Workshop on Differentiable Marine Hydrodynamics Simulations and its Applications using MarineHydro.jl

Understanding and predicting wave-structure interactions is central to the design

and operation of offshore systems—from floating wind turbines and wave energy con-

verters to autonomous marine platforms. Traditionally, these interactions are modeled

using boundary element method (BEM) solvers based on linear potential flow the-

ory. However, a major limitation of existing BEM solvers is their inability to provide

sensitivities (derivatives) of hydrodynamic quantities with respect to design variables

such as geometry, wave frequency, or device layout. These sensitivities are essential

for enabling modern workflows in design optimization, control co-design, uncertainty

quantification, and physics-informed machine learning.

Without access to gradients, engineers are forced to rely on derivative-free meth-

ods (e.g., parameter sweeps, heuristics, or surrogate models without sensitivity guid-

ance), which scale poorly and often miss optimal solutions in high-dimensional design

spaces. This bottleneck has hindered progress in applying scalable, gradient-based op-

timization methods—now common in aerospace, robotics, and machine learning—to

the marine energy sector.

This workshop introduces MarineHydro.jl, a fully differentiable BEM solver that

unlocks access to exact sensitivities of hydrodynamic coefficients through reverse-

mode automatic differentiation. Developed in Julia, MarineHydro.jl supports both di-

rect and indirect BEM formulations and includes fast and accurate Green’s function

implementations. The workshop is targeted at researchers and practitioners working

on design and control of offshore and marine structures, with an emphasis on appli-

cations where sensitivity information is essential. Participants will learn to install the

solver, perform hydrodynamic simulations, extract gradients, and apply these results to

practical case studies.

We begin by reviewing the fundamentals of differentiable programming in the con-

text of potential flow theory and boundary element methods. The solver’s architecture

supports both direct and indirect BEM formulations and includes efficient implemen-

tations of Green’s function approximations, balancing accuracy and performance.

Application I: Surrogate Modeling with Gradient-Enhanced Learning. Tradi-

tional surrogate models rely solely on function evaluations, requiring dense sampling

for accurate approximations. By incorporating gradient information directly into model

training, MarineHydro.jl enables the construction of more data-efficient and accurate

surrogates. This is particularly valuable for design optimization, where high-fidelity

models are expensive to evaluate.

Application II: Multi-Body Interaction and Layout Optimization. Hydrody-

namic interactions between multiple floating bodies—such as wave energy convert-

ers—are critical for performance, yet challenging to optimize using traditional ap-

proaches. Using MarineHydro.jl, participants will explore sensitivity-based analyses

of inter-body effects and apply gradient-based optimization to study spatial layouts

and design variables. Case studies include gradient-based power optimization of WEC

arrays, marking the first use of exact hydrodynamic gradients for such purposes.

The availability of exact sensitivities opens up new avenues in multidisciplinary de-

sign optimization (MDO), control co-design, uncertainty quantification, and physics-

informed machine learning. By integrating differentiable hydrodynamics into the de-

sign loop, MarineHydro.jl paves the way for scalable, robust, and high-performance

workflows in the next generation of marine renewable energy systems.