The first part of this work focuses on a hybrid model for aerobic fermentation systems, where temporal variation in oxygen transfer and microbial activity is significant. By learning latent dynamics that cannot be explicitly modeled from first principles, the hybrid model significantly improves predictive performance compared to standalone mechanistic or data-driven models. This approach serves as a foundation for the development of digital twins and dynamic optimization in industrial bioreactors. The second component extends hybrid modeling to systems governed by reaction-diffusion dynamics, represented by partial differential equations. Here, the spatiotemporally varying diffusivity is modeled as a function of space and time, allowing the model to adapt to local variations and nonlinear effects of cell density. Since hybrid training in such contexts can suffer from numerical instability and solution multiplicity, the next phase introduces regularization strategies to improve robustness. Building upon these modeling foundations, the next component of this dissertation develops a hybrid model-based observer and model predictive control (MPC) strategy. The observer reconstructs unmeasured states using partial measurements and parameters estimated by the neural network, while the MPC uses the hybrid model to optimize process trajectories under uncertainty. A user-friendly graphical interface is developed to support deployment and tuning, emphasizing the framework's readiness for industrial integration. The final chapter extends the hybrid modeling framework to two-timescale systems, with a focus on crystallization processes. Recognizing the stiffness and computational burden of complex mechanistic models, we develop a hybrid two-timescale model that treats fast dynamics (nucleation, growth) using method-of-moments equations and models slow, stiff aggregation terms using neural networks. This decoupled approach enables stable simulations, rapid control optimization, and precise regulation of crystal size distributions. Together, these contributions establish a generalizable methodology for constructing, training, and deploying hybrid models across diverse process systems. By bridging mechanistic insight with the flexibility of data-driven models, the framework offers a scalable path toward intelligent monitoring and control of real-world chemical and biochemical operations.","abstract_has_math":false,"creators":["Shah, Parth Jitendra 1997-"],"institution":"Texas A&M University","degree_name":"Doctor of Philosophy","degree_level":null,"degree_discipline":"Chemical Engineering","degree_department":null,"school":null,"contributors":[],"advisors":["Kwon, Joseph"],"committee_chairs":[],"committee_members":["Eduardo Gildin (egildin@tamu.edu)","Costas Kravaris (kravaris@tamu.edu)","Faisal Khan (fikhan@tamu.edu)"],"year":2025,"date_issued":"2025-08","date_published":"2025-08","updated_at":"2026-08-21T16:48:44Z","subjects":["Artificial Intelligence","Engineering, Chemical","Energy","Chemistry, Pharmaceutical"],"languages":["English"],"rights":[],"rights_urls":[],"identifier_entries":[]},"links":{"outbound_url":"https://hdl.handle.net/1969.1/1599339","outbound_label":"Handle","outbound_source":"dc:identifier.uri"},"source_record":{"url":"https://oaktrust.library.tamu.edu/server/oai/request?verb=GetRecord&metadataPrefix=dim&identifier=oai%3Aoaktrust.library.tamu.edu%3A1969.1%2F1599339","prefix":"dim"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor.advisor","label":"Advisor","values":["Kwon, Joseph"]},{"key":"dc:contributor.committeemember","label":"Committee Member","values":["Eduardo Gildin (egildin@tamu.edu)","Costas Kravaris (kravaris@tamu.edu)","Faisal Khan (fikhan@tamu.edu)"]},{"key":"dc:creator","label":"Author","values":["Shah, Parth Jitendra 1997-"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date.accessioned","label":"Dc Date Accessioned","values":["2026-02-04T22:58:59Z"]},{"key":"dc:date.issued","label":"Date","values":["2025-08"]},{"key":"dc:type","label":"Dc Type","values":["Thesis"]},{"key":"thesis:degree_discipline","label":"Discipline","values":["Chemical Engineering"]},{"key":"thesis:degree_name","label":"Degree Name","values":["Doctor of Philosophy"]},{"key":"thesis:institution_name","label":"Thesis Institution Name","values":["Texas A&M University"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["Artificial Intelligence","Engineering, Chemical","Energy","Chemistry, Pharmaceutical"]}]},{"id":"language_rights","label":"Language and Rights","entries":[{"key":"dc:language.iso","label":"Language (ISO)","values":["English"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier.uri","label":"Identifier URI","values":["https://hdl.handle.net/1969.1/1599339"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description.abstract","label":"Abstract","values":["Hybrid modeling has emerged as a promising paradigm to combine the interpretability of first-principles models with the adaptability of data-driven techniques. This dissertation develops a unified framework for hybrid modeling, estimation, and control in complex chemical and biochemical processes. At its core, this work addresses a fundamental challenge: the inability of traditional mechanistic models to capture latent, time-varying phenomena, and the tendency of purely data-driven models to lack robustness and physical fidelity. By embedding neural networks within mechanistic models to estimate hidden parameters or transport coefficients, we demonstrate enhanced predictive accuracy, numerical stability, and real-time control capability across a diverse set of applications. The first part of this work focuses on a hybrid model for aerobic fermentation systems, where temporal variation in oxygen transfer and microbial activity is significant. By learning latent dynamics that cannot be explicitly modeled from first principles, the hybrid model significantly improves predictive performance compared to standalone mechanistic or data-driven models. This approach serves as a foundation for the development of digital twins and dynamic optimization in industrial bioreactors. The second component extends hybrid modeling to systems