Design of a Multivariable Self-tuning Control Algorithm for Distributed Systems

Abstract

Optimal control strategies are very dependent on the accuracy of the model and this assumes some importance in operating plant when there is a change in characteristics. Consequently, in order to ensure good and stable performance a detailed description of the process is required. Since it is not common to incorporate tunable properties in the model to compensate for any process changes it is difficult to realize the benefit of the sophisticated strategies typical of the successes in the aerospace industry. Nevertheless, some significant improvements are possible. An example of the kind of problems which arise in respect of distributed parameter processes, is the fixed-bed catalytic reactor. Detailed models are not useful for control because it results in loss of simplicity and excessive computing requirements. Moreover, it is generally difficult and costly to identify such models because of the problems associated with the measurement of some of the relevant variables e.g. rate constants and/or catalyst activity. To be attractive, overall controller design must be based on a simple but robust control algorithm which allows for an approximate form of model. Consideration is given to the use of an adaptive closed-loop control law based on a multi-input-multi-output system having random process inputs and the observations subjected to noise. A comparison is made with the performance of a multivariable optimal regulator using a quadratic performance index. The effects of model inaccuracies, sampling times and other design parameters are also considered. The need tor this basic structure is Shown CO be important in the case of the tabular reactor to allow for the approximation of multidimensional lumped parameter representation of the process.