Machine-Learning-unterstützte Bewegungsberechnung des Kranschiffs Orion
Dr.-Ing. Florian Stempinski, Benjamin Baert M. Sc., DEME Offshore, Zwijndrecht/Belgium; Wolfgang Pauli PhD, Microsoft, Redmond/WA/USA; Maryam Tavakoli Hosseinabadi PhD, Microsoft, Redmond/WA/USA
Machine-Learning-Aided Ship Motion Calculation of the Crane Vessel Orion
The crane vessel Orion is equipped with a DP3 system, a motion-compensated pile gripper, and a main crane with an SWL of 5000t. She is designed to install mono piles, jackets, and substations for future windfarms as well as for installation and decommissioning of oil platforms. In her first project, Arcadis Ost 1, Orion has installed 27 monopiles weighing more than 2000 t with a diameter of 9.5m and a length of up to 110 m.
For safe and efficient operations, the ship motions caused by wind and sea states must be determined for a large variety of operating conditions to fully utilize the operational limits without exceeding them. Hence, a precise prediction of the ship motion response for natural seaway is imperative.
The motion analysis and subsequent dynamic simulations are conducted with Ansys AQWA that solves the radiation and diffraction problem in frequency domain. Yet, the boundary element method (BEM) is time consuming since a high resolution of the meshes of environmental conditions (e.g., wave frequency and direction) is required. The hydrodynamic coefficients that constitute added mass, damping, and wave forcing must be recalculated for each loading/ballasting condition, forward speed, and water depth.
To speed up the calculation time while optimizing accuracy, we implemented an ML-solution. We trained multiple regression models to predict pre-calculated operating conditions based on environmental conditions. We relied on automated ML methods to identify the best-fitting ML algorithm and hyperparameters for this algorithm. Specifically, we employed Azure AutoML implementation of automated ML. Quadratic roll damping due to viscous effects and bilge keels is determined in virtual roll excitation tests using the CDF solver StarCCM+. Together with the loading condition and online-stability calculation, the response amplitude operators (RAOs) are assembled. With a directional wave spectrum, the significant and MPM-motions are calculated in a very short time.
The results are not only applied for the design of sea fastenings and other temporary project-specific equipment on deck. They can also be fed into the physics-engine of the full-mission-simulator, a physical marine operations simulator on-shore, which is used to instruct and train the crew, crane driver, and lifting supervisor in preparation for upcoming projects. Finally, the fast calculation enables the utilization of diffraction analysis results in the on-board condition monitoring system. This way, it shows not only the forecasted weather/wave conditions but also the motion responses thereof. This allows the crew to make better informed decisions especially in prospect of marginal operational conditions.
