Integrated dual adaptive control of continuous chromatographic separation processes via reinforcement learning.

Continuous chromatographic processes involve nonlinear dynamics, cyclic operation, and strong interactions between operating variables, which make real-time control inherently complex. As process variability and integration increase, there is a growing need for adaptive control frameworks that can maintain performance under changing conditions. In this work, a reinforcement learning (RL) control framework is developed for adaptive operation of continuous chromatographic systems. The controller is trained based on a mechanistic simulation environment and learns to coordinate multiple process inputs based on observed system states. A phase-dependent reward formulation captures the trade-offs between product purity, yield, and operational efficiency, guiding the agent toward stable, high-performance operation from process start-up. The RL controller maintains the target purity under feed variations, flowrate disturbances, and measurement noise, achieves cyclic steady state (CSS) within 10 process cycles, improves process yield from 86 % to 91 %, and increases product recovery by 79 % compared to nominal operation. These results demonstrate that RL can deliver robust and adaptive control for complex multivariable chromatographic systems, providing a framework for intelligent and adaptive process operation.