Abstract
While flip-through imposes the highest impact pressure reportedly from physical measurements, its numerical validation with experiment has been inconsistent and velocity field evolution has not been verified yet. To validate and better understand the flip-through impact phenomenon, which is highly nonlinear and stochastic, we comparatively examine δ SPH on GPU and Incompressible SPH on CPU, where the parameters that predict the flip-through accurately are determined through Bayesian approach as well as parametric search, respectively. The four sets of results similarly present that the numerical flip-through impact pressures are well captured in agreement with the experiment data at all gauges especially in terms of the pressure time series. Taking the validated simulations for the pressure, we verify the velocity fields are in good agreement from focusing to jetting and identify the excessive impact pressure is caused when the merge occurs close to the wall. To mitigate the excessive impact pressure, we perform Bayesian optimization of geometric modification reducing the maximum pressure by 49%. We confirm that the updated features of boundary condition, time integration, and density diffusion result in marginal prediction variations of pressure time series and free surface evolution while minimizing the gap along the wall and non-uniform particles.