Optimization-based design and model-based control for coupled distillation systems

Abstract:
In chemical and petrochemical industry, distillation has been established as reference unit operation for the separation of liquid mixtures. The capability of high product purity, robustness in operation, and good predictability of the vapor liquid behavior favors its application. Further integration of distillation with reaction, extraction, or membrane separation allows for an additional improvement of the processes. Due to the application of distillation in the production of bulk chemicals, the optimal design and operation has been in the focus of industrial and academic research. In the last decades, sensitivity studies in process simulators have become the state of the art approach in conceptual process design. However, sensitivity studies are limited in the degrees of freedom that can be investigated. Especially the utilization of intensified processes, like reactive distillation, significantly increases the number of design and operational degrees of freedom and therefore requires more powerful design techniques. Systematic design methods, which incorporate shortcut methods and rigorous optimization, are a promising design alternative. They are combined in the Process Synthesis Framework (Marquardt, 2008): The distillation column configurations are optimized for total annualized cost of operation (Kraemer, 2009), incorporating operation as well as annualized investment cost. Continuous reformulation and efficient initialization strategies allow for a robust optimization of the mixed-integer column model. In this contribution, cost optimal flowsheets are determined for multi-component azeotropic separation and reactive distillation. The integrated design of separation processes is not only a challenging task in process syntheses but also in process control, as the complex design may result in more complex system dynamics such as strong nonlinearities, nonminimum-phase behavior, stiffness, or strong couplings of the process considered. As a result, state-of-the-art control technology, namely decentralized PID control, may not sufficiently satisfy control performance. Furthermore, standard system identification methods like step tests may be inappropriate to derive a system model reproducing the behavior of the real plant. Hence, in addition to the process synthesis approach, this contribution presents a systematic modeling approach, which can be used to design model-based control methods, such as full state observers and model-predictive controllers.