Ensemble- and volume-averaged stochastic models for cavitating flows
Ensemble- and volume-averaging are phase-averaged methods for disperse, bubbly flows. While built upon similar assumptions, it is challenging to assess their relative merits. Volume-averaging is an intrinsically deterministic model, for which bubbles are represented in a Lagrangian framework as advected particles, each sampled from a distribution of equilibrium bubble sizes. The dynamic coupling to the liquid phase is modeled through local volume averaging. Ensemble-phase averaging is stochastic, and uses ensemble-averaging to derive mixture-averaged equations and the field equations are evolved in an Eulerian reference frame for the associated bubble properties, each representing bins of an underlying equilibrium distribution. In both cases the equations are closed by solving Rayleigh-Plesset-like equations for the bubble dynamics as forced by the local or mixture-averaged pressure, respectively. Computationally, there are complex tradeoffs between these two approaches, especially for modern, parallel architectures. We assess their relative complexity and cost via high-resolution simulations.
S. H. Bryngelson, A. Charalampopoulos, T. Sapsis, T. Colonius, "A Gaussian moment method and its augmentation via LSTM recurrent neural networks for the statistics of cavitating bubble populations," under review International Journal of Multiphase Flow (2019)
S. H. Bryngelson, K. Schmidmayer, T. Colonius, "A quantitative comparison of phase-averaged models for bubbly, cavitating flows," International Journal of Multiphase Flow 115, 137–143 (2019)