Medical Engineering & Physics 22 (2000) 381393 www.elsevier.

com/locate/medengphy

The ow patterns within the impeller passages of a centrifugal blood pump model
S.C.M. Yu *, B.T.H. Ng, W.K. Chan, L.P. Chua
Nanyang Technological University, Thermal and Fluids Engineering Division, School of Mechanical and Production Engineering, Singapore 639798 Received 8 March 2000; received in revised form 5 July 2000; accepted 17 August 2000

Abstract The effects of impeller geometry on the performance of a centrifugal blood pump model [the MSCBP design of Akamatsu and Tsukiya (The Seventh Asian Congress of Fluid Mechanics (1997), 710) at a 1:1 scale] have been investigated both experimentally and computationally. Four impeller designs were tested for pump hydraulic performance at the operating point (i.e. 2000 rpm), using blood analog as the working uid. Each impeller has seven blades with different congurations including the radial straight blade and backward swept blade designs. The results show that both designs can achieve a stable head of about 100 mm Hg at the operating point. Subsequent investigations involved the visualization of the relative ow eld within the impeller passages via the image de-rotation system coupled with a 2.5 W argon ion laser. Flow structures in all sectors of each impeller were examined and discussed. To further quantify the possible effects of blade geometry to thrombus formation and hemolysis, computational uid dynamics (CFD) was used to simulate a simplied two-dimensional blade-to-blade ow analysis so as to estimate the shear stress levels. The results indicate that the stress levels found within the blade passages are generally below the threshold level of 150 N/m2 for extensive erythrocyte damage to occur. There are some localized regions near the leading edge of the blades where the stress levels are 60% above the threshold level. However, given such a short residence time for the uid particles to go through these high shear stress regions, their effects appear to be insignicant. 2000 IPEM. Published by Elsevier Science Ltd. All rights reserved.

1. Introduction In general, all mechanical circulatory support devices that replace or assist the physiological function of the heart must exhibit a certain acceptable level of bio-compatibility which is dened as the ability to pump blood with minimal hemolysis and thrombus formation. In addition, they must also exhibit the following features [1]. Such a device must: generate a ow rate of approximately 56 l/min of blood against the mean aortic pressure of about 100 mm Hg for left ventricular assisting devices (VADs); be inexpensive, small and compact to facilitate the implantation of the device in adults and children;
Abbreviations: CFD, computational uid dynamic; MSCBP, magnetically suspended centrifugal blood pump; PS, pressure side; RBC, red blood cell; SS, suction side; VAD, ventricular assisting device. * Corresponding author. Tel.: +65-790-5595; fax: +65-791-1859. E-mail address: mcmyu@ntu.edu.sg (S.C.M. Yu).

be bio-chemically stable within the body; provide reliable and efcient long-term cardiac support (greater than 1 year); have minimal hemolysis and thromboembolic complications; have low heat generation so that damage to surrounding tissue is minimized; not cause any complications to other organs; and enable patients to lead a near normal lifestyle. Although pumps operating on the pulsating principles are, in general, preferable, they are usually very expensive, costing between US$16,000 and $200,000 [2]. Surgical problems relating to the use of pulsatile devices have often resulted in postponement of implantation. Clearly a less complex, less expensive and easy-to-use device is desired. Researchers such as Treichler et al. [3] and Nose et al. [4] recognized many inherent features of the centrifugal blood pumps that could have contributed to an excellent implantable long-term device. Firstly, the reduced size of the non-pulsatile device,

1350-4533/00/$ - see front matter 2000 IPEM. Published by Elsevier Science Ltd. All rights reserved. PII: S 1 3 5 0 - 4 5 3 3 ( 0 0 ) 0 0 0 4 5 - X

382

S.C.M. Yu et al. / Medical Engineering & Physics 22 (2000) 381393

Nomenclature g H k P Q Re Si Sx,y,z t u u,v,w (u )2,(v x,y,z r m n q t txx,tyy s1,s2 w gravitational acceleration (m/s2) head rise (mm Hg) turbulent kinetic energy pressure (Pa) ow rate (l/min) Reynolds number source term due to internal energy momentum source term in the x and y directions, respectively time (s) velocity vector velocity in the x, y, and z directions (m/s) )2,(u v ) Reynolds stresses Cartesian coordinates density (kg/m3) dynamic viscosity (kg/ms) kinematic viscosity (m2/s) temperature (K) shear stress (Pa) normal stresses (Pa) Laplacian operator dissipation function maximum and minimum normal stresses, respectively (N/m2) rotational speed (l/s)

