Unphysical pressure oscillations
Dear all, I have been trying to set up a time-dependent 2D external flow incompressible Navier-Stokes computation in the transitional regime (separated flow around airfoil at Re~5000, aoa=12) but I get unphysical pressure oscillations across the domain. The main features of the flow, vortex shedding frequency, aerodynamic forces mean value and oscillation amplitude are captured with reasonable resolution. There appear, however, high frequency pressure oscillations whose amplitude is reduced by an increase in resolution without ever completely disappearing. The velocity field seems to be little affected by this pressure oscillations, but they are however quite annoying. I have tried, but doesn't seem to have any effect whatsoever: - many different meshes - usual outlet boundary conditions (Neumann for velocity, Dirichlet for pressure) and HOutflow BC. - nodal and modal expansions. (btw, what is the difference between GLL_Lagrange and GLL_Lagrange_SEM?) - convective and skew-symmetric forms for the advection term - 1st 2nd and 3rd order time integration with the IMEX time-stepper - Reducing the timestep Dealiasing with SpectralHPDealiasing has some effect, but not much. What seems to work quite well in smoothing the pressure oscillations is using SmoothAdvection (I am not sure of what this does), except that the computation eventually destabilises and produces NaNs. I also wanted to try the inf-sup condition and use different expansion orders for pressure and velocity, but this does not seem to be possible other than with the direct solver, which cannot be used for fully nonlinear transient computations. Another option I considered was using the unconditionally stable rotational velocity-correction scheme of Dong & Shen (JCP 2010) but I think it is not implemented in the IncNavierStokesSolver. Please find attached the files to run one of the many tests I have performed. Any help will be welcome. Many thanks in advance. Cheers, Fer
Hi Fer, Probably Douglas might be able to comment on this type of simulation since he has just been completing some runs as part of his PhD which has now been examined and accepted. I attach a copy of a paper under review which has airfoil situations at Re=1000 at this type of aoa. My only other comment is that we have seen pressure oscillations when the surface is not curved but I note from your .xml file you do have curved elements. I have not checked how smooth the curved elements you are using. Are there any issues on this side. Finally it is possible to specify an inf-sup compatible space for the velocity correction scheme. Definitely using a Modified Expansion and possibly/probably for the Lagrange basis. However the velocity and pressure quadrature spaces have to be the same. The way to specify such a basis is using an Expansion definition along the lines of: <E COMPOSITE=“C[0]” BASISTYPE=“Modified_A,Modified_A" NUMMODES="3,3" POINTSTYPE="GaussLobattoLegendre,GaussLobattoLegendre" NUMPOINTS="6,6" FIELDS="p" /> <E COMPOSITE=“C[0]” BASISTYPE="Modified_A,Modified_A" NUMMODES="5,5" POINTSTYPE="GaussLobattoLegendre,GaussLobattoLegendre" NUMPOINTS="6,6" FIELDS="u,v" /> Cheers, Spencer.
Hi Spencer, Many thanks for the quick answer. I tried first using the modal expansion as suggested with same quadrature order but different mode count for pressure and velocity, but did not to work. The other suggestion regarding the curved elements, however, left me thinking. I am using third order equispaced polynomials to represent the curved edges. To check if that could be the issue I refined the mesh around the leading edge and the oscillation disappeared completely, so it seems that the simulation was simply under-resolved in the leading edge after all. Thanks again for your help, because I had been struggling and increasing resolution everywhere except where it was really needed... Cheers, Fer On 05/10/16 18:49, Sherwin, Spencer J wrote:
Hi Fer,
Probably Douglas might be able to comment on this type of simulation since he has just been completing some runs as part of his PhD which has now been examined and accepted.
I attach a copy of a paper under review which has airfoil situations at Re=1000 at this type of aoa.
My only other comment is that we have seen pressure oscillations when the surface is not curved but I note from your .xml file you do have curved elements. I have not checked how smooth the curved elements you are using. Are there any issues on this side.
