Sei sulla pagina 1di 12

Pergamon

.I. B&mechlmics, Vol. 11, No. 4, pp. 391~402,1994

copylisht 1994 0 &&a s&Ice Lid Printed Orcar in B&in. Allrightl racrvcd
an-9290/w %.00+.00

NUMERICAL SIMULATION OF UNSTEADY LAMINAR FLOW THROUGH A TILTING DISK HEART VALVE: PREDICTION OF VORTEX SHEDDING
Z. J. HUANG,* C. L. MERKLE,t S. ABDALLAH~ and J. M. TARBELL*~~
*Department of Chemical Engineering, tMechanica1 Engineering, $Tbe Applied Research Laboratory and The Bioengineering Program, The Pennsylvania State University, University Park, PA 16802, U.S.A. Ata&act--Heart valves induce flow disturbances which play a role in blood cell activation and damage, but questions of the magnitude and spatial distribution of fluid stresses (wall shear stress and turbulent stress) cannot be readily addressed with current experimental techniques. TImefore, a numerical simulation procedure for flow through artificial heart valves is presented. The algorithm employed is based on the Navier-Stokes equations in generalized curvilinear coordinates with artificial compressibility for coupling of velocity and pressure. The algorithm applies a 5nite-diI%ence technique on a body-conforming composite grid around the. heart valve disk on which the numerical simulations are performed. Steady laminar flow over a backward-facing step and unsteady laminar 5ow inside a square driven cavity are computed to validate the algorithm. Two-dimensional, time-accurate simulation of flow through a tilting disk valve with a steady upstream Reynolds number as high as 1000 reveals the complex behavior of vortex shedding. By scaling the results at the Reynolds number of 1000 to peak systolic 5ow conditions, the maximum value of shear stress on the valve disk is estimated to be 770 dyn cme2. The apparentReynolds stress associated with vortex shedding is estimated to be as high as 3900 dyn cm-* with a vortex shedding frequency of about 26Hz. The apparent Reynolds stress value is of similar magnitude as reported in experiments but would not be expected to damage blood cells because the spatial scales associated with vortex shedding are much larger than blood cell dimensions.

INTRODUCTION

Some of the majq problems with heart valves and artificial hearts are haemolysis, thrombus formation and tissue overgrowth near the heart valves (Tarbell et al., 1989; Yoganathan et al., 1979). These problems are closely related to the flow pattern in the vicinity of the heart valves, where the occluder elements (valve disk) produce a high degree of disturbance (turbulence) in the blood flow, causing separations, recirculations and elevated fluid stresses, all in close proximity. High shear stress and turbulence (Reynolds stress) may damage the blood cells and the endothelial walls, thus, triggering the formation of thrombus. Accompanying low shear stress, stagnation, enhances the formation of aggregates in the blood flow and Quantitative overgrowth. facilitates tissue experimental studies of heart valve flow have been conducted for nearly two decades (Baldwin et al., 1989, 1990, Chandran et al., 1989; Tarbell et al., 1986, 1989; Woo and Yoganathan, 1985; Yoganathan et al., 1979). However, due to the limitations of experimental apparatus and the structure of the artificial heart valve, it is very difficult to experimentally measure the shear rate and other flow parameters in the immediate vicinity of the valve disk. On the other hand, detailed understanding of flow characteristics at virtually every point of the flow field
Received in final /on 23 August 1993. 11 Author to whoi correspondence should be addressed at Deuartment of Chemical Entieerinn, 158 Fenske L&oratory, Pennsylvania State U&ersity&iversity Park, PA 168024400, U.S.A.
M 27:4-0

can be attained by numerical simulation after appropriate physical modeling of the actual flow problem. Consequently, several different numerical simulation procedures have been attempted for the solution of the flow field in an artificial heart or heart valve. Stevenson et al. (1985) simulated steady, turbulent flow through trileaflet aortic heart valves in a straight tube using the assumption of an axisymmetric (twodimensional) geometry. The stream-function-vorticity approach was applied to couple the continuity and Navier-Stokes equation with the u-s model of turbulence in the axisymmetric form. The peak turbulent shear stresses and pressure drops were calculated from the simulated Aow field and were in good agreement with the experimental measurements. Rogers et al. (1989) developed a procedure to simulate threedimensional laminar flow in the Penn State artificial heart. Using artificial compressibility to couple the continuity and Navier-Stokes equations and upwinddiierencing for dissipation, the governing equations were solved using line-by-line iterations. Fairly good qualitative agreement of incoming-particle tracings was achieved with the experimental observations. Peskin and &Queen (1980,1983,1989) have conducted impressive simulations of laminar blood flow in the natural human heart both with and without the valve prostheses. In their studies, the Poisson pressure equation was used to couple the continuity and Navier-Stokes equations, and they were solved together with a model of elastic fibers of the heart wall. Several types of valves were studied and the resulting flow patterns were similar to those observed in experiments. However, the simulations werC conducted at

