Fluid Inlet Velocity Boundary

SRD

I need to define a "Velocity Inlet" boundary condition for fluid.
For instance, I need to define a specific fluid velocity at inlet of a pipe. I can create it by a moving piston, but I want to know if there are any other ways?

mn61

I am doing something similar right now.
You can use the initial velocity or a tank model where the fluid flows out of the tank leading to your fluid domain.
In my observation the moving piston has a negative effect on the fluid density because of the compressible sph approach.

SRD

The Initial velocity dose not work, because it is just valid at the start situation and after a while it will be affected by the other particles.

In the case of using tank, you have to increase the particle number to be able to create the velocity. It increases the cpu time and is not a good suggestion.

I think there should be another way.

mn61

Yes...you can implement the boundary condition on your own. There are some good approaches in the available SPH literature. Another possibility is to patch the velocity of the particles when they approach defined coordinates.
Both methods require source code modifications !

COertel

Hi all,

sorry for answering on such an old posting, but mn61´s comment on " patch the velocity" sounds interesting. Is this already an available function or is it something, which one needs to implement?

Greetings

Chris

alexmees

I needed exactly this for my simulation to work. I implemented a simple Boundary condition of XPeriodicIncZ but the outlet velocity was too great for the Inlet condition and this messed things up. I change the source code to alter all the velocity of particles which suffered position changes relating periodic boundaries to v=[0,0,0]. This satisfied my specific need, tough your may be different.

My solution although not elegant/very crude is this:
(find and change appropriate lines in code)

* JPeriodicGpu.h
void CheckPositionAll(bool boundchanged,float3* posg, float3* velg);
///

* JPeriodicGPU.cpp
void JPeriodicGpu::CheckPositionAll(bool boundchanged,float3* posg, float3* velg)
...
cuperi::CheckPositionAll(n,pini,posg,xrun,xmin,xmax,xperinc,yrun,ymin,ymax,yperinc,zrun,zmin,zmax,zperinc,velg);
///

* JSphGpuSingle.cpp
PeriZone->CheckPositionAll(BoundChanged,Posg,Velg);
///

*JPeriodPeriodicGpu_ker.h
void CheckPositionAll(unsigned n,unsigned pini,float3 *pos,bool xrun,float xmin,float xmax,tfloat3 xperinc,bool yrun,float ymin,float ymax,tfloat3 yperinc,bool zrun,float zmin,float zmax,tfloat3 zperinc,float3 *vel);
///

* JPeriodPeriodicGpu_ker.cu
template __global__ void KerCheckPositionAll(unsigned n,unsigned pini,float3 *pos,float xmin,float xmax,float3 xperinc,float ymin,float ymax,float3 yperinc,float zmin,float zmax,float3 zperinc, float3 *vel){
const unsigned p=blockIdx.y*gridDim.x*blockDim.x + blockIdx.x*blockDim.x + threadIdx.x; //-Number of particle.
if(p=xmax){ rpos=make_float3(rpos.x+xperinc.x,rpos.y+xperinc.y,rpos.z+xperinc.z); rvel=make_float3(0,0,0); modif=true; }
}
if(yrun){
if(rpos.y=ymax){ rpos=make_float3(rpos.x+yperinc.x,rpos.y+yperinc.y,rpos.z+yperinc.z); rvel=make_float3(0,0,0); modif=true; }
}
if(zrun){
if(rpos.z=zmax){ rpos=make_float3(rpos.x+zperinc.x,rpos.y+zperinc.y,rpos.z+zperinc.z); rvel=make_float3(0,0,0); modif=true; }
}
if(modif)pos[p2]=rpos;
if(modif)vel[p2]=rvel;
}}
///