Finally it is possible to specify an inf-sup compatible space for the velocity correction scheme. Definitely using a Modified Expansion and possibly/probably for the Lagrange basis. However the velocity and pressure quadrature spaces have to be the same. The way to specify such a basis is using an Expansion definition along the lines of:
<E COMPOSITE=“C[0]” BASISTYPE=“Modified_A,Modified_A" NUMMODES="3,3" POINTSTYPE="GaussLobattoLegendre,GaussLobattoLegendre" NUMPOINTS="6,6" FIELDS="p" /> <E COMPOSITE=“C[0]” BASISTYPE="Modified_A,Modified_A" NUMMODES="5,5" POINTSTYPE="GaussLobattoLegendre,GaussLobattoLegendre" NUMPOINTS="6,6" FIELDS="u,v" />
Cheers, Spencer.
HI Fer, OK glad it worked out in the end. When you say the modal expansion did not work, did you mean it did not solve your issue or it did not execute? Cheers, Spencer. On 7 Oct 2016, at 11:43, F Mellibovsky <fernando.mellibovsky@upc.edu<mailto:fernando.mellibovsky@upc.edu>> wrote: Hi Spencer, Many thanks for the quick answer. I tried first using the modal expansion as suggested with same quadrature order but different mode count for pressure and velocity, but did not to work. The other suggestion regarding the curved elements, however, left me thinking. I am using third order equispaced polynomials to represent the curved edges. To check if that could be the issue I refined the mesh around the leading edge and the oscillation disappeared completely, so it seems that the simulation was simply under-resolved in the leading edge after all. Thanks again for your help, because I had been struggling and increasing resolution everywhere except where it was really needed... Cheers, Fer On 05/10/16 18:49, Sherwin, Spencer J wrote: Hi Fer, Probably Douglas might be able to comment on this type of simulation since he has just been completing some runs as part of his PhD which has now been examined and accepted. I attach a copy of a paper under review which has airfoil situations at Re=1000 at this type of aoa. My only other comment is that we have seen pressure oscillations when the surface is not curved but I note from your .xml file you do have curved elements. I have not checked how smooth the curved elements you are using. Are there any issues on this side. Finally it is possible to specify an inf-sup compatible space for the velocity correction scheme. Definitely using a Modified Expansion and possibly/probably for the Lagrange basis. However the velocity and pressure quadrature spaces have to be the same. The way to specify such a basis is using an Expansion definition along the lines of: <E COMPOSITE=“C[0]” BASISTYPE=“Modified_A,Modified_A" NUMMODES="3,3" POINTSTYPE="GaussLobattoLegendre,GaussLobattoLegendre" NUMPOINTS="6,6" FIELDS="p" /> <E COMPOSITE=“C[0]” BASISTYPE="Modified_A,Modified_A" NUMMODES="5,5" POINTSTYPE="GaussLobattoLegendre,GaussLobattoLegendre" NUMPOINTS="6,6" FIELDS="u,v" /> Cheers, Spencer. Spencer Sherwin McLaren Racing/Royal Academy of Engineering Research Chair, Professor of Computational Fluid Mechanics, Department of Aeronautics, Imperial College London South Kensington Campus London SW7 2AZ s.sherwin@imperial.ac.uk<mailto:s.sherwin@imperial.ac.uk> +44 (0) 20 759 45052
It executed fine but did not solve my problem with pressure oscillations. I mean that defining the same or different expansions for velocity and pressure did not change anything. On 07/10/16 12:57, Sherwin, Spencer J wrote:
HI Fer,
OK glad it worked out in the end. When you say the modal expansion did not work, did you mean it did not solve your issue or it did not execute?
Cheers, Spencer.
