At the same as pictures, the ship hull offset data is sequential in space, which indicates that the geometry information carried by the data is associated with their sequence of values. Theoretically, the convolutional neural network (CNN), which has been widely applied in modern daily life, could also identify the ship hull's geometric features. This paper initially demonstrated that computer vision techniques (such as CNN or VGG) can recognize the ship hull offset features like human designers, and provided some support for further ship modeling investigations and research.
The geometry of hull form affects the hydrodynamic performance, safety and economics of ships significantly. To reduce the emission of ships, the research on ship hull optimization methods is concentrated in recent years, and there are increasing investigators trying to apply artificial intelligence, machine learning, and deep learning to ship design. The computer vision (CV) technology could import the recognition functionally of geometry to programs, and there are many successful engineering applications already. As the pictures and voices, the ship offset table is a matrix carrying the information associated with certain sequences. In this case, it is possible to use CV to train a neural network to recognize the ship hull form features.
Ship hull form design and optimization is always a critical challenge to researchers. Pérez and Suárez introduced a method to create a quasi-developable B-spline surface between two limit curves is presented (Pérez & Suárez, 2007). Pérez and Clemente followed the traditional design principles of naval architecture, starting with a sectional area curve (SAC) that affects hydro statistics (Pérez & Clemente, 2011). Koelman and Veelo pointed out the limitation of NURBS on ship hull modeling (Koelman & Veelo, 2013). Yang and Huang applied a simulation-based hydrodynamic design optimization of ship hull forms at the early design stage, and reduced the total drag compared to the original design (Yang & Huang, 2016). Yu and Lee indicated a design method for the efficient gooseneck bulbous bow for the middle-sized ferry (Yu & Lee, 2017). Greshake and Bronsart solved the restriction of tensor-product B-splines to regular control meshes with a subdivision method to refine polygon meshes (Greshake & Bronsart, 2018). He and his team established a gradient-based optimization framework, including a discrete adjoint method for efficient derivative computations (He et al., 2019). Katsoulis et al presented a T-splines-based parametric modeler (TshipPM) for complex ship forms, capable to provide smooth geometries at low computational modeling cost (Katsoulis et al., 2019). Wei et al illustrated a reliability-based robust design optimization (RBRDO) framework for ship hull form design (Wei et al., 2019). Bašić et al introduced an improved linear theory by considering the boundary layer effects through the tangency correction that can handle flow separation (Bašić et al., 2020). Can et al investigated the wake factor of a self-propelled container ship by Telfer's GEOSIM method based on CFD (Can et al., 2020). Matveev applied numerical simulation on the ship resistance reduction with air injecting (Matveev, 2020). Miao and Wan illustrated a hydrodynamic optimization tool, OPTShip-SJTU, containing four components, a hull form modifier, performance evaluator, surrogate model building, and optimizer module (Miao & Wan, 2020). Moreira and Soares used an artificial neural network (ANN) to estimate wave-induced ship hull bending moment and shear force from ship motions (Moreira & Soares, 2020). Ni et al applied an SBD (simulation-based design) optimization framework to optimize the design of a SWATH (small water area twin-hull) (Ni et al., 2020). Zhang et al optimized the DTMB5415 with the ISIGHT platform (Zhang et al., 2020).
