Combining turbulence modelling techniques may address traditional limitations

In the Large Eddy Simulation approach for modelling turbulence, scientists filter the velocity field to resolve features in which they are interested but need a model for the effects of unresolved features. A common approach is to use the Smagorinsky model, which includes a tunable coefficient relating the length scale of residual turbulent motion to the width of the filter.

One way to set this coefficient is to calculate it dynamically, which results in a loss of some of the physics by using an average coefficient. The other option is the conditional source-term estimation (CSE) model, which uses conditional averages for related variables and may suggest a more natural selection of the Smagorinsky coefficient. In a new paper, researchers apply CSE conditioning techniques to dynamic models.

“In real turbulent flow, the flow structures are moving around all the time, and grouping data based on a static set of regions doesn’t capture that particularly well,” said author Graham R. Hendra.

To address this limitation, the researchers combined the CSE approach with the dynamic procedure for calculating the Smagorinsky coefficient.

The researchers tested their mixed model on the Sandia flame series. Though they found no practical improvement in the model’s prediction as compared to traditional CSE and dynamic techniques, this is likely due to their lack of computational resources rather than a shortcoming of the model. The authors note that future work can address this limitation as well as investigate the viability of other model combinations.

“This work can be applied anywhere that dynamic turbulent models are currently used,” said Hendra. “By changing what information on which the data is conditioned, it could be applied in other turbulent systems.”

Source: “Conditional dynamic subfilter modeling,” by Graham R. Hendra and W. Kendal Bushe, Physics of Fluids (2019). The article can be accessed at https://doi.org/10.1063/1.5098813 .