391

392

Z. J. HUANGet al. and pressure fields as follows:

low Reynolds numbers, due to the very limited resolution of the grid near the boundary, which was restricted by the large amount of CPU memory and computer time required. One shortcoming of these previous models is that the details of flow near the boundaries such as the valve disk/leaflet and tube walls were not determined accurately due to lack of grid resolution. Although the main bulk flow may be of great interest, the interao tion with the valve disk greatly influences the flow pattern in the whole system, as observed in our studies. The complex flow phenomenon of vortex shedding, which is known to arise in flow fields behind blunt bodies at high Reynolds numbers (Hoerner, 1958) and has been postulated to exist in the outflow tract of heart valves (Reif et al., 1990), has not been predicted by previous numerical studies of heart valves. The heart valve modeled in this study (BjGrkShiley convexo-concave/monostrut tilting disk valve) is comprised of a tilting valve disk and a circular ring with superstructures (struts) for restraining the movement and maximum tilting angle of the valve disk. Since blood flow in the vicinity of the heart valve involves separation and recirculation, the gradients of fluid velocity and stress are highest near the tilting valve disk. Thus, a body-conforming curvilinear grid must be generated around the valve disk in order to evaluate effectively the detailed characteristics of the blood flow, andalarge number of grid points have to be used in order to generate unskewed body-conforming grids due to the complex geometry of the heart valve. Using a (developed) numerical scheme based on the artificial compressibility method for solving the Navier-Stokes and continuity equations with a bodyconforming grid of high resolution, the following specific questions will be addressed in this study: What are the highest wall shear stress values in the vicinity of the heart valve and where do they occur? Does flow through an open valve induce vortex shedding at high Reynolds number? What are the highest fluctuating velocities (apparent stress) induced by the valve and where do they arise? Are the fluid stresses induced by the valve in the range which might cause blood cell damage?
METHODS

~=p(v.o),
where g is the artificial compressibility coethcient and T is the pseudo-time. In the case of two-dimensional flow, after non-dimensionalization and coordinate transformation to generalized coordinates r=r(x,Y), 11=rl(x*yk (4)

the continuity equation and momentum equations can be written as a system of first-order equations: !$ -(!!!+!!)+(!$!+$)_8_ir, (5)

where fi is defined as the residual vector, fi is the sink term related to physical-time derivatives:

ii= 11
0

$
au

Lat 1

and a new set of vectors are defined as

where .I is the Jacobian of the transformation, J= a(5,tl) 8(x, Y) U and V are the contravariant u=Lu+r& velocities: v=?fXu+~yv.

For laminar Newtonian flows, the momentum fluxes due to molecular transfer & and] (the subscript v denotes viscous fluxes) are given by

For an incompressible fluid, the continuity equation and momentum equations are as follows: v*P=o, ,r;+,.v+ (1)

-VP-V.? ,

(2)

where 0 is the velocity vector, z is the viscous stress tensor, p is the pressure and p is the density. The artificial compressibility approach uses a pseudo-time derivative of the pressure added to the continuity equation in order to couple the velocity

-(vq vt)au -- - vII) au (vu -- t=;

J at+ J mwav+mwav --J at J

all

atl I

(6)

Prediction of vortex shedding where Re is the Reynolds number

393

Re,PUoLo -.
p and p are the density and viscosity of the fluid, respectively, and Lo and U. are the characteristic length and velocity which define the relationship between nondimensional variables and dimensional ones (denoted by the tilde over them)

pv

f p=-j&
2 f=LdU,

vo

Although our current scheme does not contain a turbulence model and can only describe laminar flows, unsteadiness in the flow can produce apparent Reynolds stresses which might be interpreted as turbulence. Therefore, we will also compute the apparent Reynolds shear stress puv, where u and u are the deviations of the axial and radial velocity from their time-averaged values (the overbar denotes a time average). The dimensionless apparent Reynolds shear stress is obtained by scaling pl(0 by pUz. The wall shear stress is calculated by three-point one-sided finite-differencing which is second-order accurate. In the preceding equations, the mat&s of the coordinate transformation were represented by equations of the form

and are evaluated from the following (Pulliam and Steger, 1978):

formulas

ay r,=g J-l, i
&=--&

ff=-E

ayJ-l,
i

ax i

J--l,

qy=-g

ax

J-.

For steady-state flows, the sink term fi in equation (5) disappears. In the time-accurate formulation for unsteady flows, the physical-time derivatives of the sink term fI are evaluated using a second-order, threepoint, backward-differencing formula

au

c~~~+~+c~u~+c_~u~-~

at

At

where the superscript n denotes the physical-time level at t = t [the current time is at the (n + 1)th level] and

