|
Numerical Simulations of Transitional Flow
in an Arteriovenous Graft
|
Sang-Wook Lee
Department of Mechanical & Industrial Engineering
University of Illinois, Chicago, IL
Francis Loth
Department of Mechanical & Industrial Engineering
University of Illinois, Chicago, IL
Paul F. Fischer
Mathematics and Computer Science Division
Argonne National Laboratory, Argonne, IL
Hisham Bassiouny
Department of Surgery
University of Chicago, Chicago, IL
|
 |
|
Spectral Element Mesh Generation
Our spectral element algorithms rely upon hex-based meshes for the efficiency and the accuracy.
In addition to orthogonality and aspect ratio considerations, the desirable properties of a spectral
element method mesh include a smooth description of the surface. This is particularly important for
computing quantities such as wall shear stress, which are very sensitive to geometry roughness.
Hence, we first reconstruct a smooth surface based on CT/MRI image of plastic cast using low pass filter to
remove pixilation effects.
Final high accuracy meshes are developed based on isotemperature surfaces from
the resulting solutions of a sequence of conduction (Poisson) problems in which two ends of the bifurcation
are set to constant temperature, while the side walls and the third end are insulated.
Conduction (Poisson) solution for mesh generation
Comparison of mesh quality
Spectral element mesh of AV graft(K=2640)
more detail of meshing algorithm...
.
Pulsatile flow based on in vivo flow waveform
A computational fluid dynamics study was conducted by direct numerical
simulation to elucidate three-dimensional complex flow structure with realistic pulsatile flow and geometry
conditions based on animal study. The unsteady incompressible Navier-Stokes equations were solved using the spectral element method.
The spectral element method is a high-order weighted residual technique and is ideally suited to flows at transitional
Reynolds numbers as it results in minimal numerical dissipation and dispersion.
Newtonian fluid behavior and rigid walls were assumed. Flow simulations were conducted on the NSF Pittsburgh
Supercomputing Center TCS1 with 256 plus 1024 parallel processors (P). The simulations were initiated at Re << 1
and the Reynolds number was ramped up to the diastolic value by exponentially decreasing the viscosity.
After initialization, the flow problem was simulated for six cardiac cycles with a polynomial degree of basis
function N = 12. The volumetric flow rates measured in DVS and PVS during the animal study were phase-averaged for
12 heartbeat cycles. These flow waveforms were imposed as velocity boundary conditions with a Womersley solution
which is an analytic solution of laminar fully developed pulsatile flow on both inlet boundaries.
The minimum and maximum Reynolds numbers entering the graft, based on averaged velocities
(over the cross-section) and graft diameter were 875 and 1235, respectively.
Computational domain of venous anastomosis of AV graft
Flow waveform for inlet boundary condition
Vector plots in midplane
Axial velocity distribution at four axial positions
Coherent vortical structures (Movie)
.
Importance of flow division on the incipience of transion to turbulent flow
The steady state studies were done to quantify the effect that flow division
had on transition to turbulence. Four Reynolds numbers (based on graft inlet) ranging from 800 to 1400 were
simulated under three different flow division conditions between the PVS and DVS. Numerical simulations demonstrated that various flow division conditions
significantly alter the transitional nature of flow in an arteriovenous graft despite having low Reynolds number
compared with critical Reynolds number of 2300. High velocity and pressure fluctuations were observed for the
PVS:DVS=70:30 case and absent from the 100:0 case at Reynolds number of 1200. Coherent structures were parallel
to the axial direction and highly organized under the 100:0 flow division. Under the 70:30 flow division, these
vortical structures broke down to a smaller scale generating transverse vortices. The 70:30
case had a less favorable pressure gradient developed in PVS compared with the 100:0 case. Transition to turbulent
flow was observed for Reynolds numbers greater than 1000 with the 85:15 flow division. LDA measurements confirmed
these numerical findings with good agreement with respect to time average and r.m.s. of the axial velocity.
Coherent structures with three different flow division condtions
Transverse vorticity distribution (Movie)
Coherent structures for different flow division (Movie)
.
Turbulence in an arterial anastomosis
The flow in the arterial anastomosis (AA) was simulated and then fed into
the PTFE graft in order to investigate flow physics upstream of the venous anastomosis. Results indicated that
turbulence is readily generated at the arterial anastomosis due to the samll diameter of artery.
However, the turbulence generated in arterial anastomosis is not sustained through the 20 cm PTFE graft segment
at RePTFE = 1200.
Transverse vorticity generated in arterial anastomosis (Movie)
Transverse vorticity convection through PTFE graft (Movie)
.
|
|