Robust Identification Scheme in an Adaptive Track-Controller for Ships

Abstract

We describe the method of parameter estimation used in an adaptive track controller for ships, which we developed for commercial production. The choice of the algorithm was dominated by two main requirements: the algorithm had to be robust against temporarily poor excitation of one of the model parts and its complexity was limited by the moderate computational capacity in our specific hardware environment. The system model consists of a part describing the low frequency ship motion and a high frequency disturbance model. The identification problem is in general state space form with some states not directly measurable. The standard linear model form can be recovered, if the states are assumed to be known. As an approximation they can be reconstructed with an observer. Instead of using the current parameter estimates for designing a suboptimal observer, we use certain fixed average parameter values. This results in a very simple structure of the estimation algorithm with stability properties that are more easily to investigate in practise. In particular the obtained separation of the two model parts allows to taylor their respective parameter tracking algorithms according to the different patterns of excitation.