Machine Learning Cuts Ship Design Time and Cost

Scientists are increasingly focused on optimising ship hull designs to reduce resistance and improve efficiency, a complex undertaking particularly when considering advanced propulsion systems. Moloud Arian Maram from J.M. Voith SE & Co. KG, Georgios Bletsos and Thomas Rung from the Institute for Fluid Dynamics and Ship Theory at Hamburg University of Technology, alongside Thanh Tung Nguyen and Michael Palm from J.M. Voith SE & Co. KG, and Ahmed Hassan working with colleagues at the Institute for Fluid Dynamics and Ship Theory, have developed a novel optimisation framework addressing these challenges. Their research introduces a Conditional Variational Autoencoder (CVAE)-based surrogate model to represent the time-averaged flow field generated by Voith Schneider Propellers, significantly reducing computational demands traditionally associated with adjoint-based shape optimisation. This collaborative work demonstrates that accurately modelling propulsion systems is critical for effective hull design, achieving over an 8% reduction in resistance compared to designs neglecting propeller effects and offering a pathway towards more efficient and sustainable maritime transport.

Scientists are tackling the immense computational burden of designing more efficient ships. Optimising hull shapes for vessels with complex propulsion systems traditionally demands prohibitive computing power and storage. A novel machine learning technique promises to accelerate this process, delivering substantial reductions in drag and fuel consumption.

Scientists have developed a new optimisation framework for ship hull design that leverages machine learning to overcome longstanding computational hurdles. The work addresses a critical need in naval architecture: minimising ship resistance while accounting for the complex interactions of advanced propulsion systems. This surrogate model accurately replicates the time-averaged flow field generated by the Voith Schneider Propeller, effectively replacing the detailed propeller geometry and its time-dependent behaviour with a data-driven approximation.

Initial tests demonstrate that this approach significantly reduces computational costs while maintaining the necessary accuracy for reliable hull optimisation. The team’s work reveals that neglecting the influence of the propulsion system during hull design can lead to suboptimal results, even increasing resistance compared to the initial hull shape.

In contrast, the proposed method, integrating the CVAE-based propulsion surrogate, achieves resistance reductions exceeding 8%. This advancement promises to unlock more efficient ship designs and contribute to reducing the environmental impact of maritime transport by lowering fuel consumption and associated emissions.

Hull resistance minimisation via propeller-aware shape optimisation

Optimisation studies utilising the proposed framework achieved more than an 8% reduction in total ship resistance compared to baseline hull forms. The surrogate model accurately replicates the time-averaged flow field induced by the Voith Schneider Propeller, replacing the need for geometrically and temporally resolving the propeller itself.

Primal flow verification confirmed that this data-driven approximation maintains necessary accuracy while delivering significant computational savings. Specifically, the research demonstrates that neglecting the influence of the propulsion system during hull optimisation can lead to designs exhibiting increased resistance when evaluated under realistic, propelled conditions.

Initial shapes, optimised without considering propeller effects, showed a performance degradation when propulsion was reintroduced. Shapes generated using the AI-assisted method consistently outperformed these, achieving the aforementioned 8% resistance reduction. Residual blocks and self-attention mechanisms were incorporated to further enhance the accuracy of the surrogate model’s output. Data transfer between the machine learning model and the optimisation study is restricted to flow velocities, utilising a meta grid extending beyond the propeller’s swept area. This machine learning approach circumvents the need for lengthy, time-resolved simulations by learning to replicate the time-averaged flow field generated by a Voith Schneider Propeller, a type of marine propulsion system with rotating and pitching blades.

Rather than directly simulating the propeller’s geometry and motion at each time step, the surrogate model provides a data-driven approximation, significantly reducing computational demands. The methodology begins with detailed simulations of the Voith Schneider Propeller operating within a representative flow environment. The encoder network compresses the complex flow data into a lower-dimensional latent space, while the decoder reconstructs the flow field from this compressed representation.

To validate the surrogate model’s accuracy, primal flow verification studies were conducted, comparing the flow fields predicted by the surrogate model with those obtained from full, geometrically-resolved CFD simulations. This ensured that the data-driven approximation maintained sufficient fidelity for use in the subsequent optimisation process. The advantage of this approach lies in its ability to replace a computationally expensive, time-dependent propeller model with a fast, static representation, enabling optimisation studies that would otherwise be impractical.

Furthermore, the research deliberately avoids replacing the propeller with a body force or uniform inflow, recognising that such simplifications can lead to suboptimal hull designs. Instead, the surrogate model captures the nuanced interaction between the propeller and the hull, allowing for a more accurate assessment of resistance and a more effective optimisation process. This careful methodological choice ensures that the resulting hull shapes genuinely benefit from the integration of the propulsion system.

Machine learning streamlines Voith Schneider Propeller modelling for enhanced ship resistance prediction

The relentless pursuit of hydrodynamic efficiency in ship design has long been constrained by computational bottlenecks. For decades, naval architects have relied on painstakingly detailed simulations to refine hull shapes, but the addition of complex propulsion systems, vital for realistic performance prediction, has pushed those simulations to the breaking point.

The problem isn’t simply one of processing power, it’s the exponential growth in data and the time required to model transient effects accurately. This work offers a clever sidestep, employing machine learning not as a replacement for physics, but as an intelligent proxy. By training a variational autoencoder on the flow field generated by a Voith Schneider Propeller, the researchers have created a ‘digital twin’ capable of mimicking the propeller’s influence with a fraction of the computational cost.

This isn’t about sacrificing accuracy; the results demonstrate a significant reduction in resistance, over eight percent, without compromising fidelity. More importantly, it highlights the danger of simplifying the problem by ignoring the propeller altogether, a common practice that can lead to suboptimal designs. The real potential lies not just in optimising individual hull forms, but in creating a dynamic design space where countless variations can be rapidly evaluated, paving the way for a new era of bespoke ship design tailored to specific operational needs and environmental constraints.