cl-J=

with At=t+--t and At,=t-r-. When At,=At (i.e. equal spacing in physical time), equation (9) becomes the well-known formula au 3un+1 -4u+u-- (11) at 2At . The implementation of boundary conditions in our studies is accomplished by standard means with viscous methods being used for the wall boundary conditions and inviscid methods for inflow+utflow conditions. No slip conditions are imposed on the walls by setting the velocity components to zero and augmented by using the normal momentum equation to compute the wall pressure. Inflow and outflow boundary conditions are enforced by employing the method of characteristics (MOC) procedure (Merkle and Tsai, 1986). The introduction of an artificial compressibility term in the continuity equation causes the system to become hyperbolic so that the well-known MOC methods can be applied. Briefly, the number of boundary conditions is determined by the sign of the eigenvalues of the Jacobians of the convective fluxes. A positive (negative) eigenvalue indicates a wave propagating in the positive (negative) direction. Each wave propagating from the interior of the computational domain to the boundary needs a characteristic equation to extract information from the interior itself, while the rest of the information should come from outside the computational domain, in the form of a number of specified boundary variables. At the inflow (upstream) boundaries, there will always be an upstream propagating wave from inside the computational domain. The corresponding characteristic equation is constructed from the system of governing equations [for details, see Merkle and Tsai (1986)]. The remaining variables (corresponding to the downstream propagating waves) are specified as boundary conditions. For the present results, we have chosen to specify the velocity components at the inlet plane, according to the experimental measurements or theoretical calculation. At the outflow (downstream) boundaries, there is one wave propagating from outside the computational domain. Therefore, in our studies, the pressure is specified and all other variables are determined from the characteristic equations. The four-stage Runge-Kutta time-marching scheme has been used to solve the incompressible flow problem previously (Merkle and Tsai, 1986; Tsai, 1988). A local scalar stability analysis indicates that a fourth order artificial dissipation term must be added to the coupled system of equations (5) in order to preserve the stability of the scheme, especially for flow at high Reynolds number, when the effect of dissipation by the viscous term becomes less important. The modified first-order system of equations becomes

394

Z. J. HUANG al. et

where_&,is the fourth order artificial dissipation factor and R denotes the combined residual vector. The four-stage Runge-Kutta scheme, which is equivalent to a fourth-order temporal Taylor expansion, may then be expressed as

Q.._+~~.* ip*+p**,
Q
*m+l =pt__AZ&4cH,

(13)

where the superscript m denotes the pseudo-time marching step and the superscripts *, ** and +** denote the intermediate stages. The iterative procedure in pseudo-time is carried out until the deviations (Ao=6mm+i -@) drop below certain predefined criteria. For steady-state flows, this indicates that the steady-state solution has been obtained. For time-varying flows, the results can be regarded as the transient solution at the current physical-time level and the iterative procedure can be advanced to the next physical-time level. To transform the system of partial differential equations from continuous physical space to a set of linear/nonlinear algebraic equations in the discrete computational domain, the structured grid mesh for the finite-difference techniques must be generated. For the complex multiconnected geometry near a valve disk, it is necessary to divide the physical space into several simply connected zones and apply a generic grid generator in each zone. The grid generator used is an elliptic system (Thompson, 1985), which is based on the orthogonality of stream and potential lines, and modified as follows

(14) where P and Q are the control functions for the distribution of grid points, whose selection has proven to be more of an art than a science. After several mathematical manipulations, equation (14) can be rewritten in a more comprehensive form.

a2x
Q22 -$+gll~-%l,

azx

aq
azy

&+g(P$+Q$)=O,

922 ,rz+s11 -&y--2812 z+g(p~+Q~)=O,


aratl

azy

where

After the boundary points of each zone are assigned, equation (15) is solved using a successive overrelaxation @OR) scheme to generate the grid mesh. The initial guess for the grid mesh is created by using an algebraic transfinite solver (Thompson, 1985). As stated earlier, in some cases, it is necessary to use two sets of overlapping structured grid meshes in order to obtain a body-conforming unskewed smooth grid and ensure the stability of numerical schemes and resolution of the flow field. Thus, information must be passed between the two sets of grids. In the current study, a computing procedure has been established for linearly interpolating variables at the grid interfaces [modified from Snyder (197811. Upon obtaining a set of grid data, the grid mesh is automatically divided into triangular elements, aligned with the shorter diagonal line in each cell. Then a screening procedure is activated to determine the triangular element (in the other grid mesh), where every given grid point of the specified grid mesh is located. Afterwards, using a formula derived from Taylor expansion, linear interpolation coefficients for the given grid point with second-order accuracy are calculated, stored, and used later for the interpolation of data points between the two grid meshes. The major limitation of the overlapping-grid scheme is the CPU time needed for the screening procedure, which usually is proportional to the product of the number of grid points in the two grid meshes. In our implementation of the overlappinggrid scheme, certain optimization techniques (e.g. automatically dividing the physical space into many smaller searching subzones) are used to speed up the screening procedure. The CPU usage of our optimized screening procedure for the overlappinggrid scheme is proportional to only the number of grid points of one grid mesh and, thus, it is feasible to apply the overlapping-grid scheme in the large scale, especially when fine resolution of results is demanded. In order to validate our numerical simulation method for heart valves which produce flow patterns of separation and stagnation, the solution procedure described above was applied to a related test case-flow through a backward-facing step, for which several experimental works have been reported (Armaly et al., 1983; Denham and Patrick, 1974). In our studies, laminar flows through two different backward-facing steps with expansion ratios (ratio of downstream height to upstream height) of 1.50 and 1.94 were extensively explored. For each expansion ratio, several sets of uniform or slightly stretched rectangular grid meshes, with various lengths of the upstream and downstream sections and/or different distribution patterns for the grid points, were generated. Simulations were started using the grid meshes with the shortest lengths of the upstream (downstream) sections (140 by 30 grid, upstream length 2.0,