void CheckPositionAll(unsigned n,unsigned pini,float3 *pos,bool xrun,float xmin,float xmax,tfloat3 xperinc,bool yrun,float ymin,float ymax,tfloat3 yperinc,bool zrun,float zmin,float zmax,tfloat3 zperinc,float3 *vel)
{
if(n){
dim3 sgrid=GetGridSize(n,PERIODICBSIZE);
if(xrun){ const bool runx=true;
if(yrun){ const bool runy=true;
if(zrun)KerCheckPositionAll <<<sgrid,PERIODICBSIZE>>> (n,pini,pos,xmin,xmax,Float3(xperinc),ymin,ymax,Float3(yperinc),zmin,zmax,Float3(zperinc),vel);
else KerCheckPositionAll <<<sgrid,PERIODICBSIZE>>> (n,pini,pos,xmin,xmax,Float3(xperinc),ymin,ymax,Float3(yperinc),zmin,zmax,Float3(zperinc),vel);
}
else{ const bool runy=false;
if(zrun)KerCheckPositionAll <<<sgrid,PERIODICBSIZE>>> (n,pini,pos,xmin,xmax,Float3(xperinc),ymin,ymax,Float3(yperinc),zmin,zmax,Float3(zperinc),vel);
else KerCheckPositionAll <<<sgrid,PERIODICBSIZE>>> (n,pini,pos,xmin,xmax,Float3(xperinc),ymin,ymax,Float3(yperinc),zmin,zmax,Float3(zperinc),vel);
}
}
else{ const bool runx=false;
if(yrun){ const bool runy=true;
if(zrun)KerCheckPositionAll <<<sgrid,PERIODICBSIZE>>> (n,pini,pos,xmin,xmax,Float3(xperinc),ymin,ymax,Float3(yperinc),zmin,zmax,Float3(zperinc),vel);
else KerCheckPositionAll <<<sgrid,PERIODICBSIZE>>> (n,pini,pos,xmin,xmax,Float3(xperinc),ymin,ymax,Float3(yperinc),zmin,zmax,Float3(zperinc),vel);
}
else{ const bool runy=false;
if(zrun)KerCheckPositionAll <<<sgrid,PERIODICBSIZE>>> (n,pini,pos,xmin,xmax,Float3(xperinc),ymin,ymax,Float3(yperinc),zmin,zmax,Float3(zperinc),vel);
else KerCheckPositionAll <<<sgrid,PERIODICBSIZE>>> (n,pini,pos,xmin,xmax,Float3(xperinc),ymin,ymax,Float3(yperinc),zmin,zmax,Float3(zperinc),vel);
} } }}
///

This to my understanding copies to device memory the velocity vector and might increase computational times, tough I'm not sure how much.

alexmees

What I REALLY wanted was a inlet velocity I could control parametrically in the XML file.

What I imagined was something to enhance the JSph class definition with:

*JSph.h
/// Structure with inlet information v = v0 + c.(|r0 - r|)^b
typedef struct {
bool InletResetOption; ///reset velocities yes/no
tfloat3 InletVelCenter; /// [r0] inlet velocity center
tfloat3 InletVelBase; /// [v0] base velocity at inlet center
float InletVelExp; /// [b] velocity exponent
float InletVelCoef; /// [c] 0 = Constant velocity; Negative = Maximum at the center; Positive = Minimum at the center
}Inlet;

and also:

void JSph::LoadCaseConfig(){
...
if(eparms.Exists("InletReset")&&eparms.GetValueInt("InletReset",true,0)==1)
{
InletReset.InletResetOption = true;
InletReset.InletVelBase = TFloat3(0);
if(eparms.Exists("InletVelBaseX"))
{ InletReset.InletVelBase.x = eparms.GetValueFloat("InletVelBaseX",true,0); }
if(eparms.Exists("InletVelBaseY"))
{ InletReset.InletVelBase.y = eparms.GetValueFloat("InletVelBaseY",true,0); }
if(eparms.Exists("InletVelBaseZ"))
{ InletReset.InletVelBase.z = eparms.GetValueFloat("InletVelBaseZ",true,0); }
if(eparms.Exists("InletVelCenterX"))
{ InletReset.InletVelCenter.x = eparms.GetValueFloat("InletVelCenterX",true,0); }
if(eparms.Exists("InletVelCenterY"))
{ InletReset.InletVelCenter.y = eparms.GetValueFloat("InletVelCenterY",true,0); }
if(eparms.Exists("InletVelCenterZ"))
{ InletReset.InletVelCenter.z = eparms.GetValueFloat("InletVelCenterZ",true,0); }
if(eparms.Exists("InletVelCoef"))
{ InletReset.InletVelCoef= eparms.GetValueFloat("InletVelCoef",true,0); }
if(eparms.Exists("InletVelExp"))
{ InletReset.InletVelExp= eparms.GetValueFloat("InletVelExp",true,1); } }

This way I could change the velocity of incoming particles according to a set o rules: inlet center + particle position and velocity equation.
I did not implemented this and do not know if it would work.

alexmees

What I REALLY REALLY wanted was to create particles at the inlet every time step with predefined velocities, but for that I have no ideas...