Robust Decentralized Control for Flexible Plant Operation: Conceptual Design and Optimal Tuning

In this work we describe a conceptual approach which has been recently developed to design efficient decentralized control structures for flexible chemical plant operation. The approach combines concepts from thermodynamics, inventory networks and process control to construct a set of stable and robust decentralized control candidate structures. Such candidates consist of inventory control loops designed so to ensure convergence of the mass and energy inventories to given references and thus guaranteeing stability of the extensive properties despite plant disturbances or even parameter uncertainty. Since the loops are defined over inventories, they contain almost every possible decentralized alternative. In this way, and by physically realizing the inventory control loops over the available degrees of freedom, the approach can be employed to construct a superstructure only containing stable (and robust) decentralized control candidates. It is on this set that NLP or MINLP problems can be defined to select the best operational alternative. The different aspects of the conceptual design approach will be illustrated on the well known Tennessee-Eastman Process. In particular the benchmark case will be employed to illustrate the definition of the stable inventory alternatives, physical realization and optimal selection and tuning by an efficient and recently developed MINLP solver.