Adaptive decoupling of multivariable systems

Abstract
For many industrial processes, it is of interest to design a decoupling precompensator. The precompensator makes it possible to design controllers based on single-input-single-output models of the process. A model of the process must be known to design the precompensator. This paper shows how the precompensator can be designed adaptively using input-output measurements. The precompensator is first designed for the case when the process is known. The adaptive precompensator is then constructed using the certainty equivalence principle. The convergence properties and the implementation of the adaptive decoupler are discussed. It is shown that the adaptive scheme will converge provided the system is persistently excited and that a suitable model structure is used in the estimation.