ShipNET – AI Ship Modelling Approach for Multidimensional Design
2024 – 2028

With the ambitious goal of decarbonising the shipping industry and reducing greenhouse gas emissions by at least 50% by 2050 compared to 2008 levels, the International Maritime Organization (IMO) has introduced regulatory measures such as the Energy Efficiency Existing Ship Index (EEXI), the Energy Efficiency Design Index (EEDI), and the Carbon Intensity Indicator (CII). These measures enforce stricter standards for ship design and operation, compelling shipowners and designers to adopt innovative solutions that improve energy efficiency and environmental performance.

In this context, retrofitting has become an important market area to reduce hydrodynamic resistance and fuel consumption for existing fleets. Vessel shape optimization not only helps with regulatory compliance, but also contributes to operational cost savings and a reduction in environmental impact.

Traditionally, the development of hull design is highly dependent on the designer’s experience, making the process dependent on his skills, difficult to standardize and limiting the systematic search for optimal design solutions.

In modern ship design, parametrically based CFD optimization methods are considered state of the art. However, they have inherent limitations. Performing high-fidelity CFD simulations for each design iteration is computationally intensive and impractical, especially in complex design spaces using the Design of Experiments (DOE) approach. Recent developments in commercial CAE programs, enable fully parameter-driven, systematic and automated workflows for ship design. These tools allow designers to explore a wider range of design configurations while reducing manual effort. However, significant challenges remain, based on the need for the optimization algorithm to explore the entire parameter space. This is accompanied by the calculation of many non-practicable designs. Identifying these designs in the optimization process is a key issue. Without a structured approach to eliminate impractical or unusable designs early on, computational resources are wasted evaluating unfeasible configurations, reducing the overall efficiency of optimization.

The ShipNET project aims to overcome these limitations by developing an AI-driven, object-oriented, and modular workflow for ship design, in which adaptive AI agents act as collaborative co-pilots throughout the optimization process. These agents autonomously generate and refine hull variants, assess feasibility, predict hydrodynamic performance, and ensure compliance with physical and regulatory constraints. The objective is to create an intelligent, self-adaptive design pipeline that accelerates exploration, reduces manual intervention, and enables efficient navigation of complex design spaces.

As part of the preliminary study and development phase, two AI agents are to be developed:

• Reinforcement learning agent for automated and autonomous generation of design variants, enabling the agent to steer design exploration toward feasible, high-performing regions of the design space while avoiding non-viable configurations.
• AI-surrogate prediction agent for rapid evaluation of design variants, providing fast performance predictions to replace expensive CFD simulations within the design loop.