시스템식별법과 신경회로망을 이용한 동유체력 모델 정립

Abstract

In a few decades ago, there have been many researches on establishment of mathematical model of hydrodynamic force and moment acting on a ship by using system identification method and free running model test. But most of researches have been focused on the estimation of hydrodynamic coefficients not the model structure. The coefficients in the conventional polynomial models could not be estimated well because of, so called, simultaneous drift phenomenon. In this paper, we suggest the neural network model trained by using data resulted from free running model test. After setting up the equations of motion of four degrees of freedom such as surge, sway, roll, and yaw, unmeasured motion variables and hydrodynamic force and moment are estimated by applying the extended Kalman filter and fixed interval smoother. And then estimated motion variables and measured rudder angles are used for inputs of neural network model and estimated hydrodynamic force and moment for outputs. In conclusion, while we could estimate unmeasured motion variables and hydrodynamic force successfully, we have failed the confirmation of the neural network model to describe hydrodynamic force and moment because there has been still remained the simultaneous drift phenomenon and strong sensitivity in roll moment. Difficulties in establishing the model by neural networks would be thrown out here.