Copyright ® IFAC Control Applications in Marine Systems, Fukuoka, Japan, 1998 A SENSITIVITY APPROACH TO IDENTIFICATION OF SHIP DYNAMICS FROM SEA TRIAL DATA Mogens Blanke and Morten Knudsen Department of Control Engineering, Aalborg University,Fredrik Bajersvej 'lC, DK-9220 Aalborg, Denmark E-mail:{blanke.mk}@control.auc.dk Abstract: Non-linear mathematical models of ships comprise one hundred param- eters or more, and differences between full-scale trials and model tests are difficult to associate with the individual terms. Direct identification of parameters would be advantageous. The paper employs a sensitivity approach in an attempt to achieve this. Using the method on full-scale data from a containership, a good fit in roll and yaw is obtained, but the method reveals that this does not imply good determination of individual parameters. The sensitivity method is found to be easily applied for both identification and evaluation of the reliability of parameter estimates. Copyright @ 1998 IFAC Keywords: marine systems, modeling, system identification, non-linear ship models 1 INTRODUCTION Parameters in non-linear mathematical models of ship dynamics are used for simulation of ship mo- tions, and for the design of closed loop control systems. Model structure and values of param- eters can be estimated in model basin tests, but full-scale verifications are required anyway. They are employed to improve model to full scale pre- dictions and investigate dynamic phenomena with new hull forms and control devices. Despite two decades of experience with system identification, it is still a fairly difficult task to determine param- eters in the non-linear manoeuvering equations. The reasons are that the non-linear models have extremely high complexity and excitation is fairly limited in full scale. The paper adopts a sensitiv- ity approach to ease this identification task. The mainstream of system identification has tra- ditionally been dealing with discrete time , linear models. These are of limited interest in the mar- itime field since hydrodynamic phenomena give rise to significant non-linearities in the manoeu- 241 vering regime. Furthermore , assessment of model- basin results require determination of physical pa- rameters . With a known structure of the differen- tial equations for ship motion, the requirement is to identify parameters for non-linear continuous- time models with known structure. Methods for parameter identification in continuous-time systems have been studied earlier and results are also available for non-linear systems. In the marine systems case, four main problems need to be solved, however. One is the complexity of mathematical models and over-parameterization in the equations of motion . This renders some parameters heavily correlated in certain experiments. The second is the limited excitation possibilities in full scale. The conse- quence is that we need to combine different sets of experiments to determine a model. The third is parameter errors due to under-modelling because some dynamic effects are disregarded. The fourth obstacle is parameter bias from wave and wind disturbances.