Studies Of Ship Turning Maneuverability Based On Numerical Simulation And Machine Learning Method

Since the world entered the "Industry 4.0" period,ship construction of digitalization,informatization,and intelligence has become a hot research topic.Moreover,the development of intelligent ships will improve sailing safety.The variation of environmental conditions and the ship’s state,such as loads and speeds is very likely to affect the maneuvering capability while sailing on the sea.At this point,not only does the error of ship motion described by the traditional mathematical model inevitably increase,but the parameters in the model drift as well.Given the above problems,the intelligent technology presents a method to dynamically evaluate the navigation state as well as realize the real-time prediction of the ship’s maneuvering performance.In this paper,the ship maneuverability digital evaluation method was studied.Firstly,the maneuverability prediction method based on the MMG model was studied.Taking the international general model KVLCC2 tanker as the research object,the CFD method was applied to conduct the ship model test(static oblique towing test and dynamic circular motion test),capturing the ship hydrodynamic force and torque.The viscous hydrodynamic derivatives of the hull in the 3-DOF MMG equation were obtained by the polynomial least square fitting method.The propeller hydrodynamic and rudder hydrodynamic were obtained by empirical formulas.Subsequently,the fourth-order Runge-Kutta method in FORTRAN language was chosen to solve the motion differential equation,which settles the turning motions in calm water and waves.It indicates that the ship turning trajectory and maneuverability index are in good agreement with the experimental data,showing the effectiveness of the present method.On this basis,the sensitivity of the maneuverability index to the hull viscous hydrodynamic derivatives was studied.The results show that the first-order yaw hydrodynamic derivatives and the resistance coefficient have a significant effect on the turning motion.The influence of the low-order derivatives is generally stronger than that of the high-order derivatives.Secondly,the CFD method was also used to conduct the self-propulsion test.The propeller effect was carried out with the adoption of the virtual disk volume force method.The rudder and hull grid interface was established to realize the steering motion.By allowing the ship to travel straight ahead,the self-propulsion point was obtained,in the meantime,the vorticity field around the stern was analyzed.Then,the self-propelled steady turning motion was conducted.After comparing the trajectory to the modle test,the predictions reveal that the maneuverability index relative error is satisfied,and the simulation accuracy of the right rudder is higher than the left rudder.The vorticity field and pressure field around the rudder,as well as velocity flow field,were also visualized.Finally,maneuverability prediction was studied with the machine learning method.On the one hand,while treating the 3-DOF MMG equation as part of the loss function,the neural network model based on numerical simulation data was built,which applied a multilayer feedforward neural network.With great prediction accuracy,generalization research was further investigated.It shows that the increase of the testing dataset and the variable amount is beneficial to improving training accuracy.On the other hand,in view of piles of figures generated by the numerical simulation method,the black-box identification modelling was set up to predict the ship maneuverability in calm water and waves,with the utilization of long and short-term memory neural networks.The prediction results agree well with the original data,and the accuracy is improved while including the rudder angle as an additional input.Furthermore,the sea voyage data of the 400-thousand-ton bulk carrier VLOC was utilized as the training dataset.The results reveal that the ship maneuvering motion can be well predicted by using the history data,thus providing a preliminary reference for the dynamic prediction of ship maneuvering performance in the real sea.