Interactive ship flow simulation enhanced by neural network model in a web environment

Abstract

Hydrodynamic simulation of marine structures is a complex and time-consuming task that requires large, refined models to accurately estimate the behavior of ships. During the conceptual phase, therefore, these estimations may be more efficient if done with a mix of surrogate models and simplified simulations. We believe that AI and the web environment can contribute to providing a more precise answer and fast solution, especially when the design domain can be narrowed and properly estimated. This paper shows an attempt in this direction, describing a web-based real-time flow simulator that is composed of a Tenforflow.js-based convolutional neural network model with an image-based hull form representation. Some case studies demonstrate the advantages of a novel web-based prototyping environment in the conceptual and initial design of ships. The image-based hull form representation method with a convolutional neural network enables the design of not only main dimensions but also local shapes in an interactive web-based concurrent engineering environment. Our approach extends the neural network model of wake flow estimation to models of the prediction of resistance and pressure distributions on the hull surface and develops a novel web-based prototyping environment for the conceptual and initial design of ships.