Hi all
I am trying to simulate turbulent pipe flow using LES method.
Literature:
Eggels (1994) used L/D = 5 for his studies.
In 1999, Blackburn used L/D = 2pi().
Hwang et al (2016) used L/D = 22 for their LES studies on pipe flow.
Q1. How do we arrive at required L/D of a pipe or length of a channel flow etc. when LES is used. As LES is computationally expensive, we cannot do a grid sensitivity. Does the two point correlations of required parameters (velocity, pressure etc.) help to obtain the minimum length required.
Generally one sees at the spatial correlations, if they reaches zero, the lenght is sufficient.
If you are using periodic streamwise condition, the lenght of the domain must be as sufficient as to contain the vortical structure having the integral lenght. That is a physical consideration, valid for both LES and DNS.
Conversely, if you are prescribing inflow and outflow conditions, the lenght must be sufficient to let the flow physically correlate after the numerical transient at the inflow.
' If you are using periodic streamwise condition, the lenght of the domain must be as sufficient as to contain the vortical structure having the integral lenght '
In the case of a pipe, the largest eddy will be of size D and hence the domain should be minimum of length D. ie L/D =1. But how can we confirm whether the length of domain (D) is sufficient enough. Minimum L/D seen for LES is L/D=5 by Eggels.
For eg: In the case of channel flows, integral length of the eddies can be of the length of the channel itself. So how to limit the computational domain for a periodic infinite length channel flow is the doubt.
Whether two point correlations are the way to arrive at a conclusion OR is it the only way?
Ref: LES for Acoustics, P. Sagaut.
Generally one sees at the spatial correlations, if they reaches zero, the lenght is sufficient.
Dear Joemon, I agree with Prof. Denaro, see Veersteg-Malalasekera, Introduction to Computational Fluid Dynamics, Chap. 3, section "Flat plate bounday layer and pipe flow" for info on topic. Gianluca
1. Reducing the computation domain is never our goal. Our goal is to find the minimal computation domain sufficient.
2. In terms of getting 1st- and 2nd-order moments accurate, the computation domain can be very small. See, "The minimal flow unit in near-wall turbulence". For channel flow, x^+ ~ 1000 should be enough to give your correct Reynolds stress, so does pipe flow I presume.
3. If you want to well capture the coherent structures, the two-point correlation is a necessary requirement, but not sufficient. Energy spectrum is the ultimate tool to quantify the effects of computation domain size. See what I did in "Direct numerical simulation of turbulent flow in a spanwise rotating square duct at high rotation numbers".
Hi All
From literature studies, for a pipe flow, L/D = 12.5 is sufficient to capture almost all turbulent characteristics.
Ref: DNS of Turbulent Pipe flow at moderately high Re numbers, Khoury et al., Flow Turbulence Combust (2013).
Hi All
Spatial correlations are the right method to be used for finding the minimum spanwise length for an LES simulation.
For flow around cylinders, I have seen published works with L/D = 2 to L/D = 2pi(). No paper was seen beyond that L/D range as it is computationally expensive.
- Badges
- Science topic
- Similar topics
- Thermodynamics
- Pressure