Machine Learning-Based Resistance Prediction of AMECRC Hull

Abstract: Computational fluid dynamics (CFD) is a pivotal tool for naval architects seeking to optimize ship designs by accurately predicting resistance. However, the computational demands of CFD simulations pose significant challenges. This study focuses on developing a digital tool enabling the prediction of resistance components of the Australian Maritime Engineering Cooperative Research Center (AMECRC) hull. The application of machine learning (ML) techniques in ship design is becoming increasingly popular in recent days. By leveraging ML techniques, we developed a surrogate model based on experimental data of 14 hulls from the AMECRC systematic series across a range of Froude numbers (0.3-1.0). Linear regression model, efficiently trained linear regression model, support vector machines (SVM), Gaussian process regression (GPR) model, Kernel approximation regression model and neural network-based model are compared to conclude that the neural network-based surrogate model performs best in predicting resistance components while significantly reducing computational time compared to CFD simulations. Moreover, a user-friendly GUI application is also developed to assist naval architects in predicting resistance components for any hull within the parameter space of the AMECRC systematic series. This tool enhances efficiency and empowers designers to make informed decisions, ultimately leading to the development of more optimized ship designs. Keywords: Machine Learning, Computational Fluid Dynamics, AMECRC Hull, Resistance