Image-Based Hybrid Neural Network Model for Naval Ship Wave Resistance Prediction

Abstract

Accurate prediction of ship wave-making resistance is a critical challenge in naval architecture, particularly during the preliminary design stage. This study presents a comprehensive hybrid artificial intelligence (AI) framework for predicting the wave-making resistance coefficient (CW) of the DTMB 5415 naval hull model, integrating both numerical hull parameters and image-derived hydrodynamic features. A systematic parametric study was conducted by varying the hull’s principal dimensions—length, beam, and draft—by ±25% from their nominal values, resulting in 135 distinct hull configurations, where for each combination, CW is computed using Maxsurf software (Academic Version 25). Corresponding wave fields are captured as images and preprocessed through resizing, grayscale conversion, contrast enhancement, and edge detection to emphasize key hydrodynamic characteristics for AI training. A dual neural network architecture is employed, combining a Feed-Forward Artificial Neural Network (CW) for numerical inputs with a Convolutional Neural Network (CNN) for image-based feature extraction. The hybrid model demonstrated superior predictive performance, achieving a coefficient of determination (R2) exceeding 0.99, significantly outperforming standalone FFAN and CNN models. This study contributes a novel, physically interpretable AI framework capable of capturing complex nonlinear interactions between hull geometry and wave patterns, providing a reliable and computationally efficient alternative to towing tank experiments and high-fidelity CFD simulations.