Prediction of vortex shedding

395

downstream length 5.0) and then carried out with the inlet (outlet) moving further away from the step until further increase of the upstream (downstream) section lengths (up to 600 by 80 grid, upstream length 5.0, downstream length 25.0) had no effect on the flow near the step. It was determined that as the Reynolds number of the flow increased (from 73,125,191 to 229), a longer downstream section length was needed to achieve results independent of length, while the minimum upstream section length was adequate at all Reynolds numbers. The time-accurate version of the algorithm developed was tested by application to flow inside a square cavity, driven by a moving solid upper boundary (lid), and then compared with the computational results of Soh and Goodrich (1988). In our calculation, an even-spaced rectangular grid (50 x 50) was used for the square driven cavity in both the case of an impulsively started lid and a periodically oscillating lid at a Reynolds number of 400, based on the height of the cavity. Actual measurements of the dimensions of a BjiirkShiley convexo-concave tilting disk valve were used to generate the grid mesh employed in our simulation. It should be noted that the disk geometries of the convexo-concave and monostrut valves are identical. Figure 1 shows the two-dimensional composite grid mesh around the tilting-disk valve mounted in a straight tube at its fully opened position (57.5). Although a straight tube is not an anatomically correct geometry to simulate valve replacement in the natural heart, it is a good approximation to the actual geometry of the aortic outflow tract in the Penn State

artificial heart [see Baldwin et al. (1988, 1989, 1990)]. In addition, many fluid mechanical tests of heart valves have been carried out in straight tubes [see Yoganathan et al. (1979) and Woo and Yoganathan (1985)]. Figure 2 is an expanded view for details of the grid at the trailing edge of the valve disk. The bodyconforming grid is used in order to get accurate results in the vicinity of valve disk, where gradients of many important variables (such as velocity, pressure, etc.) are very high. As in our test case of flow over a backward-facing step, grid meshes with various lengths for the upstream (downstream) sections were tested in order to eliminate the impact of upstream (downstream) boundaries on flow near the valve disk. Different patterns of grid point distribution were also explored to ensure the resolution of the flow field. For the results presented below, the composite grid mesh contains 92,672 grid points (771 x 121, excluding the virtual points) with lengths for the upstream and downstream sections being 5 and 20 valve diameters, respectively. Time-accurate simulations have been conducted for unsteady blood flow in the vicinity of the valve disk at open-tube Reynolds numbers of 10,50, 100,200,500 and 1000, with the upstream inlet flow condition set as a steady parabolic velocity profile. This velocity profile was used in the upstream tube because the Reynolds numbers were in the laminar range and parabolic profiles would be expected. Although only flows with steady inlet conditions were simulated, they provide a good approximation of the systolic phase of physiological flows during which the valve is fully open.

0.80

1. Display of the in the of the X Y-axis values are dimensionless based on to high

at the of 2.38

of 57.5.

396

Z. J. HUANGet al.

Fig. 2. Details of the gridmesh near the trailing edge of the valve disk. Part of Fig. 1 has been enlarged 2.5 times to provide a clear view of the grid point distribution near the valve disk surfaces, where the gradients of velocity and other flow properties are the highest.

12.0

*
10.0

o* + *
P t

0,
4.0

+ * *

2.0

0.0
0.0

1..

I.

I..

.I.

50.0

loo.0

150.0

200.0

250.0

300.

Reynolds Number 0 --Present * ---

Dye Trace

+_--

Laser Anemometer

Fig. 3. Comparison of calculated and experimental data for a backward-facing step. The dimensionless reattachment length (based on the inlet height) for the primary separation zone is plotted against the Reynolds number (based on the inlet height and bulk flow velocity). The experimental data are from Denham and Patrick (1974). Good agreement between calculations and experiments provides validation of the computational method for separated flows which occur downstream of heart valves.

Prediction of vortex shedding ItRWLTs

391

The computational method we developed to study flow through heart valves provides accurate predictions of the separation reattachment length for flow over a backward-facing step (see Fig. 3) and drag forces on the lid of a square cavity driven either impulsively or periodically and, therefore, the pre-

dieted flow fields for the heart valve model should be accurate. Time-accurate simulations of flow through the Bj&k-Shiley valve show that, at Reynolds numbers above SO,the flow becomes separated and forms severa1 recirculation zones (vortex street) just downstream of the valve disk (see Fig. 4). At low Reynolds numbers (up to !OO), the instantaneous solutions at

1.00

0.50

0.00
-2.00 -1.50 -1.00 -0.50 0.00 0.50 1.00

1.50

2.00

2.50

3.00
A: O.OOCKtO B: 0.10000

1.00

c:O.2OWfl D: 0.30000 E 0.4OooO

0.50

0.00 3.00 3.50 4.00 4.50 5.00 5.50 6.00 6.50 7.00 7.50 8.00

F: 050000 G: 0.6OOoO H: 0.7OoOO K: 0.8oooo

1.00

L: 0.9mo M: 1.m

0.50
N: 0.55233 p: 1.22562

0.00 8.00 8.50 9.00 9.50 10.00 10.50 11.00 11.50 12.00 12.50 13.00

Q: l.loooO R: 1.2OOW S: -0.08787

1.00

0.50

0.00

.._

_.._

....