which is due to the absence of heart valves and a large compliance chamber, facilitates the ease of insertion into the body without dissection of the patients diaphragm. Secondly, the simplicity in design of the non-pulsatile pump simplies the manufacturing process tremendously and hence greatly reduces the cost of a non-pulsatile centrifugal blood pump. Although there are reports, for example by Nakata et al. [5] and Sezai et al. [6], that continuous blood ow provided by the centrifugal pump will have several detrimental physiological effects, Allen et al. [7] have indicated that these effects are only temporary. Despite these controversies on the effect of non-pulsatile and pulsatile blood ow, it is certain that non-pulsatile ow is not a limiting factor to maintain life as reported by Yada et al. [8] and Minami et al. [9]. Thus, clinical centrifugal blood pumps have gained increasing roles in providing support for patients with a failing heart, due to their ease of application in the heartlung machine, relative low cost, simplicity, small size and ability to generate high ow rates. These include management of patients with cardiogenic shock by Minami et al. [10], shock after myocardial infarction (MT) by Noda et al. [11] and as a bridge to transplantation by Deleuze et al. [12]. Uses of these pumps were estimated to replace the roller pumps in 30% of the cardiac surgical procedures [13]. One of the ways to overcome the problems of thrombus formation and heat generation at the shaft/seal inter-

face in a centrifugal blood pump would be to reduce the complications at the interface (or to eliminate the shaft) and develop a completely seal-less centrifugal pump. Many research groups like Hart et al. [14] and Khanwilkar et al. [15] are constantly working on a better and more compact centrifugal blood pump system. Despite these continual efforts, all centrifugal blood pumps still have high potential for blood trauma due to the high level of energy being transferred from the fast spinning rotor to the blood. Hemolysis occurs when blood is subjected to high shear forces and sudden ow directional change in the pump housing. Thrombus formation occurs in low velocity and re-circulating regions. Therefore to design a good centrifugal blood pump with minimal hemolysis and thrombosis, high shear rate, ow re-circulation and stagnation regions have to be avoided or minimized. Akamatsu et al. [16] reported that a magnetically suspended centrifugal blood pump (denoted as MSCBP) is proven to have a satisfactory pump performance. The advantage that the MSCBP has over the commercial centrifugal blood pump is the use of magnetic suspension that eliminates leakage and sealing problems. It also prevents blood clots on the shaft and heat generation by the shaft and bearings. Subsequent hemolytic tests performed by Park et al. [17] and Nishimura et al. [18] on Akamatsu et al.s design showed that the index of hemolysis was signicantly lower in the MSCBP than in other

S.C.M. Yu et al. / Medical Engineering & Physics 22 (2000) 381393

383

non-magnetically suspended centrifugal blood pumps. In fact, clinical application of the MSCBP for long-term ventricular support (more than 200 days) has been achieved in a sheep as reported by Nishimura et al. [18]. However, the MSCBP is not exempted completely from problems such as hemolysis and thrombosis, which are common in all blood pumps. Thrombus formation appeared at the narrow gap between the impeller and the volute as observed by Yamada et al. [19]. Though a larger gap would induce more regurgitant ow through the gap and thus reduce the thrombus formation, it will result in lower pump efciency. Too narrow a gap will induce higher shear stress in the gap resulting in hemolysis. Park et al. [17] reported that maximum pump efciency occurs for a gap range between 0.15 and 0.25 mm, but this does not guarantee minimum blood trauma. Other possible sites for blood trauma to occur include the impeller passages where ow recirculation and stagnation exist. Akamatsu and Tsukiya [20] attempted to study the ow patterns within the impeller passages in the MSCBP at the off-design speeds of 300 and 600 rpm. To achieve Reynolds principle of similarity, the uid viscosity was modied. Similar to the scaled-up model, the advantage of this technique is the reduction of pump rotational speed through the assumption of similarity. However, it should be noted that the results may not be the true representation of the ows at higher rotating speed (usually at around 2000 rpm at the design point). The objectives of the present investigation are two fold. Flow visualization was conducted rst to examine any regions of ow stagnation and recirculation within the impeller passages. The results were subsequently compared with the output from the computational uid dynamics (CFD) analysis. After establishing the qualitative agreements between the experiments and CFD results, further investigations using CFD were carried out to examine quantitatively the effects of ow stagnation and recirculation to the thrombus formation and hemolysis. All impellers considered here have the same blade numbers. The relative ow patterns within the impeller passages were examined at the operating condition, i.e. at the design speed of 2000 rpm, with blood analog as the working uid. The following section describes the blood pump model and the experimental set-ups including the image de-rotating system for ow visualization. A brief description of the CFD code used (Fluent V4.3 [21]) will also be provided. These sections will be followed by presentation and discussion of the results. The paper ends with a summary of the most important ndings. 2. Pump congurations and instrumentation 2.1. Pump geometry and test rig A transparent acrylic model of the centrifugal blood pump (1:1 scale) was machined and tested in a closed

