A Framework for Real-Time Catalytic Process Optimization via Thermodynamic Modeling

Thesis Abstract

The acceleration of catalytic process development hinges on real-time optimization capable of adapting to dynamic reaction environments, yet existing frameworks inadequately integrate thermodynamic constraints with rapid sensor-derived feedback to guide control decisions. This study addresses the gap by proposing a framework that couples real-time thermodynamic modeling with adaptive process optimization to enhance yield, selectivity, and energy efficiency in industrial catalytic reactors. The aim is to develop and validate a thermodynamics-informed optimization framework that operates in real time, leveraging in situ sensor data to continuously refine reactor operation strategies. The specific objectives are (1) to formulate a dynamic thermodynamic model that incorporates heat, mass transfer, catalytic kinetics, and reactor design features; (2) to implement a data assimilation scheme that updates model states using real-time measurements from online gas chromatography, infrared thermography, and pressure-enthalpy sensors; (3) to develop a model-predictive control (MPC) algorithm constrained by thermodynamic feasibility to optimize conversion and selectivity; (4) to evaluate the framework against conventional dashboard-based optimization using historical plant data; and (5) to quantify robustness under feedstock variability and catalyst deactivation scenarios. The methodology adopts a mixed-methods research design combining quantitative modeling with simulation-based validation and a case study in a pilot-scale fixed-bed reactor system. The population comprises industrially relevant catalytic processes, with a representative dataset drawn from a petrochemical refinery pilot plant featuring a 2–3 m bore fixed-bed reactor working with a syngas-to-olefins pathway. A sample of 12 months of continuous operation data, including hourly measurements of temperature, pressure, flow rates, gas compositions, and catalyst activity indicators, will be used for model parameterization and validation. Data collection instruments include inline FTIR and GC analysis for species concentrations, quartz crystal microbalance for deposition trends, thermocouples for temperature profiles, and infrared thermography for surface heat flux monitoring. The framework integrates a canonical thermodynamic model based on Gibbs free energy minimization and reaction equilibrium constraints with kinetic submodels calibrated to reported intrinsic rate data, and a heat and mass transfer module capturing axial and radial gradients. Data assimilation will employ a sequential Bayesian updating technique to fuse sensor data with state estimates, ensuring robust handling of measurement noise and model mismatch. The optimization component utilizes a model-predictive control (MPC) strategy with a horizon aligned to reactor dynamics, formulated to minimize a composite objective function that balances conversion, selectivity, energy consumption, and thermal safety margins. Sensitivity analyses will assess the influence of feedstock variability and catalyst aging on performance, while stochastic simulations will explore robustness under perturbations. Key expected findings include improved reactor-wide selectivity and yield by 8–15% relative to benchmark operations, a reduction in energy intensity by 6–12%, and enhanced process stability under feed variability due to real-time avoidance of thermodynamically infeasible operating regions. The study anticipates demonstrating that the thermodynamics-constrained MPC outperforms traditional PI-based control in maintaining optimal operating envelopes, particularly during catalyst deactivation events where reaction equilibria shift. The contribution to knowledge lies in the explicit integration of real-time thermodynamic feasibility into an adaptive optimization framework for catalytic processes, expanding the theoretical discourse on data-driven process control with fundamental thermodynamic constraints and providing a transferable methodology for diverse catalytic systems. The study will advance practical understanding of how thermodynamic bounds and dynamic kinetics interact under real-time control, offering a replicable blueprint for implementing thermodynamics-informed optimization in industrial settings. The main conclusion is that a thermodynamics-informed real-time optimization framework can significantly enhance performance and resilience of catalytic processes without compromising safety or feasibility, provided that accurate state estimation and robust data assimilation are employed. Recommendations include extending the framework to multi-reactor configurations, incorporating catalyst regeneration dynamics, and validating the approach across different reactor geometries and reaction chemistries. Future work should explore integration with digital twin architectures and the incorporation of machine learning surrogates to reduce computational overhead while preserving thermodynamic integrity.