Gaussian process regression for the side-by-side foil pair

Abstract

The mutual interaction among multiple fish during schooling has significant implication on motion pattern control and hydrodynamic optimization. However, the collective motion of multiple objects in a flow field forms a vast parameter space, causing difficulty in comprehensively analyzing and considering each parameter. To address this issue, the problem is simplified to a foil pair oscillating in a side-by-side configuration in a two-dimensional flow. Moreover, the Gaussian process regression predictive algorithm is combined with the fast and robust boundary data immersion method CFD algorithm to form a iteration loop for value prediction of the large parameter space. Through a relatively small number of simulations (around 1000 data points), we obtained predictions for the entire four-dimensional parameter space that consists of more than 160 000 parameter sets, greatly improving the computational efficiency. After obtaining the predicted space, we analyzed the interactions between different parameters and specially described the mechanism that gives rise to the unique effect of phase difference on the efficiency of the overall system and individual foils.