circuit loop. A schematic layout of the test rig is presented in Fig. 1(a) and Fig. 1(b) shows the test rig together with the ow visualization set-up. The closeloop circuit shown in the gure consists of a uid reservoir, a throttle clamp, pressure transducers at the inlet and outlet of the pump and a digital owmeter. The pump was connected to the reservoir with a 25.4 mm diameter siliconplatinum cured tubing. The working uid was supplied to the centrifugal pump inlet from a large reservoir. A schematic view of the model blood pump is shown in Fig. 2(a) with detailed dimensions given and is similar to the MSCBP developed by Akamatsu et al. [16]. The main difference between the present model pump and the pump developed by Akamatsu et al. [16] is that the present model pump is not magnetically suspended. The main drawback of the MSCBP was that optical access was not possible inside the impeller passages due to the presence of gap sensors and electromagnets. The model pump was therefore designed to be shaft-driven by a servomotor with one side of the volute optically accessible. Furthermore, the shaft-driven design can eliminate any possible effects of impeller meandering so that a more stable condition can be obtained for the study of ow patterns between impeller passages. The model blood pump consists of two main parts: an impeller and a volute. The pump had a double volute casing, which comprised of two circular arc-shaped enclosures surrounding the impeller. The pump inner inlet and discharge pipe diameters were 12 and 16 mm, respectively. The volute was manufactured in two parts via the CNC machine, i.e. lower and upper volutes as shown in Fig. 2(b). The impeller was sandwiched between the volutes and was allowed to rotate freely without rubbing the inner surfaces of the volutes. Blood analog was used as the working uid. Two parts of pure glycerine were mixed with three parts of distilled water in order to achieve the required composition which has a dynamic viscosity of 3.5 centipoise. 2.2. Geometry of the impellers The impellers for the present investigation were closed-type shown in Fig. 3(a). Each impeller has an inner and outer diameter of 13 and 25 mm, respectively. It was formed by sandwiching the seven vanes between an upper and lower shroud. The blade proles and the lower shroud of the impeller were machined from a solid aluminum block by the CNC machine. The impeller was then anodized (black) to eliminate optical interference during ow visualization. The upper shroud, which was made of clear acrylic sheet, was polished and mounted onto the lower shroud by means of three M1 screws. The vanes of the impeller had a uniform height of 3.5 mm. This impeller was then mounted onto the shaft, which was driven by a motor unit providing a constant

384

S.C.M. Yu et al. / Medical Engineering & Physics 22 (2000) 381393

Fig. 1.

Schematic of the experimental set-up.

rotational speed of the centrifugal blood pump (to within 0.5%). This motor unit, with a built-in feedback control to regulate the speed, had maximum rated output, speed and torque of 100 W, 3000 rpm and 0.32 N m, respectively. A total of four impellers with different blade proles were studied using the same volute. As shown in Fig. 3(b), the rst impeller (R7) is a straight radial blade impeller design with inlet and exit blade angles of 90. The other three impellers are backward-facing impellers (denoted as A3, A4 and B2, respectively) with inlet and outlet blade angles at 6.74 and 30, respectively. The proles of the blade curvatures vary for all three cases. Further details of the blade prole design can be obtained from Li [22]. Measurement of the relative static pressure difference was located at approximately two diameters upstream of the pump inlet and ve diameters downstream of the pump outlet (by convention). Their difference is measured by the pressure transducer (Druck DPI 260 series Digital Pressure Indicator) via silicon tubing. The pressure transducer displays differential pressure from 0 to 250 mm Hg with accuracy of 0.04% full scale.