_...

..._

....

....

....

,...

....

13.00

13.50

14.00

14.50

15.00

15.50

16.00

16.50

17.00

17.30

18.00

Fig. 4. A snapshot of the vortex street downstream of the valve disk. Stream function contours are displayed for an opemtube Reynolds number of loo0 at a dimensionless physical time of 61.50 (based on the valve disk diameter and inlet bulk flow velocity). Letters on contours are defmed by the 1eRend the right on and represent values of the stream function. Vortices, which at&t the shear stress distribution near the valve and fluid mechanical interaction with blood, are observed for all flows with Reynolds numbers above so.

398

Z. J. HIJANG

et al.

successive

t=M.So
u.w
I
1111-11-11-11~11

-1.00

-0.50

0.00

0.50

1.00

1.50

2.00

2.50

3.00

350

4.00

450

5.00

1.00 OS0 0.00


-1.00 -0.50 0.00 050

MN.75

1.00

l.SO

2.00

250

3.00

3.50

4.00

4.50

5.00

1.00 0.50
lJ.w

t=62.00 iI

II-

1'~"1~"'1"~'1""l""lm'l~~l-~l~~~~l~~~~~~~~~l~~~~l~~~~~

-1.00 -0.50

0.00

0.50

1.00

l.SO

2.00

250

3.00

3.50

4.00

4.50

5.00

1.00 0.50 0.00

t=42.2S

-1.00 -0.50

0.00

0.50

1.00

1.50

2.00

2.50

3.00

3.50

4.00

4.50

5.00

1.00 0.50 0.00


-1.00 -0.50 0.00 0.50

t=62.S0

1.00

l.SO

200

250.

3.00

350

4.00

4.So

5.00

Fig. 5. Display of vortex shedding from the trailing edge of the valve disk. Stream function contours are displayed for an open tube.Reynolds number of 1000 at different physical times (through a shedding cycle) as defined by ihe legend on the right. Stream function contour values are the same as in Fig. 4 (increment of 0.10) except for the contour attached to the valve disk surface whose value varies with different physical times. Vortex shedding, which produces flow unsteadiness and apparentReynolds stresses, is observed for all flows with Reynolds number above 200.

Prediction of vortex shedding stage between the steady and unsteady states, but no attempt was made to determine the critical transition Reynolds number. In addition to the vortex street on the upper and lower walls shown in Fig. 4, periodic vortex-shedding

399

a transitional

from the edge of the valve disk was also observed (see Fig. 5) at Reynolds numbers of 200,500 and 1000, with the vortex street extending further downstream on both walls as the Reynolds number increased. The vortex which sheds from the trailing edge of the valve disk quickly dissipates in the flow field because of the presence of the upper and lower walls. This in turn causes the formation of moving vortex streets on both walls, which have vortex forming frequencies almost exactly half of the shedding frequency from the trailing edge. As vortices move downstream along the walls, they become weaker and weaker, and finally dissipate. The maximum shear stress occurs at the leading edge of the valve disk, and its magnitude displays almost perfectly periodic behavior (see Fig. 6) with the amplitude and mean being elevated with rising Reynolds number (peak (mean) dimensionless maximum shear rate values: 377 (373) at Re=200, 480 (460) at Re = 500,580 (500) at Re = 1000). By extrapolating simulated shear stress results to peak systolic flow conditions (bulk flow velocity of 61.07 ems-, a disk diameter of 2.38 cm and a fluid viscosity of 3.5 cp), it is estimated that the maximum shear stress on the valve disk is about 770 dyn cm-. The maximum value of the dimensionless apparent Reynolds shear stress increases from 0.220 at Re= 200-0.657 at Re= 500-0.874 at Re= 1000, and
its location is very close to the trailing edge of valve

disk. Assuming a dimensionless value of 1.0 at the peak systolic flow Reynolds number of 4360, we estimate a maximum apparent Reynolds shear stress of about 3900 dyn cmm2 under the actual flow conditions. The transition from one steady oscillatory flow pattern to another after the upstream flow rate is stepped up requires about four vortex shedding cycles (see Fig. 7) to achieve a new steady oscillation pattern which, as will be pointed out in the discussion, is quite rapid.
DISCUSSION

