Mitigating control loop interactions: Disturbance observers versus decouplers

This paper examines the feasibility of using disturbance observers (DOs) to compensate for interactions between control loops in non-linear multivariable systems. A distillation column application is considered. Traditionally, both steady-state and dynamic decouplers, designed using a linear model of the nonlinear system, are used to improve the multi-loop control of distillation columns. The multi-loop control strategy used here is proportional and integral (PI) feedback control on each loop with the controller parameters being derived from the linear model of the column. Steady-state and dynamic decoupling work well on linear systems but less so on nonlinear systems when there is a degree of mismatch between the linear model used to design the decouplers and the underlying non-linear system. Previous work on DOs has indicated their ability to compensate for system nonlinearities, implying that the use of DO's for decoupling in nonlinear multivariable systems might provide an improvement over the traditional decoupling strategies. Control studies on the non-linear distillation column model indicate that the dynamic decoupler is extremely sensitive to model mismatch, leading to severe deterioration in performance when applied to the non-linear system. The steady-state decoupler was less sensitive to model mismatch and indeed, tends to be the approach used practically. DOs, on the other hand, alleviated the effects of loop interaction quite well even in the presence of modeling errors. This suggests that DOs will work well in a non-linear environment in the presence of model mismatch, perhaps providing much better robust performance than that displayed by the traditional decoupling approaches.