Modelling the yaw dynamics of an uninhabited surface vehicle for navigation and control systems design

Abstract

Marine and Industrial Dynamic Analysis Research Group, Advanced Engineering Systems and Interactions, School of Marine Science and Engineering, Plymouth University, UK In many navigation and control systems a key element in both their software architectures is a model which describes the dynamic characteristics of the given vessel. Thus, to ensure uninhabited surface vehicles are capable of successfully completing missions, it is essential that the marine control systems designer has available bona fide dynamic models that are accurate and possess good prediction capabilities. Since hydrodynamic modelling is usually very expensive, time-consuming and requires the use of specialist equipment in the form of a tank testing facility, it was considered more appropriate to model the vehicle yaw dynamics using Black-box identification techniques. In this paper five new models of the yaw dynamics of the Springer uninhabited surface vehicle are developed and presented. Two of the models that are based on auto-regressive with exogenous variable and autoregressive moving average with exogenous variable techniques are linear in composition and expressed in their state space formats; while the others are nonlinear and described by multi-layer perception, radial basis function and recursive neural networks. In all cases, the models were optimised using a steady-state genetic algorithm. The models are compared using a Mean Square Error criterion and residual correlation analysis. From the evaluation undertaken, it is shown that the yaw dynamic model based on the recursive neural network outperformed the others in terms of predictability and exactness. Therefore, it is concluded the recursive neural network model is the best candidate for navigation and control systems design studies involving the Springer vehicle. © 2014 2012 Taylor and Francis Group LLC.