2.3. Image de-rotating technique Analysis of the relative ow patterns between the blades of an impeller is difcult when an observer remains stationary relative to the impeller. However, when the observer rotates with the impeller at the same speed, the relative ow eld can be observed. The derotator optically freezes the image of the rotating impeller, which is then analyzed by ow visualization to provide ow patterns within the blades. The principle of the rotating 459045 or the Porro prism is shown schematically in Fig. 4(a). When the upright object F is placed on the left side of the XX axis of the prism, a laterally inverted image (mirror image) is formed on the right side of the XX of the prism axis (c.f. Fig. 1(b) for the locations of prisms). This can be better understood by studying the diagram in Fig. 4(b) which shows the path of light forming the inverted image. Regardless of the orientation of the object F, a laterally inverted image of the object F can be obtained. However, when the object rotates by 90, the prism needs to rotate only by 45 in the same direction to obtain an inverted but upright image. Simi-

S.C.M. Yu et al. / Medical Engineering & Physics 22 (2000) 381393

385

Fig. 2.

Schematic of the blood pump model.

larly, if the object rotates by 180 with the prism rotating to 90, the image formed will continue to be inverted and upright. Therefore, for an object rotating at a certain angular velocity, the image of a rotating object will always be upright but inverted and stationary in real time if the image de-rotator rotates in the same direction but at half the speed of the object. In order to render the ow pattern visible, polystyrene spheres of 50 m were used as tracers. A 2.5 W argonion laser operated in continuous mode was employed as a light source (c.f. Fig. 1(b)). Through a cylindrical lens, the laser beam was transformed into a light sheet of 1mm thickness to illuminate the ow. The tracer particles in the plane of the light sheet were then observed and photographed in the direction normal to the illuminated plane. The ow patterns within the illuminated region of interest, i.e. the blades of the impeller, can be viewed by a CCD video camera via a beam-splitter. The camera used has a frame rate of 25 per second with monochrome

high resolution of 768 by 576 pixels and a 2/3 CCD array (JAI, CV-3000). It was capable of electronic shutter speeds up to l/10,000 s. An image acquisition board (MV-1000) was used to digitize the analog camera video output into 8 bits per pixel at rates up to 40 million samples per second. These digitized images were then stored in a personal computer for further image analysis.

3. Numerical method A commercial CFD code Fluent (V4.3) [21] was used to perform the numerical investigation. The fundamental governing equations in describing two-dimensional, steady, compressible uid ow can be written as below: Continuity equation: div(ru) 0 (1)

386

S.C.M. Yu et al. / Medical Engineering & Physics 22 (2000) 381393

div(riu)

rdiv(u) div(k q)

Si

(4)

Fig. 3. Geometry of the impeller blades investigated.

Fig. 4. Optical principle of image de-rotator (taken from Ohki et al. [23]).

Momentum equations: (ruu) x (rvu) y p div(m u) Sx x p div(m v) Sy y (2) (3)

Internal energy equation:

where is the Laplacian operator, p is the pressure, i is the internal energy, t is the time, u is the velocity vector, k is the thermal conductivity, r is the density, m is the dynamic viscosity, q is the temperature, u and v are the velocities in the x and y directions respectively, is the dissipation function, Si is the source term due to internal energy and Sx,y is the momentum source term in the x and y directions, respectively. Fluent uses the Semi-Implicited Method for PressureLinked Equations (SIMPLE) algorithm and SIMPLEConsistent (SIMPLEC) with convergence techniques such as block correction and multi-grid method, and provides the velocity and pressure elds as solutions for the discretized form of the above governing equations. Fluents ability to model ow in a rotating frame of reference involving complex geometry with high robustness and efciency makes the code suitable for calculating ow in rotating machinery. Further details can be found in the Fluent V4.3 manual [21]. In this initial stage of numerical analysis, the focus will be on the prediction of relative ow eld between two blades of an impeller without considering the inuence of the volute. As the ow through a real turbomachine is three-dimensional, unsteady and viscous, it is necessary that all these characteristics be taken into account during the design stage [24]. In spite of the availability of powerful computer hardware, solutions to such complex equations are time consuming and expensive owing to limitations in CPU and memory. Therefore approximations to simplify the problem are necessary and are made by reducing the three-dimensional to twodimensional problems, which are amendable to analysis. It is a well-known fact that in centrifugal impellers, the axial velocity component can be neglected as compared to the radial and angular components. Hence, it was reasonable to approximate the passage ow as a two-dimensional problem. Only a segment of the impeller blades was modeled (see Fig. 3(a)). The total number of nodes for each blade passage was about 6283. There were 21 nodes in both the inlet and outlet extensions and 61 nodes on the blade surface, giving a total of 103 grid points in the radial direction. In the circumferential direction, there were 61 grid points. The calculation was performed on a Pentium-II 400 MHz processor and it required about 10,00015,000 time-steps, which was approximately 67 h of central processor unit (CPU) time. At the inlet of the computational domain, an inlet relative velocity was prescribed with the assumption of no pre-rotation. The inlet velocity consisted of a normal velocity component, which was in the radial direction and specied according to the ow rate of 5 l/min, and a tangential velocity component based on the design rotational speed. At the outlet, the normal velocities were adjusted to satisfy the overall

