In the last three decades, process automation systems have evolved into a modular and networked architecture, due to the increasing demand from the industry for reliable and flexible plant operation. The last three decades also brought along powerful tools for the solution of large-scale, multivariable optimization and control problems. At this point, an important task for process systems engineers is to combine these tools on the algorithm side, with the networked automation systems on the infrastructure side. Achieving this task requires three important questions to be addressed.

•How should the actual plant and the corresponding models be decomposed into subsystems?

•How should the tasks of the same nature i.e. multivariable control with multivariable control, interact (homogenous networks)?

•How should the tasks of a different nature i.e. process monitoring with real time optimization, interact (heterogeneous networks)?

Answering the first question involves the exploration of possible decompositions, additional requirements such as observability and controllability, and also a balance between computational efficiency due to parallelization and communication load on the network due to decentralization. The second question involves the development of complex, multi-rate algorithms for distributed optimization, estimation and control. Finally, the third question requires the study of coordination and arbitration procedures.

In the present work, a distributed model predictive control framework is demonstrated as a homogeneous network and the connection of these controllers with a real time optimization module is studied as a heterogeneous system. Model decomposition problem is discussed for a highly interacting, multivariable process. Finally, the performance of a networked automation system for this process is tested according to several simulation scenarios.