Machine learning strategy for viscous calibration of fully-nonlinear liquid sloshing simulation in FLNG tanks: Fid-Bau Portal

Machine learning strategy for viscous calibration of fully-nonlinear liquid sloshing simulation in FLNG tanks

Zhang, Chongwei / Tan, Jie / Ning, Dezhi

Highlights • Fully-nonlinear simulation of liquid sloshing with viscous effects is considered. • A machine learning strategy is proposed to adaptively determine the damping coefficient. • Database of network training and testing is established from newly conducted physical experiments. • This strategy for viscous sloshing calibration has extremely high probability for accurate predictions.

Abstract This study considers the fully-nonlinear simulation of liquid sloshing in FLNG tanks. The mathematical model is established based on the potential-flow theory. Instantaneous boundary conditions are applied to track the large-amplitude free-surface deformation. An artificial damping method is introduced to involve the viscous dissipation effect in the sloshing process. The machine learning strategy is proposed to adaptively calibrate the associated damping coefficient based on the back-propagation neural network. The network is trained and tested using the database built from newly conducted physical experiments. Through statistical analyses, an optimised three-layer network with a hidden layer of seven neurons is formed. With the damping coefficients determined, fully-nonlinear simulations of the liquid sloshing are carried out in time domain based on the boundary element method. Pressure histories predicted by the numerical method are compared with those measured in physical experiments. The machine learning strategy shows a high probability to accurately predict the damping coefficient for nonlinear sloshing simulations. The present model has great potential to reproduce the nonlinear sloshing behaviour in a variety of cases.