The computational method and grid scheme presented in this paper were developed so that the flow

field near a heart valve disk, which is difficult to access experimentally, could be examined in detail. To validate the computational method for separated flows, our simulation results for the backward-facing step at several Reynolds numbers were compared with the experimental data of Denham and Patrick (1974), and the separation reattachment lengths were found to be in very good agreement with the experimental data (see Fig. 3). The calculated streamline patterns of the square cavity with the impulsively started lid and the periodically oscillating lid were virtually identical with those of Soh and Goodrich (1988) who used a 40 x 40 rectangular grid. A comparison of the drag forces on the periodically oscillating lid computed by the two methods was excellent. Although only flows with steady inlet conditions were simulated, previous studies indicate that steady

300.0 SO.0 520

.., 54.0

_ 56.0

. .., 58.0

.., 60.0

. . . . 62.0

I 64.0

.., 66.0

. . . . . . . . 68.0 70.0

Fig. 6. Shear rate at the leading edge of the valve disk as a function of time at Reynolds numbers of 200,500 and 1000 (lower, middle and upper curves, respectively). The dimensionless shear rate is based on the diameter of the valve disk and the inlet bulk flow velocity; the dimensionless physical time is as defined in Fig. 4. The onset of unsteadiness associated with vortex shedding at Reynolds number of 200 could be mistakenly interpreted as turbulence, but is unlikely to be damaging to blood because of the large scale of the motion (see Fig. 5). It is obvious that the intensity of the flow oscillation and apparent turbulence increase with Reynolds number.
BM 27:4-C

Z. J. HUANGet al.

0.0

20

4.0

6.0

8.0

10.0

120

DimensionlessTime Fig. 7. Shear rate at the leading edge of the valve disk as a function of time after the Reynolds number was stepped up, from 500 to 1000. The scaling is the same as in Fig. 6. The transition to a new flow state occurs rapidly and suggests that vortex shedding will occur during pulsatile flow.

flows provide a good approximation of the systolic phase of physiological flows during which the valve is fully open and the fluid stresses and associated blood damage potential are highest. For example, Chandran et al. (1983) have shown that, for a Bjiirk-Shiley convexo+zoncave valve with the disc tilting toward the lateral walls of the aorta (normal orientation), the velocity and turbulent stress profiles for steady flow at the peak systolic flow rate have the same basic shape as profiles obtained under pulsatile flow conditions at the peak of systole [see Chandran et al. (1983, Figs 8 and 9)]. The peak velocities in the major and minor orifice jets for the steady and pulsatile flow conditions agree to within 25%. The maximum turbulent normal stresses were 1250 dyncm- for steady flow and 1400 dyncm-* for pulsatile flow. Khalighi et al. (1983) studied the same valves in steady flow in an aortic arch model, while Chandran et al. (1985) conducted a similar investigation using pulsatile flow. The velocity profiles in the two studies display maxima in the major and minor orifice jets which agree to within 25% [compare Fig. 3 of Khalighi et al. (1983) with Fig. 3 of Chandran et al. (1985)]. They did not report turbulence measurements. These previous studies suggest that steady inlet flow simulations provide a reasonable, although not perfect, approximation to physiological pulsatile flow during systole. The most interesting prediction of the computer simulations is the onset of vortex shedding for Reynolds number above 200. The mechanism of vortex shedding was studied by Hoerner (1958) and is considered to be the result of the inertial forces overwhelming the viscous forces in flow fields behind blunt bodies at high Reynolds numbers. The intensity of vortex shedding is usually measured in terms of

the Strouhal number:

S=f$
where f is the vortex-shedding frequency, h is the dimension (height or thickness) of the body producing the vortex, and V is the bulk velocity. In our studies, h was taken as the diameter of the valve disk. For the typical shedding cycle associated with the vortex at the trailing edge of the valve disk, the Strouhal number does not change much with Reynolds number (1.10 at Re=200, 1.14 at Re=500 and 1.02 at Re= 1000). The Strouhal numbers for the vortex streets on the upper and lower walls are almost exactly half of the above figures (about 0.5). Osswald et al. (1994) simulated flow over a backward-facing step at a Reynolds number of 2000 and observed a moving vortex street on the wall as vortices kept forming just downstream of the step. The Strouhal number was calculated as 0.38 based on the height of the channel. The vortex streets on walls that we have observed show similar patterns and Strouhal numbers of 0.55,0.57 and 0.51 based on the channel height for Reynolds numbers of 200,500 and 1000, respectively. According to Hoerner (1958), the Strouhal number will not change much at high Reynolds numbers, and is about 0.4 for unbounded flow around blunt bodies. However, in his studies, the projected height or thickness perpendicular to the flow direction was used as h. If we adapt the same definition of h, the Strouhal numbers for vortex shedding from the trailing edge in our studies will be 0.59,0.61 and 0.55 for Reynolds numbers of 200,500 and 1000, respectively. These values are still somewhat higher than the ones suggested by Horner for blunt bodies, but the deviation is

