Session: 08-06-01 non-presentations

Paper Number: 128013

128013 - Leveraging Physics-Informed Neural Network to Estimate Ship Resistance 

The biggest problem we are facing right at this moment is climate change. The maritime industry faces the pressing challenge of reducing carbon emissions to combat climate change issues. Ships powered by marine bunker fuels contribute significantly to greenhouse gas emissions, motivating the pursuit of innovative solutions for emission reduction and green maritime practices. This research aims to address this global crisis by leveraging the power of Physics-Informed Neural Networks (PINNs), a data-driven deep learning framework coupled with complex governing physical laws to estimate ship resistance, which is the fundamental factor in fuel consumption and carbon emission. Prominent maritime nations, including India, have pledged to reduce ship emissions by at least 20% by 2030, as agreed upon in the United Nations International Maritime Organisations's 2023 conference.
Furthermore, from India's perspective, its unique geographical landscape offers a plethora of untapped potential for maritime trade and transportation, including inland waterways, riverine, and coastal passages. However, designing new ships optimized for these diverse waterways is a complex and challenging endeavor, where resistance encountered by the ship hull form plays a crucial role. Green shipping can be achieved by reducing emissions beyond their current levels by enhancing their design, which involves minimizing resistance encountered by the ships while navigating the waters. Nowadays, designers mostly rely on traditional approaches like Computational Fluid Dynamics simulation, numerical simulation using potential flow theory, model tests and experimental findings. Resistance estimation through hull form optimization becomes particularly challenging when dealing with various ships with distinct geometry. Due to the considerable computational time and cost involved, it is nearly impractical for Naval Architects to manually create physical and computer-generated meshed models for every ship model proposed. Physics-Informed Neural Network automates the entire algorithm of ship resistance estimation and uncovers intricate relationships between hull parameters, ship geometry, and flow field. The neural network coupled with physical laws is utilized as a data-driven solution framework, which will take several input parameters like ship geometry, body coordinates, water parameters, hull form parameters, fluid-structure interaction points or the collocation points. Reynolds Averaged Navier Stokes (RANS) with K-Epsilon Turbulence model will be the governing physical laws, which will constrain the actual data-driven losses incurred from the boundary and initial conditions. Limited memory - BFGS optimizer is used to train the overall PINNs framework. The PINNs framework will predict pressure and wall shear stresses and include turbulent kinetic energy (K) and turbulence dissipation (E) to address the dynamic energy cascading near the ship's bow. Based on the output parameters, wall resistance encountered by the ship will be estimated. The resistance estimates for different hull forms will be validated against results obtained from simulations in OpenFOAM or commercial codes like Star-CCM. In conclusion, this research endeavours to provide a pioneering and computationally efficient solution to ship resistance estimation, contributing to reducing emissions and a green maritime frontier. Automating the entire process through applying PINNs mimicking the human brain promises to revolutionize the ship design industry.

Presenting Author: Sayan Chowdhury Indian Institute of Technology Madras

Presenting Author Biography: I am a PhD scholar in the Department of Ocean Engineering recently awarded the prestigious Prime Minister's Research Fellow (PMRF). My research aligns with ship resistance estimation, hull form optimisation, Climate change and Data-driven models using deep learning framework.

Authors:

Sayan Chowdhury Indian Institute of Technology Madras
Abhilash Somayajula Indian Institute of Technology Madras