S.C.M. Yu et al. / Medical Engineering & Physics 22 (2000) 381393

387

mass balance for the computation domain. At the blade surface, the wall was dened with zero velocity. It was assumed that the uid was circumferentially periodic and invariant with ow passages, thus CYCLIC boundary conditions were specied at the two sides of the extension areas. All velocity components, pressure, turbulence kinetic energy (k) and dissipation rate (e) were the same on the CYCLIC boundaries. This permitted the simulation of only one of the seven blade passages of an impeller. The values of density (r) and dynamic viscosity (m) of the medium were taken to be 1050 kg/m3 and 3.510 3 kg/m s, respectively, to approximate blood properties under normal body temperature. The Reynolds number based on the angular velocity and the outer impeller diameter was approximately 40,000. The ow inside the blade passage may be turbulent due to the geometrical complexity and rotational effect. Therefore, a laminar ow model was not used to simulate the highly complex rotational ow in the impeller. The standard e turbulent model was therefore adopted.

4. Results and discussion 4.1. Pump performance Fig. 5 shows the variation of the head rise (H) against the ow rate (Q) for respective impeller designs. The measured head produced by all these impellers are found to be stable at each ow rate point measured (to within 1%). Also shown in the gure for comparison are the theoretical head curves obtained from the Eulers equation for outlet blade angle at 30 and 90, respectively. As may have been expected, the head generated by each of the four impeller designs gradually fell as the ow rate (Q) increased. The head at the design point (5 l/min) for the radial blade was found to be the highest

Fig. 5. Pump performance curves for respective impeller blade designs ( Eulers analysis: blade trailing edge angle at 90; Eulers analysis: blade trailing edge angle at 30).

at about 100 mm Hg. The performance curves for the straight radial blade and backward swept blade show some differences, to within 10%, with the former one being slightly higher. For the backward swept blade design, keeping the inlet and outlet blade angles constant, some further improvements on the head generated across the pump can be obtained by varying the blade proles. Maximum increase was around 12% at the design point; for example, compare impellers A3 and B2. A few important observations can be extracted from Fig. 5. First, increasing the outlet blade angle would lead to higher head in the performance curves. The performance curves estimated from the Eulers equation show that the backward facing prole with an outlet blade angle of 30 would produce a negative slope and the head generated at the design point would be below that of the straight radial blade. This is further supported by comparing the performance curves of R7 and other backward swept blade designs (although there is small difference in the inlet angle between the two designs). The performance of impeller R7 is expected to improve further if the inlet blade angle can match more closely with the inlet ow angle. Secondly, it can also be seen that the prole of impeller blade is an important factor in generating a higher head. Comparing impellers A3, A4 and B2, despite having the same inlet and outlet blade angles, the achievable head increased from 84 mm Hg for impeller A3 to 88 mm Hg for impeller A4 and nally to 95 mm Hg for impeller B2. There is about a 13% increase from impeller A3 to impeller B2. Therefore changing the blade curvature while keeping both the inlet and outlet blade angles constant can enhance the pump performance. Thirdly, the pump characteristic is closely related to the losses within the pump. Comparing the theoretical head and the head generated by impeller R7, a head loss of 105 mm Hg is noted. For impeller A3, the difference is about 111 mm Hg. The head loss for impeller A4 reduced to 107 mm Hg and to 100 mm Hg for impeller B2. The difference between the theoretical and measured is a result of losses and of the ow from the blades. The deviation (referred to as slip in the pump literature) is a result of the relative rotation, a nite number of blades and viscous effects. Head loss occurring in the MSCBP pump are summation of several factors according to Akamatsu et al. [16]. They are rstly the mechanical losses caused by the physical contact between the shaft/seal interface and the shaft bearings. In addition, there are disk friction losses at the gap between the impeller and the volute. Losses in the form of uid leakage that occurred when part of the uids leaving the impeller exit returned back to the impeller inlet through the gap and the losses in the volute also reduce the pump efciency. Finally, hydraulic losses in the impeller passages as a result of