Prediction

of vortex shedding

401

believed to be due to the presence of the bounding walls. Assuming a value of 1.0 for the Strouhal number based on the valve disk diameter [since the Strouhal number is relatively constant at high Reynolds numbers, according to Hoerner (1958)], a tube Reynolds number of 4360 (typical of peak systole), a disk (and tube) diameter of 2.38 cm, a fluid viscosity of 3.5 cp and a fluid density of 1.05 gmcm- j, the frequency of vortex shedding is estimated to be 26 Hz. Vortex shedding has been noted in a two-dimensional monoleaflet valve model (Reif et al., 1983), and has been postulated to play a role in the flutteer of bileaflet valves in the range of 2-200 Hz (Reif et al., 1990). It has also been clearly observed downstream of a 90% stenosis experiencing physiological oscillatory flow using spectral decomposition of velocity measurements at a frequency of about 25 Hz (Lieber, 1990). Thus, there is reason to believe that the vortex shedding observed in our simulations will actually occur under phsiological flow conditions. The maximum wall shear stress predicted for a peak Reynolds number of 4360 is 770 dyncm- and is located at the leading edge of the valve disk. Because of the difficulty in making wall shear measurements at any position along the valve disk, there are no experimental data available in the literature for comparison. However, it should be noted that shear stresses of this magnitude are unlikely to cause haemolysis, although they may be of sufficient magnitude to cause platelet activation (Blackshear and Blackshear, 1987). The apparent Reynolds shear stress of 3900 dyn cm- predicted for the peak Reynolds number of 4360 is surprisingly close to experimental measurements of the maximum Reynolds shear stress (3400 dyn cmm2) downstream of a Bjiirk-Shiley tilting disk valve reported by Woo and Yoganathan (1985). However, it must be emphasized that the apparent Reynolds shear stress we have calculated is not due to turbulence since a turbulence model was not incorported in our simulation. The apparent stress is due entirely to large scale, laminar vortex shedding. This observation raises the interesting possibility that reported values of Reynolds stresses downstream of heart valves contain contributions from large-scale organized structures (i.e. vortex shedding) as well as fine-scale turbulent motions. That is, remnants of the large-scale vortices which shed at lower Reynolds stress. Thus, decomposition of the apparent Reynolds stress into its organized and turbulent contributions becomes important because the larger scales associated with the organized motion are much larger than cellular dimension and, therefore, less likely to be damaging to blood cells than the smaller scales of the turbulent motion (Papoutsakis, 1991). Methods for performing the decomposition and more detailed discussion of organized structures in turbulent physiological flows have been presented by Lieber (1990). Two-dimensional, time-accurate simulations for a fully open valve with a steady upstream flow condi-

tion were performed in our current studies. However, the actual physiological flow through a Bjiirk-Shiley tilting disk valve is three-dimensional with a timevarying upstream flow condition. How likely is it that the phenomena of vortex shedding which we have observed will persist in the physiological flow? First, let us consider the time-varying nature of the inlet flow. Figure 7 shows the development over time of the shear rate at the leading edge of the disk as the inlet Reynolds number is stepped up from 500 to 1000. It is apparent that the flow takes about three to four cycles to reach a new oscillatory pattern after the sudden change of inlet condition. Realizing that the valve will typically be in the fully open position for more than 90% of the 400 ms systolic phase; the inlet blood flow maintains a near-constant flow rate for about 300 ms; and that the vortex-shedding frequency is estimated to be about 25 Hz, there would appear to be ample time for the shedding phenomena to develop. For the actual three-dimensional flow with a pulsatile flow rate, the vortex shedding may be suppressed until higher Reynolds number than the two-dimensional case in a manner similar to the difference observed between vortex shedding from cylinders and spheres. However, the very low Reynolds number (between 100 and 200) of starting vortex shedding in the two-dimensional case suggests that similar unsteadiness would be very likely for the actual three-dimensional case with a Reynolds number over 4000. The frequency of vortex shedding and the interactions with the pulsatile flow could have important implications for blood cell damage. The advanced techniques of hot-wire or multicomponent Laser Doppler Anemometry (LDA) are needed to confirm this ordered unsteadiness experimentally.
work was supported by NASA Joint Research Interchange NCA-291 and in part by the Pittsburgh Supercomputing Center Grant P41-RR06009 from the NIH National Center for Research Resources.
Acknowledgements-This

REFERENCES

Armaly, B. F., Durst, F., Pereira, J. C. F. and Schonung, B. (1983) Experimental and theoretical investigation of backward-facing step flow. J. Fluid Mech. 127.473-496. Baldwin, J. Ty, Deutsch, S., Geselowitz, D. B: and Tarbell, J. M. (1990) Estimation of Reynolds stresses within the Penn state left ventricular assist device. ASAIO Trans. 36, M274-M278. Baldwin, J. T., Tarbell, J. M., Deutsch, S. and Geselowitz, D. B. (1989) Mean flow velocity patterns within a ventricular assist device. ASAIO Trans. 35, 429-433. Baldwin, J. T., Tarbell, J. M., Deutsch, S., Geselowitz D. B. and Rosenberg, G. (1988) Hot-film wall shear probe measurements inside a ventricular assist device. J. biomech. Engng 110, 326333. Blackshear, K. B. and Blackshear, G. L. (1987) Mechanical hemolysis. In Handbook of Bioengineering (Edited by Skalak, R. and Chien, S.). McGraw-Hill, New York. Chandran, K. B., Cabell, G. N., Khalighi, B. and Chen. C.-J.

