| dc.description.abstract | A chemical plant may have thousands of measurements and control loops. By the term plantwide control it is not meant the tuning and behavior of each of these loops, but rather the control philosophy of the overall plant with emphasis on the structural decisions. The structural decision includes the selection/placement of manipulators and measurements as well as the decomposition of the overall problem into smaller subproblems (the control configuration). Based on a review of the existing methods, a plantwide control design procedure is proposed.
The procedure starts with a top-down analysis of the plant. Where the emphasis is on selecting controlled variables, which will give an easy and more robust optimization. This is achieved by controlling the active constraints, and for the unconstrained degrees of freedom variables with a flat optimum is preferred. A flat optimum indicates that an implementation error or a disturbance will have a small effect on the economic performance. The next step is to choose the throughput manipulator.
The top-down analysis is followed by a bottom-up design of the control system. The bottom-up design is guided by controllability analysis. The goal is first to stabilize the plant (including nearly unstable poles), such that it is possible to operate the plant manually. This is the regulatory control layer. Finally, the supervisory control layer is designed.
One issue that needs to be resolved is if such a control hierarchy can impose new and fundamental limitations, which is not present in the original plant. It is shown, that if the setpoints and measurements of the lower layer are available to the next layer and that the lower layer controller is stable and minimum phase, it is not possible to introduce new fundamental limitations. When the lower layer measurements and/or the lower layer setpoints are unavailable it is possible to introduce new limitations.
The procedure is applied on several applications:
1. A liquid phase reactor with a distillation column and recycle.
2. A gas phase reactor with separator, compressors and recycle.
3. The methanol synthesis loop (a special case of application 2).
4. The Tennessee Eastman challenge problem.
5. An industrial heat integrated distillation columns (from the methanol plant).
All these plants have in common that their behavior is changed by recycle or heat integration.
In the case of a liquid phase reactor, there is no economic penalty for increasing the holdup in the reactor. In fact, the holdup should be as large as possible in order to increase the conversion per pass, which will make separation cheaper. Luyben has proposed a control strategy in which he uses the reactor holdup as a throughput manipulator. This will give an economic penalty which most other authors so far have neglected. For the remaining degree of freedom there is a flat optimum for all but a few variables. One of the conclusion is that the “Luyben rule”, i.e. “fix one flow in recycle” has bad self-optimizing properties and should not be applied to this plant.
For the gas phase plant, the situation is different. Due to compression costs there are a cost associated with the hold-up (pressure). In fact, the optimum is unconstrained in this variable. Control of recycle-rate, purge fraction, or reactor pressure gives a system with good self-optimizing properties. This is linked to the behavior of these variables as conversion increases. As expected purge flow is a bad alternative as a controlled variable. More unexpectedly, inert composition in the recycle turned out to have bad self-optimizing properties. This is also explainable by the behavior of this variable when conversion is increased. The results for the simple gas phase reactor carries well over to the methanol case study.
The Tennessee Eastman problem is a well-studied test problem, but few have studied the selection of controlled variables based on the economics of the plant. In addition to the constrained variables, reactor temperature, C in purge and recycle flow or compressor work, should be controlled. A very common claim is that it is necessary to control the inventory of inert components, this is not true. The shape of the objective function is very unfavorable, and a small implementation error leads to infeasibilities.
The heat-integrated distillation columns are similar too simple distillation in many aspects. But there are differences, e.g. the number of degrees of freedom are different. We argue that the heat transfer area between the two columns, top compositions (of valuable products), and pressure in the low-pressure column should be controlled at their constraints. There is one unconstrained degree of freedom, and for this particular case control of a temperature in the lower part of the column shows good self-optimizing properties.
It is shown that it is not given that poles at the origin will not show up in the relative gain array. It may happen if it is possible to stabilize the pole with two different control loops. This should be seen as an argument for using the frequency dependent relative gain array.
The emphasis in this thesis has been on case studies. By the use of systematic tools for analysis, some “rules” that have been presented in the process control community are shown to have had a weak theoretical basis. The thesis has improved the understanding of the control of a large scale processing plants. | nb_NO |
