A Multiscale Framework for Predicting Reactive Distillation Dynamics

Thesis Abstract

Reactive distillation combines separation and reaction in a single unit, offering potential energy savings, reduced equipment footprints, and intensified process performance; however, its predictive reliability remains limited due to the multiscale interactions among reaction kinetics, phase equilibria, mass transfer, and column hydraulics. This study addresses the challenge by developing a multiscale framework to predict reactive distillation dynamics with coherent integration across molecular, unit, and process scales, enabling accurate forecasting of temperature profiles, conversion, selectivity, and product purity under transient and steady-state operation. The aim is to formulate and validate a multiscale framework that couples molecular-level kinetic parameters and thermodynamic models with rigorous unit- and process-scale simulations to predict reactive distillation performance. Specific objectives are (i) to quantify how microscopic reaction mechanisms and activity coefficients influence macroscopic column behavior under varying feed compositions and light-ends constraints; (ii) to develop a modular upscaling approach that links detailed reaction-thermodynamic models with rigorous dynamic distillation simulators via a transfer-learning interpolation scheme; (iii) to implement a sensitivity and uncertainty analysis to identify dominant scales and parameters affecting predictive accuracy; (iv) to evaluate the framework on benchmark reactive systems (e.g., methanol-to-olefins surrogate, esterification with in-situ dehydration) using experimentally validated kinetic data; and (v) to provide design guidelines and operational strategies for improved controllability and energy efficiency. A mixed-methods methodology is adopted. The population consists of three representative reactive distillation configurations a simple reactive distillation with equilibrium-limited reactions, a heterogeneous catalytic reactive distillation, and a reactive azeotropic system. A sample of five to seven available kinetic-thermodynamic datasets will be compiled from literature and in-house experiments to calibrate molecular-scale models. At the molecular level, density functional theory (DFT) and transition state theory (TST) calculations will estimate reaction pathways and activation energies, while activity coefficients will be retrieved via NRTL or UNIFAC models, with parameter estimation performed through nonlinear least-squares fitting against reported vapor-liquid equilibrium data. At the unit/packaged model level, a dynamic stage-by-stage distillation model will be implemented in a validated process simulator, augmented by a reduced-order sub-model for phase-equilibria and heat and mass transfer resistances. At the process level, a time-dependent, closed-loop control framework will be built to study startup, load changes, and feed disturbances. Data collection instruments include high-resolution thermocouples and pressure transducers within experimental test rigs for model validation; online gas chromatography-midilli and high-performance liquid chromatography (HPLC) analyses for product distribution; fiducial experiments to capture dynamic responses under step disturbances; and literature-compiled kinetic and thermodynamic data for calibration. The method of analysis comprises hierarchical Bayesian calibration to fuse molecular-scale parameters with process-scale data, maximum likelihood estimation for model parameters, and global sensitivity analysis using Sobol indices. Model validation will be performed with time-series data employing root mean square error (RMSE), Theil’s inequality coefficient, and concordance correlation coefficient (CCC). Hypothesis testing will assess whether the multiscale framework significantly reduces predictive error relative to conventional single-scale models. Expected findings indicate that incorporating scale-coupled kinetics and thermodynamics reduces dynamic prediction errors by 25–40% for transient distillation profiles and improves steady-state purity predictions by 1–2 mole percent for complex feedstocks. The framework is anticipated to reveal critical scale interactions, notably the influence of activity coefficient variations on stage holdup and temperature trajectories, and to identify regimes where reduced-order models retain high fidelity. The study will contribute novel methodological insights into multiscale model integration for reactive separations and provide a transferable framework applicable to diverse reactive distillation configurations. The contribution to knowledge includes a validated multiscale framework that seamlessly integrates molecular-level reaction-thermodynamic phenomena with macro-scale distillation dynamics, offering systematic pathways for design optimization and robust control. Conclusions will emphasize the practical implications for energy-efficient reactor-separator configurations and improved operability under perturbations, with recommendations for extending the framework to pilot-plant validation, integration with real-time optimization, and incorporation of catalytic deactivation and fouling phenomena.