## Plantwide- and Self-Optimizing Control of a Reactor with Recycle Process

##### Master thesis

##### Permanent lenke

http://hdl.handle.net/11250/248534##### Utgivelsesdato

2013##### Metadata

Vis full innførsel##### Samlinger

##### Sammendrag

Plantwide control is a very important topic in today?s process plants in a lot of different industries. With a large number of interacting process units, it is not enough to design a control structure for each single unit by itself without considering the whole system. Control decisions highly affect the economics of a plant, and that is why it is such an important topic. For example, a huge part of the energy costs can be saved by selecting a proper control system.The objective of this work was to find a control structure for an entire process plant, including which variables to control, which variables to be manipulated and which variable to measure. Skogestad?s procedure was applied to a reactor and distillation column process with recycle. This procedure included finding an economic cost function, finding the active constraints, defining the degrees of freedom, and identify self-optimizing variables based on economic loss.Two different cases were studied in this thesis. Case I had a given reactor feed rate with the objective of minimizing the vapour boilup in the column, while case II had the objective of maximizing the feed rate with a given vapour boilup.Matlab models of the reactor and distillation column were made, and the process was optimized for both of the studied cases. Following this, the nominal optimal values of compositions and flows were compared between the cases and also compared with existing literature. The optimization results were consistent with the literature written on the same process. Three different methods were tested in order to find which self-optimizing variable to use for the last remaining degree of freedom in the system. These were the Brute force method, the null space method and the exact local method. In the Brute force method, the economic cost was computed for all the candidate controlled variables by keeping them constant and applying disturbances and implementation errors. The losses resulting from this were plotted to see which variable had the smallest loss. This ended up being the flow ratio L/F, and was therefore chosen as the self-optimizing variable for both case I and II.Measurement combinations to be held constant were found by applying the null space and exact local method, followed by calculating the economic loss caused by this. Dynamic simulations were also performed in order to find the loss with the null space and exact local method. The dynamic model gave the same nominal optimization results as the steady-state model.Control structures and pairings between controlled variables and manipulated variables were suggested for the two cases based on the results from Sigurd?s procedure. For control of the self-optimizing variable L/F, the reflux flow L was used as manipulated variable for the single control loop. This was because the reflux L was the only remaining unconstrained degree of freedom after the control analysis.