388

S.C.M. Yu et al. / Medical Engineering & Physics 22 (2000) 381393

ow separation are also an important factor. Akamatsu et al. [16] had estimated the losses contributed by each of the three factors mentioned above in the MSCBP and had identied that the hydraulic loss is comparable to the disc friction loss at the design point, contributing about 35% of the total losses. The leakage loss is the lowest among the three losses (about 20%). The subsequent focus of the present investigation will be on hydraulic loss due to ow separations within the impeller passages. 4.2. Flow visualization results The relative ow patterns of the seven passages of the four impellers (R7, A3, A4 and B2) were obtained using the ow visualization set-up described earlier. The ow patterns were taken at the mid-height of the vanes (c.f. Fig. 3(a)). Since the ow patterns for the three backward swept designs are very similar, only the results for the A3 design will be presented here. Figs. 6 and 7 show the relative ow patterns in every sector (S1, S2, S3, S4, S5, S6 and S7) between the blades of the R7 and A3 designs. In each gure, the blade rotates in the clockwise direction causing the side of the blade that leads the rotation to be the pressure side (PS) while the side of the blade that follows the rotation is the suction side (SS). The diagram in the center of the gures shows the schematic of the pump and impeller, and position of impeller with respect to the volute. All ow visualizations were conducted at the design point, i.e. at a rotational speed of 2000 rpm and a ow rate of 5 l/mm.

The shutter speed of the camera was set to 1/500th of a second. 4.2.1. Radial impeller (R7) In every sector of the impeller R7 in Fig. 6, a pair of large-scale recirculating vortices near the suction side was observed while the bulk ow went smoothly along the pressure side. In the case of radial blade impeller where the inlet blade angle is at 90, the direction of the relative velocity vectors at the inlet differ from that of the blade angle, thus giving rise to early ow separations. Fig. 8a shows a sketch of the inlet velocity triangle and the ow patterns between any two blades. From the sketch it is noticed that regardless of the rotational speed of the impeller, the relative velocity components at the inlet of the blade passage will always differ from the inlet blade angle. The vortex which appeared near the leading edge of the suction side is largely due to the mismatch between the inlet ow and blade angles. The second vortex near the trailing edge of the suction side is due to the adverse pressure gradients developed along the blade. It should be mentioned that the two recirculating vortices do not represent dead water regions. It was clearly shown during the visualization tests that tracer particles entering the blade passages went smoothly through the recirculation regions and left the passage. The ow phenomena within the passages is not completely two-dimensional as indicated by the criss-crossing of streaklines observed in the passages. At the impeller inlet region (c.f. Fig. 2(a), where the inlet center cone

Fig. 6.

Flow visualization at different impeller passage of the blood pump model with impeller design R7.

S.C.M. Yu et al. / Medical Engineering & Physics 22 (2000) 381393

389

Fig. 7.

Flow visualization at different impeller passage of the blood pump model with impeller design A3.

changed the direction of ow from axial to radial), the uids may not enter the passage with uniform velocity distribution along the vane height and as a consequence, causing the criss-crossing of the streaklines. The ow structures in all the sectors are qualitatively similar to one another. This implies that the interaction of the volute and the ow within the impeller passages may not be signicant at the operating point. However, some local effects of the volute on certain sectors of the impellers exist. This can be observed in the spatial extent of the recirculation vortices formed. Although the vortices appeared mainly on the suction side, the area occupied by the vortices varied at different sectors. For example, in sectors S4 and S5 it appeared that the size and strength of the vortices were larger than those in the other sectors due to the presence of the splitter plate. The splitter plate had restricted the ow from leaving the passage causing ow reversal at the passage outlet. In sector S1, the streaklines shown in the ow pattern near the impeller passage exit seem to be shorter indicating a lower velocity. This is due to the blockage effect by the cut-water (or tongue region in Fig. 2(a)). Their presence restricted the amount of uids leaving the passage. It should be noted that the uid that departed from the passage had to squeeze through the space between the tongue and the vanes, subjecting itself to higher velocity gradients. For sectors S2, S3, S6 and S7, the uids left the passage into the volute without any restrictions. 4.2.2. Backward facing impeller (A3) The change in the blade shape from the straight radial type to a backward facing blade altered the ow patterns. This caused a reduction in the number of vortices from

Fig. 8. Sketch of the inlet velocity triangles and ow patterns between the blades of respective impeller designs (mirror image view).

