Reduced-order modeling of temporal local scour dynamics beneath a submerged cylinder in a steady flow using autoencoders

Document Type

Article

Publication Date

8-1-2025

Abstract

Understanding and predicting local scour beneath submerged cylinders is critical for assessing structural stability under both operational and extreme conditions, and for designing effective countermeasures. Given the transient nature of scour initiation and development, models capable of capturing temporal dynamics are preferred. Although two-phase flow models can resolve these features, they are computationally expensive and impractical for rapid predictions. To address this challenge, this study investigates the effectiveness of advanced data-driven techniques—Proper Orthogonal Decomposition with Long Short-Term Memory networks (LSTM) and β -Variational Autoencoders with LSTM—for emulating sediment transport simulations. Using a dataset generated with sedFoam, we evaluate these models in terms of accuracy and computational efficiency.

Publication Source (Journal or Book title)

Physics of Fluids

This document is currently not available here.

Share

COinS