402

Z. J. HUANGet al. difference simulations of three-dimensional flow. AIAA Paper 78-10. Reif, T. H. and Huffstutler Jr, M. C. (1983) Design considerations for the omniscience pivoting disc cardiac valve prosthesis. ht. J. Artif Organs 6, 131-138. Reif, T. H., Schulte, T. J. and Hwang, N. H. C. (1990) Estimation of the rotational undamped natural frequency of bileaflet cardiac valve prostheses. J. biomech. Engng
1X2,327-332.

(1983) Laser anemometry measurements of pulsatile flow past aortic valve prostheses. J. Biomechanics 16, 865-873. Chandran, K. B., Khalighi, B. and Chen, C.-J (1985) Experimental study of physiological pulsatile flow past valve prostheses in a model of human aorta-II. Tilting disc valves and the effect of orientation. .I. Biomechanics 18, 773-780. Chandran, K. B., Schoephoerster, R. and Dellsperger, K. C. (1989) Effect of prosthetic mitral valve geometry and orientation on flow dynamics in a model human left ventricle. .I. Biomechanics 22, 51-65. Denham, M. K. and Patrick, M. A. (1974) Laminar flow over a Downstream-facing step in a two-dimensional flow channel. Trans. Inst. them. Enars 52. 361-367. Hoerner, S. F. (1958) Fluid-Dynakc D&g. Published by the Author, New Jersey. Khalighi, B., Chandran, K. B. and Chen, C.-J. (1983) Steady flow development past valve prostheses in a model human aorta-II. Tilting disc valves. J. Biomeehanics 16, 1013-1018. Lieber, B. B. (1990) The decomposition of apparent stresses in disturbed pulsatile flow in the presence of large scale organized structures. J. Biomechanics 23, 1047-1060. Merkle, C. L. and Tsai, P. Y-L. (1986) Application of Runge-Kutta schemes to incompressible flow. AIAA 24th Aerospace Science Meeting, January 1986, Reno, NV, AIAA Paper 86-0553. Osswald, G. A., Ghia, K. N. and Ghia, U. (1994) Direct simulation of unsteady two-dimensional incompressible internal flow using Navier-Stokes equations. J. comput. Phys. (in press). Papoutsakis, E. T. (1991) Fluid-mechanical damage of animal cells in bioreactors. TIBTECH 9, 427-438. Peskin, C. S. and McQueen, D. M. (1980) Modeling prosthetic heart valves for numerical analysis of blood flow in the heart. J. comput. Phys. 37, 113-132. Peskin, C. S. and McQueen, D. M. (1983) Computer assisted design of pivoting disc prosthetic mitral valves. J. 7horacic Cardionasc. Surg. 86, 126-135. Peskin, C. S. and McQueen, D. M. (1989) A three-dimensional computational method for blood flow in the heart. J. comput. Phys. 81, 372-405. Pulliam, T. H. and Steger, J. L. (1978) On implicit finite-

Rogers, S. E., Kwak, D. and Kiris, C. (1989) Numerical solution of the incompressible Navier-Stokes equations for steady-state and time-dependent problems. AIAA 27th Aerospace Science Meeting, January 1989, Reno, NV, AIAA Paper 894463. Snyder, W. V. (1978) Contour plotting [J6]. ACM Trans.
math. Software 4, 290-294.

Soh, W. Y., Goodrich, J. M. (1988) Unsteady solution of incompressible Navier-Stokes equations. J. comput. Phys.
79,113-134.

Stevenson, D. M., Yoganathan, A. P. and Williams, F. P. (1985) Numerical simulation of steady turbulent flow through trileaflet aortic heart valves. J. Biomechanics 18,
899-926.

Tarbell, J. M., Baldwin, J. T., Jarvis, P., Franciochelli, D., Frangos, J. A., Deutsch, S. and Geselowitz, D. B. (1989) Fluid mechanics and cellular interaction in the Penn state artificial heart. BioProcess Engineering Symposium, 1989, ASME, New York. Tarbell, J. M., Gunshinan, J. P., Geselowitz, D. B., Rosenberg, G., Shung, K. K. and Pierce, W. S. (1986) Pulsed ultrasonic Doppler velocity measurements inside a left ventricular assist device. J. biomech. Engng 108, 232-238. Thompson, J. F. (1985) Numerical Grid Generation, pp. 188-202, 310-315. Elsevier, New York. Tsai, P. Y-L. (1988) Time-marching schemes for solving the Euler equation. Ph.D. thesis, Penn State University, PA. Woo, Y. R. and Yoganathan, A. P. (1985) In vitro pulsatile flow velocity and turbulent shear stress measurements in the vicinity of mechanical aortic heart valve prostheses.
Life Support Systems 3, 283-312.

Yoganathan, A. P., Corcoran, W. H. and Harrison, E. C. (1979) In vitro velocity measurements in the vicinity of aortic prostheses. J. Biomechanics 12, 135-152.

Potrebbero piacerti anche