390

S.C.M. Yu et al. / Medical Engineering & Physics 22 (2000) 381393

two in impeller R7 to one in impeller A3. Similar observations were also found in the other two backward swept designs. Unlike impeller R7, where the vortices were formed at both the leading and trailing edges of the suction blade, there is no vortex formation at the leading edge of impeller A3 due to the smaller difference between the inlet ow and blade angles. From Fig. 7(a) and (e), it is evident that ow separation did not occur at the leading edge of the suction side because the ow within the impeller A3 at design condition seemed to be better aligned with the inlet blade angle. It is observed that the ow patterns within the blade passages of impeller A3 are qualitatively similar to each other with the vortex being conned to the region near the trailing edge of the suction blade while the main ow remained attached along the pressure side in each sector. The formation of the vortex was likely to be the result of adverse pressure gradients causing the ow to separate. A sketch of the ow pattern between the blades of impeller A3 is shown schematically in Fig. 8(b). Along the pressure side, the streaklines in the blade passage of impeller A3 are in general longer than those observed in impeller R7 indicating higher velocities. For a given blade number, though the circumferential blade passage area is the same, the blade prole would certainly affect the ow pattern. In the case of impeller A3, uid particles have to accelerate through the curved blade passage as compared to impeller R7. Higher velocity gradients would appear on the pressure side and as a result incur higher levels of shear stresses. The longer blade length in impeller A3 may be a potential source of blood trauma. The ow patterns in all sectors are qualitatively similar to one another. However, similar to those found in impeller R7, some local effects on the ow passages due to the volute/impeller interaction may exist. This is shown by the variation in the size of the vortices. The vortices in sectors S1, S2, S3, S6, and S7 were about the same size. In sector S4, the vortex seemed to be of larger size compared to the rest of the sectors, blocking most of the ow passage. The large vortex size could be attributed to the blockage of uids from leaving the passage by the presence of the splitter plate. There were two vortices present in sector S5. A pair of vortices, comprising of a large vortex and a smaller one, appeared at the trailing edge of the suction blade. The existence of the smaller vortex might be due to induction by the bigger vortex. Near the cut-water region at sector S1, the uid did not seem to be able to exit the passage into volute. At the passage outlet where the volute area was very limited, the uids got bounced off the volute wall and reversed back into the impeller passage. This can be observed by the streaklines criss-crossing each other at the corresponding region. In general, for the other backward facing impeller

designs, i.e. impeller A4 and B2, the ow patterns were qualitatively similar to those observed in impeller A3. Vortex formation was found at the trailing edge of the suction side while ow remained attached along the pressure side. The volute had some local effects on the ow patterns in several blade sectors. From the ow visualization studies of the four impellers, it is postulated that the number of vortices could be reduced if the relative ow angle can be aligned better with the inlet blade angle. However, vortex formation seemed to be unavoidable at the suction side of the blade due to the build-up of the adverse pressure gradients. 4.3. Numerical simulation of ow within the impeller passages 4.3.1. Velocity vectors The computed velocity distributions in a single passage of impellers R7 and A3 are presented in Fig. 9. The gure shows the relative ow eld of the blade passages of the three impellers. The blades are rotating in the counter-clockwise direction. This results in the suction side (SS) on the upper surface and pressure side (PS) on the lower surface. In the case of impeller R7 and Fig. 9(a), a pair of strong re-circulating vortices indicated by the letters A and B is observed within the blades. The ow is restricted to the pressure side of the blade passage. The simulation clearly shows that the formation of vortex A is mainly due to the difference in the relative ow angle and blade angle, while vortex B is due to the adverse pressure gradients appearing on the suction side. Similarly, a dominating vortex B and a small vortex A appearing near the impeller outlet are observed for impellers A3, A4 and B2. Smooth ow is also restricted to the pressure side. In general, the experimental ow visualization is qualitatively similar to the predicted ow eld for all the impellers revealing the dominance of vortex B. It should be mentioned that the smaller vortex A predicted by CFD is not obvious in the experimental ow visualization of these impellers. Numerical simulation and ow visualization have shown the difference in ow patterns due to the different vane proles. The agreement between the numerical simulation and experimental ow visualization results is reasonably good. Through observations of ow patterns, blood trauma can be anticipated by identifying the region of re-circulation. However, this does not quantitatively indicate the level of damage imposed onto the blood cells. It is therefore desirable to establish quantitatively the effects of shear stress on blood caused by different blade proles. 4.3.2. Shear stress analysis The maximum shear stress tmax acting on the blood cell in a two-dimensional analysis is the resultant of two

