Deep Learning-Driven Nonlinear Reduced-Order Models for Predicting Wave-Structure Interaction

TL;DR

This paper develops a physics-informed deep learning model combining LSTM and reduced-order modeling techniques to accurately predict wave-structure interactions, validated on a 2D wave basin scenario.

Contribution

It introduces a novel DEIM-LSTM and physics-informed LSTM approach for efficient and accurate wave-structure interaction predictions.

Findings

DEIM-LSTM effectively reduces system complexity.

Physics-informed LSTM improves prediction accuracy.

Model validated on 2D wave basin data.

Abstract

Long Short-Term Memory (LSTM) network-driven Non-Intrusive Reduced Order Model (NROM) for predicting the dynamics of a floating box on the water surface in a wavemaker basin is addressed in this study. The ground truth or actual data for these wave-structure interactions (WSI) problems, namely box displacements and hydrodynamic forces and moments acting on the box due to wave interaction corresponding to a particular wave profile, are computed using the Smoothed Particle Hydrodynamics (SPH). The dimensionality of the system is first reduced using the Discrete Empirical Interpolation Method (DEIM) and the LSTM is applied to the reduced system resulting in a DEIM-LSTM network for developing a surrogate for prediction. The network is further enhanced by incorporating the physics information into the loss function resulting in a physics-informed LSTM (LSTM-PINN) for predicting the rigid body dynamics of box motion. The performance of predictions for these networks is assessed for the two-dimensional wave basin WSI problem as a proof-of-concept demonstration.