On 7 Oct 2016, at 11:43, F Mellibovsky <fernando.mellibovsky@upc.edu <mailto:fernando.mellibovsky@upc.edu>> wrote:
Hi Spencer,
Many thanks for the quick answer. I tried first using the modal expansion as suggested with same quadrature order but different mode count for pressure and velocity, but did not to work. The other suggestion regarding the curved elements, however, left me thinking. I am using third order equispaced polynomials to represent the curved edges. To check if that could be the issue I refined the mesh around the leading edge and the oscillation disappeared completely, so it seems that the simulation was simply under-resolved in the leading edge after all. Thanks again for your help, because I had been struggling and increasing resolution everywhere except where it was really needed...
Cheers, Fer
On 05/10/16 18:49, Sherwin, Spencer J wrote:
Hi Fer,
Probably Douglas might be able to comment on this type of simulation since he has just been completing some runs as part of his PhD which has now been examined and accepted.
I attach a copy of a paper under review which has airfoil situations at Re=1000 at this type of aoa.
My only other comment is that we have seen pressure oscillations when the surface is not curved but I note from your .xml file you do have curved elements. I have not checked how smooth the curved elements you are using. Are there any issues on this side.
Finally it is possible to specify an inf-sup compatible space for the velocity correction scheme. Definitely using a Modified Expansion and possibly/probably for the Lagrange basis. However the velocity and pressure quadrature spaces have to be the same. The way to specify such a basis is using an Expansion definition along the lines of:
<E COMPOSITE=“C[0]” BASISTYPE=“Modified_A,Modified_A" NUMMODES="3,3" POINTSTYPE="GaussLobattoLegendre,GaussLobattoLegendre" NUMPOINTS="6,6" FIELDS="p" /> <E COMPOSITE=“C[0]” BASISTYPE="Modified_A,Modified_A" NUMMODES="5,5" POINTSTYPE="GaussLobattoLegendre,GaussLobattoLegendre" NUMPOINTS="6,6" FIELDS="u,v" />
Cheers, Spencer.
Spencer Sherwin McLaren Racing/Royal Academy of Engineering Research Chair, Professor of Computational Fluid Mechanics, Department of Aeronautics, Imperial College London South Kensington Campus London SW7 2AZ
s.sherwin@imperial.ac.uk <mailto:s.sherwin@imperial.ac.uk> +44 (0) 20 759 45052
Dear Spencer, After writing that the problem was solved by increasing the resolution I realised that I was wrong. I had mistakenly left the option "SmoothAdvection" on, which damped the oscillations but crashed inadvertently after some time. Reducing the timestep does not solve the issue, and I am not sure of what the smoothadvection option does all in all. So my two next natural steps would be: 1) What does "SmoothAdvection" do? Is there any article where this is explained? Is there any some know circumstances under which it may destabilise a simulation? I have realised that the computation seems to crash when the first strong vortices reach the domain exit. Could the boundary conditions there (I use houtflow) somehow destabilise the computation? 2) You mentioned in your email that curved elements that are not sufficiently smooth can be troublesome. I am using gmsh to generate the geometry and the mesh. I have been using splines for the geometry and the high order mesh option to implement curved edges (polynomial on equispaced points) on the airfoil surface. Are there any known issues with curved edges produced with gmsh? When I visualise solutions on paraview, the interelement points seem to follow a smooth curved edge, but I don't know how smooth the edges are in reality. Many thanks in advance, and sorry for coming back with the same problem. Cheers On 05/10/16 18:49, Sherwin, Spencer J wrote:
Hi Fer,
Probably Douglas might be able to comment on this type of simulation since he has just been completing some runs as part of his PhD which has now been examined and accepted.
I attach a copy of a paper under review which has airfoil situations at Re=1000 at this type of aoa.
My only other comment is that we have seen pressure oscillations when the surface is not curved but I note from your .xml file you do have curved elements. I have not checked how smooth the curved elements you are using. Are there any issues on this side.
