Sloshing suppression with a controlled elastic baffle via deep reinforcement learning and SPH simulation

TL;DR

This paper presents a novel approach combining deep reinforcement learning and SPH simulation to optimize elastic baffle control for sloshing suppression, achieving over 80% reduction in surface amplitude.

Contribution

It introduces a DRL-based control strategy for elastic baffles in fluid tanks, outperforming traditional rigid baffle methods and providing a robust, adaptable solution for sloshing mitigation.

Findings

Elastic baffle control reduces sloshing amplitude by over 80%.

DRL-derived policies outperform rigid baffle control at higher frequencies.

Control effectiveness remains stable across varying water depths.

Abstract

This study employed smoothed particle hydrodynamics (SPH) as the numerical environment, integrated with deep reinforcement learning (DRL) real-time control algorithms to optimize the sloshing suppression in a tank with a centrally positioned vertical elastic baffle. Compared to rigid baffle movement and active strain control methods, the active-controlled movable elastic baffle, which remains undeformed at its base, achieved the best performance with an 81.63% reduction in mean free surface amplitude. A cosine-based expert policy derived from DRL data is also extracted, resulting in a comparable 76.86% reduction in a three-dimensional (3D) numerical simulation. Energy analyses showed that elastic baffle motion effectively decreased system energy by performing negative work on the fluid, reducing kinetic and potential energy. The DRL-based and expert policies also demonstrated robust performance across varying excitation frequencies and water depths. Specifically, rigid baffles proved more effective at frequencies below the system's first natural frequency, while elastic baffles exhibited superior performance at higher frequencies. Changes in water depth minimally affected the effectiveness of control policies, though they significantly influenced elastic baffle deformation behavior. Overall, the sloshing suppression efficiency consistently ranged between 70% and 80%, confirming DRL-informed control methods' versatility and effectiveness under diverse operating conditions.