Intelligent design of distillation columns integrating detailed tray geometry using data-driven model

Distillation is one of the most widely utilized unit operations in the process industry, associated with substantial capital and operational costs. Although systematic design methods exist, conventional approaches still rely on engineers' expertise, leading to inconsistent designs and inefficiencies, especially for repetitive distillation tasks. This research aims to address these challenges by proposing a smart design and optimization framework that leverages machine-learning techniques to reduce reliance on human intervention, thereby enhancing design quality, ensuring design consistency, and improving work and design efficiency. The key novelty lies developing automated framework for simultaneous optimization of column internals and tray efficiency prediction, eliminating the assumption-prediction gap that characterizes traditional sequential approaches, while enabling comprehensive design space exploration through data-driven models significantly reduce computational burden and make multi-variable optimization practically feasible. Moreover, with help of the data-driven models focusing on the optimization of column internals, particularly valve-tray columns, the proposed methodology tackles the complexity of multiple degrees of freedom and stringent hydraulic constraints through an integrated hybrid approach combining machine learning with detailed first-principles hydraulic correlations. By integrating distillation simulation results through data-driven models, rigorous hydraulic correlations, and detailed tray efficiency predictions, the framework ensures operational feasibility through comprehensive hydraulic constraint validation, including jet flooding, downcomer flooding, and weir loading limits. Application of this approach to an industrial valve-tray column for separating C4 hydrocarbons demonstrates good performance improvements. The optimized design achieves a 17% reduction in Total Annualized Cost (TAC) compared to an industrial base case designed following conventional approaches across various investment scenarios while maintaining reasonable hydraulic performance and enhancing tray efficiency through systematic optimization of column internal design parameters, demonstrating the practical advantages of the automated framework over traditional experience-dependent methods.