governed by reaction-diffusion dynamics, represented by partial differential equations. Here, the spatiotemporally varying diffusivity is modeled as a function of space and time, allowing the model to adapt to local variations and nonlinear effects of cell density. Since hybrid training in such contexts can suffer from numerical instability and solution multiplicity, the next phase introduces regularization strategies to improve robustness. Building upon these modeling foundations, the next component of this dissertation develops a hybrid model-based observer and model predictive control (MPC) strategy. The observer reconstructs unmeasured states using partial measurements and parameters estimated by the neural network, while the MPC uses the hybrid model to optimize process trajectories under uncertainty. A user-friendly graphical interface is developed to support deployment and tuning, emphasizing the framework's readiness for industrial integration. The final chapter extends the hybrid modeling framework to two-timescale systems, with a focus on crystallization processes. Recognizing the stiffness and computational burden of complex mechanistic models, we develop a hybrid two-timescale model that treats fast dynamics (nucleation, growth) using method-of-moments equations and models slow, stiff aggregation terms using neural networks. This decoupled approach enables stable simulations, rapid control optimization, and precise regulation of crystal size distributions. Together, these contributions establish a generalizable methodology for constructing, training, and deploying hybrid models across diverse process systems. By bridging mechanistic insight with the flexibility of data-driven models, the framework offers a scalable path toward intelligent monitoring and control of real-world chemical and biochemical operations."]},{"key":"dc:format.mimetype","label":"Dc Format Mimetype","values":["application/pdf"]},{"key":"dc:title","label":"Title","values":["Advances in Hybrid Modeling: Parameter Estimation and Control for Chemical and Biochemical Processes"]}]}],"canonical_facts":{"dc:contributor.advisor":["Kwon, Joseph"],"dc:contributor.committeemember":["Eduardo Gildin (egildin@tamu.edu)","Costas Kravaris (kravaris@tamu.edu)","Faisal Khan (fikhan@tamu.edu)"],"dc:creator":["Shah, Parth Jitendra 1997-"],"dc:date.accessioned":["2026-02-04T22:58:59Z"],"dc:date.issued":["2025-08"],"dc:description.abstract":["Hybrid modeling has emerged as a promising paradigm to combine the interpretability of first-principles models with the adaptability of data-driven techniques. This dissertation develops a unified framework for hybrid modeling, estimation, and control in complex chemical and biochemical processes. At its core, this work addresses a fundamental challenge: the inability of traditional mechanistic models to capture latent, time-varying phenomena, and the tendency of purely data-driven models to lack robustness and physical fidelity. By embedding neural networks within mechanistic models to estimate hidden parameters or transport coefficients, we demonstrate enhanced predictive accuracy, numerical stability, and real-time control capability across a diverse set of applications. The first part of this work focuses on a hybrid model for aerobic fermentation systems, where temporal variation in oxygen transfer and microbial activity is significant. By learning latent dynamics that cannot be explicitly modeled from first principles, the hybrid model significantly improves predictive performance compared to standalone mechanistic or data-driven models. This approach serves as a foundation for the development of digital twins and dynamic optimization in industrial bioreactors. The second component extends hybrid modeling to systems governed by reaction-diffusion dynamics, represented by partial differential equations. Here, the spatiotemporally varying diffusivity is modeled as a function of space and time, allowing the model to adapt to local variations and nonlinear effects of cell density. Since hybrid training in such contexts can suffer from numerical instability and solution multiplicity, the next phase introduces regularization strategies to improve robustness. Building upon these modeling foundations, the next component of this dissertation develops a hybrid model-based observer and model predictive control (MPC) strategy. The observer reconstructs unmeasured states using partial measurements and parameters estimated by the neural network, while the MPC uses the hybrid model to optimize process trajectories under uncertainty. A user-friendly graphical interface is developed to support deployment and tuning, emphasizing the framework's readiness for industrial integration. The final chapter extends the hybrid modeling framework to two-timescale systems, with a focus on crystallization processes. Recognizing the stiffness and computational burden of complex mechanistic models, we develop a hybrid two-timescale model that treats fast dynamics (nucleation, growth) using method-of-moments equations and models slow, stiff aggregation terms using neural networks. This decoupled approach enables stable simulations, rapid control optimization, and precise regulation of crystal size distributions. Together, these contributions establish a generalizable methodology for constructing, training, and deploying hybrid models across diverse process systems. By bridging mechanistic insight with the flexibility of data-driven models, the framework offers a scalable path toward intelligent monitoring and control of real-world chemical and biochemical operations."],"dc:format.mimetype":["application/pdf"],"dc:identifier.uri":["https://hdl.handle.net/1969.1/1599339"],"dc:language.iso":["English"],"dc:subject":["Artificial Intelligence","Engineering, Chemical","Energy","Chemistry, Pharmaceutical"],"dc:title":["Advances in Hybrid Modeling: Parameter Estimation and Control for Chemical and Biochemical Processes"],"dc:type":["Thesis"],"thesis:degree_discipline":["Chemical Engineering"],"thesis:degree_name":["Doctor of Philosophy"],"thesis:institution_name":["Texas A&M University"]},"updated_at":"2026-08-21T16:48:44Z"}