Date of Award

1995

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Chemical Engineering

First Advisor

Armando B. Corripio

Abstract

New control strategies are based on the model of the process and it is thus necessary to identify the systems to be controlled. It is also often necessary to identify them during closed-loop operation in order to maintain efficient operation and product quality. Some results of multivariable closed-loop identification carried out on a simulated 2 x 2 linear time-invariant system, using two new versions of instrumental variable methods called IV4D and IV4UP as the identification methods, are presented. In each case pseudorandom binary signals (PRBS), or dithers, are applied to the outputs of the feedback controllers. Algorithms IV4D and IV4UP are created in a four step environment where iterations are performed to obtain the best possible estimated model. For IV4D only the dither is used as part of the instrument. For IV4UP only the part of the input that comes from the dither is used for the instrument. This is obtained with the estimated model and with the description of the controllers using the closed-loop transfer function between the dither and the input to the process. The implementation is made to be run in MatLab and it uses several of the functions defined in its System Identification Toolbox (Ljung, 1991). Both instrumental variable (IV) algorithms perform very well identifying closed-loop multivariable systems under the influence of white noise and correlated noise disturbances. The two new instrumental variable methods are compared with the prediction error method, PEM, and with IV4, the regular instrumental variable open-loop algorithm, both of them are obtained from the MatLab System Identification Toolbox. IV4 does not perform well in closed-loop operation. From the simulated results, the performances of the new IV algorithms are the best but, PEM's performance is very close. Finally, real plant data are analyzed with IV4D and its results are compared with the results of other identification methods, PEM and Dynamic Matrix Identification (DMI) (Cutler and Yocum, 1991). For this closed-loop real plant data PEM is the best that performs followed by IV4D, while DMI does not perform well.

Pages

204

Share

COinS