Large eddy simulation of orientation and rotation of ellipsoidal particles in isotropic turbulent flows

Large-eddy simulation of turbulent dispersed flows: a review of modelling approaches

“…(30) is the contribution parallel to p needed to keep the strain from changing the magnitude of p. It is natural to expect significant errors in the calculation of p over time when the exact velocity gradients (available in DNS) are replaced with those computed from the resolved velocity fielded yielded by LES. In addition, SGS effects are expected also on the rotational dispersion coefficient of the particles: This coefficient depends on the turbulent energy dissipation rate, which is reduced when SGS velocity fluctuations are not accounted for [22]. Indeed, results from a priori tests demonstrate that SGS fluctuations mostly affect particle rotation, resulting in weaker particle alignment with the vorticity field and reduced particle rotational energy if neglected.…”

“…Direct comparison against DNS results shows that the stochastic model (LES-SDE) provides poor prediction of particle alignment and over-prediction (under-prediction) of the reconstructed rotational energy at large (small) aspect ratios. This is probably due to the fact that the stochastic part of the model used by [22] is Gaussian and uncorrelated in time: Therefore, the model cannot take into account any effect due to the anisotropy of particle rotation dynamics, which are correlated with preferential alignment phenomena, and any effect on particle rotational diffusivity due to the different correlation timescales characterizing the particle angular velocities [75]. The ADM-based model has little influence on particle alignment compared to a priori (FDNS) and a posteriori (LES) results, but improves prediction of the rotational energy for aspect ratios larger than unity: This is ascribed to the capability of ADM to recover fluid enstrophy near the cut-off scale [22].…”

“…This is probably due to the fact that the stochastic part of the model used by [22] is Gaussian and uncorrelated in time: Therefore, the model cannot take into account any effect due to the anisotropy of particle rotation dynamics, which are correlated with preferential alignment phenomena, and any effect on particle rotational diffusivity due to the different correlation timescales characterizing the particle angular velocities [75]. The ADM-based model has little influence on particle alignment compared to a priori (FDNS) and a posteriori (LES) results, but improves prediction of the rotational energy for aspect ratios larger than unity: This is ascribed to the capability of ADM to recover fluid enstrophy near the cut-off scale [22]. Overall, the difference between the reference DNS results and the results obtained with or without any of the particle SGS models is either marginal or increased indicating that further work is required to improve model predictions with non-spherical particles.…”

“…Most of the numerical simulations performed so far are based on DNS of turbulence, but LES is becoming more and more attractive in view of the continuous improvements in the SGS models for both the fluid and the particles. In particular, a recent application of LES to non-spherical particles in the EL framework has been presented by Chen et al [22]. These authors have considered the effect of including a subgrid closure in the equation of motion of small ellipsoidal particles evolving in HIT and have quantified this effect in terms of particle alignment within the flow and particle orientational distribution.…”

“…To recover these effects, both a stochastic SGS model and a model based on ADM have been tested. It should be noted that adopting a stochastic closure for the fluid velocity seen (which typically involves a Wiener process with uncorrelated independent increments and continuous trajectories that are nowhere differentiable) would yield a velocity field that is non-differentiable and velocity gradients would be unavailable: Therefore, a Lagrangian stochastic model for the SGS velocity gradient tensor seen by the particles was adopted (details about this model can be found in [22] and references therein). An example of the performance of the different models is shown in Fig. 14, where particle orientation within the flow is quantified by means of two observables.…”