Model-based control has been demonstrated as an effective strategy for high performance process control. A process model can be used as the basis of the development of alternative techniques for control, and then allow the control designer to analyse the process in a systematic way. Most applications have been devoted to individual operating units of a process plant. However, any modern process plant is made up of multiple processing units, which are commonly coupled and interactive due to the presence of recycle streams. The progression made in moving from a single unit processing plant to a multi-unit processing plant considerably complicates the control system design.
The application of single-unit controllers to a multi-unit processing plant might result in poor overall closed-loop performance if the interactions present were not considered in the controller design. The performance degradation would vary depending upon the strength of those interactions. Current strategies for controlling a multi-unit processing plant rely on treating the whole plant as a large single unit. A more systematic framework for multi-unit control analysis and design has not been studied.
Motivated by these facts, control of multi-unit processing plants has been addressed in this thesis. The aims of this study were to gain a better understanding of multi-unit control design, and to develop a systematic approach to multi-unit control analysis and design through the use of a process model.
A three step procedure has been proposed as a general framework to analyse and design a multi-unit control system.
The first step was to decompose a multi-unit process plant into subsystems of smaller dimension. This approach was motivated by the high dimensionality and complexity of multi-unit processes. Additionally, this approach produced a decentralised control system for the whole plant, which is popular in the chemical process industries.
The second step was to analyse the achievable closed-loop performance when the controller was designed for this plant decomposition.
The third step was to design the controller for each subsystem such that the overall performance objectives were satisfied.
To implement the proposed procedure, two approaches to plant decomposition have been proposed. The first technique was based on the physical unit operations whereby the controlled and manipulated variables were associated with each physical unit. By using this technique, the effect of recycle streams on the overall performance could be analysed such that the performance of single-unit controllers applied to a multiunit processing plants could be investigated.
The second technique was based on the dynamics of the variables to be controlled, with no regard to the physical units in which these variables occurred. Hence, alternative plant decompositions might be produced for a multi-unit processing plant from which the best plant decomposition could be determined.
An approach to interaction analysis has been developed. In this approach, the interactions were represented as normalised coprime factor uncertainty. As a result, two indicators, namely the maximum stability margin bmax and the gap β, have been proposed for screening alternative plant decompositions. The best plant decomposition could be determined by comparing the indicators resulting from alternative plant decompositions.
Furthermore, the stability and achievable performance of decentralised control could be predicted from these indicators. A sufficient stability condition of decentralised control systems has been derived, however this condition might be conservative.
Finally, a controller was independently designed in each subsystem for the best plant decomposition. The H∞ control technique was used to design each individual controller. Hence, following the proposed procedure would lead to the best control strategy for a multi-unit processing plant.
Some examples and two large case studies have been presented : a Supercritical Fluid Extraction (SFE) process and a Reactor-Separation process. These studies highlighted the following points :
1. The strength of steady-state coupling and the process dynamics were factors that made the effect of interactions on the overall closed-loop performance significant. Therefore, both of these factors should be considered in plant decomposition.
2. The main issue in multi-unit control design was not the controller complexity in each subsystem. Increased controller complexity did not guarantee better performance. Rather, the manner of decomposing the plant might improve the overall performance significantly.
3. Plant decomposition based on the physical unit operations did not always produce better performance. This implied that the use of single-unit controllers for controlling a multi-unit processing plant would not always give good overall performance.
4. A realistic performance specification was also an important factor in the success of decentralised controllers for multi-unit processing plant. The indicators, β and bmax could be used as a guideline for posing a realistic performance specification for a given plant decomposition.
5. The use of indicator β could be further exploited for the integration of process design and process control.
In conclusion, a systematic approach to multi-unit control analysis and design has been the main contribution of this study. Additionally, some new insights into multi-unit control design have been gained.