Finally it is possible to specify an inf-sup compatible space for the velocity correction scheme. Definitely using a Modified Expansion and possibly/probably for the Lagrange basis. However the velocity and pressure quadrature spaces have to be the same. The way to specify such a basis is using an Expansion definition along the lines of:
<E COMPOSITE=“C[0]” BASISTYPE=“Modified_A,Modified_A" NUMMODES="3,3" POINTSTYPE="GaussLobattoLegendre,GaussLobattoLegendre" NUMPOINTS="6,6" FIELDS="p" /> <E COMPOSITE=“C[0]” BASISTYPE="Modified_A,Modified_A" NUMMODES="5,5" POINTSTYPE="GaussLobattoLegendre,GaussLobattoLegendre" NUMPOINTS="6,6" FIELDS="u,v" />
Cheers, Spencer.
Hi Fer, After writing that the problem was solved by increasing the resolution I realised that I was wrong. I had mistakenly left the option "SmoothAdvection" on, which damped the oscillations but crashed inadvertently after some time. Reducing the timestep does not solve the issue, and I am not sure of what the smoothadvection option does all in all. So my two next natural steps would be: 1) What does "SmoothAdvection" do? Is there any article where this is explained? Is there any some know circumstances under which it may destabilise a simulation? I have realised that the computation seems to crash when the first strong vortices reach the domain exit. Could the boundary conditions there (I use houtflow) somehow destabilise the computation? The SmoothAdvection method was put in by Alessandro Bolis to emulate Prism. I believe it is just performing a C0 projection and then does a backward transform which for a nodal quad basis has a diagonal mass matrix and is simple to implement. It is not very straight forward for non quadrilateral or modal expansions. Looking at the code in ContField2D this indeed does seem to be what it does. Are you using Dealiasing and/or SpectralVansishingViscosity since these may both help. 2) You mentioned in your email that curved elements that are not sufficiently smooth can be troublesome. I am using gmsh to generate the geometry and the mesh. I have been using splines for the geometry and the high order mesh option to implement curved edges (polynomial on equispaced points) on the airfoil surface. Are there any known issues with curved edges produced with gmsh? When I visualise solutions on paraview, the interelement points seem to follow a smooth curved edge, but I don't know how smooth the edges are in reality. That does sound OK. I leave it to others to comment on gmsh since I get mixed results on how well the high order stuff works. Cheers, Spencer. Many thanks in advance, and sorry for coming back with the same problem. Cheers On 05/10/16 18:49, Sherwin, Spencer J wrote: Hi Fer, Probably Douglas might be able to comment on this type of simulation since he has just been completing some runs as part of his PhD which has now been examined and accepted. I attach a copy of a paper under review which has airfoil situations at Re=1000 at this type of aoa. My only other comment is that we have seen pressure oscillations when the surface is not curved but I note from your .xml file you do have curved elements. I have not checked how smooth the curved elements you are using. Are there any issues on this side. Finally it is possible to specify an inf-sup compatible space for the velocity correction scheme. Definitely using a Modified Expansion and possibly/probably for the Lagrange basis. However the velocity and pressure quadrature spaces have to be the same. The way to specify such a basis is using an Expansion definition along the lines of: <E COMPOSITE=“C[0]” BASISTYPE=“Modified_A,Modified_A" NUMMODES="3,3" POINTSTYPE="GaussLobattoLegendre,GaussLobattoLegendre" NUMPOINTS="6,6" FIELDS="p" /> <E COMPOSITE=“C[0]” BASISTYPE="Modified_A,Modified_A" NUMMODES="5,5" POINTSTYPE="GaussLobattoLegendre,GaussLobattoLegendre" NUMPOINTS="6,6" FIELDS="u,v" /> Cheers, Spencer. Spencer Sherwin McLaren Racing/Royal Academy of Engineering Research Chair, Professor of Computational Fluid Mechanics, Department of Aeronautics, Imperial College London South Kensington Campus London SW7 2AZ s.sherwin@imperial.ac.uk<mailto:s.sherwin@imperial.ac.uk> +44 (0) 20 759 45052
participants (3)
-
F Mellibovsky
-
Fer Mellibovsky
-
Sherwin, Spencer J