S.C.M. Yu et al. / Medical Engineering & Physics 22 (2000) 381393

391

The resultant shear stress can be obtained by rearranging Eqs. (5)(8) into Eqs. (9)(11). s1 s2 tmax txx+tyy txx+tyy cos 2a txy sin 2a 2 2 txx+tyy txx+tyy cos 2a txy sin 2a 2 2 |s1s2| 2 (9) (10) (11)

where r is the uid density, m is the dynamic viscosity, (u )2,(v )2, and (u v ) are the independent Reynolds stresses, s1 and s2 are the maximum and minimum normal stress values, respectively. Hence by discretizing the equations, the maximum shear stress tmax can be calculated at every cell within the blade passage. Fig. 10 shows the shear stress distribution for impellers R7 and A3. From the shear stresses contours, it is

Fig. 9. Computed velocity vectors at the mid-height of respective impeller passage.

normal stresses txx, and tyy and the two shear stresses act on the faces of a control volume. Shear stress in viscous ow is divided into laminar and turbulent parts, and can be calculated from the velocity vectors based on the following equations: txx u u v 2 2m + m + x 3 x y laminar shear v tyy 2m y txy m tan 2a u v y x 2txy txytyy u v 2 m 3 x y ru v +r(u )2 turbulent shear r(v )2 (6) (7) (8)
Fig. 10. Shear stress contours at the mid-height of respective impeller passage.

(5)

392

S.C.M. Yu et al. / Medical Engineering & Physics 22 (2000) 381393

observed that the majority of the ow passage had shear stress magnitude below 40 N/m2. This is well below the threshold level of about 1500 dynes/cm2 or 150 N/m2 for extensive erythrocyte damage to occure [25]. These results are agreeable to low index of hemolysis found in the MSCBP by Park et al. [17]. However, it should be noted that there are actually a few small regions (less than 1/10 of the blade length) within the blade passage for impeller R7 and A3 that are associated with high shear stresses. One particular high shear stress region is at the wall of the pressure side close to the blade leading edge. It is the stagnation point where uid particles rst come into contact with the rotating blades. Another region that induced high shear stress is along the pressure side. The uid near the wall of the pressure side produces a very high velocity gradient due to the viscous effects. This results in a high shear stress region at certain parts along the blade on the pressure side. The interface between the pair of re-circulating vortices is another site of high shear stress magnitude. The maximum shear stress that occurred within the blades for impellers R7, A3, A4 and B2 are approximately 230, 220, 273 and 267 N/m2, respectively, and these are mostly found at the leading edge of the pressure side. Although these results are above the threshold level of 150 N/m2, it is unlikely that they would cause irreversible damage to blood cells, as these happened only for a short period of time (the transit time for a uid particle to go through the pump impeller passage is estimated to be about 0.03 s). The pump has an output of 5 l/min at the design point and if we assume the blood pump patient also has a blood volume of 5 l, all the blood will go through the pump once every minute. If we assume all RBC spend 0.03 s inside the impeller passage for every minute in which 1/10 of their time (in proportion to the length of the blade) is in the high shear stress region near the leading edge, i.e. 0.003 s, for every 24 h, the RBC would experience a total of 4.32 s of shear stress greater than 250 N/m2. It will take approximately 833 days before any possible rupture of the RBC could occur (when 250 N/m2 for 60 min is required for cell rupture, [26]). It should be mentioned that the above analysis is very crude and in blood pumps the blood is exposed to subhemolytic shear stresses may times over a long period of time. Their effects cannot be quantied easily by the present analysis.

5. Concluding remarks The present investigation focuses on the ow patterns within the impeller passages of a centrifugal blood pump. Four different types of blade geometry have been tested including the radial straight blade and backward swept blade designs. Both designs can achieve a head of about 100 mm Hg at the design point. Modifying the

curvature of the blade may improve the backward swept blade design performance but the effects appear to be marginal. Subsequent investigations focused on visualizing the relative ow patterns in all the seven sectors in each impeller design. Recirculation vortices were found in all the designs and they appeared mainly at regions close to the suction side. Moreover, the recirculation regions did not seem to give rise to any ow stagnant (dead water) regions whereby thrombus is likely to accumulate. The ow patterns in all the seven sectors were similar but localized effects from the volute existed. For example, near the cutwater region, the outward ow from the impeller passages was restricted and bounds back inside into the passage. This blockage effect however is considered transitory and smooth outward ow will resume when that passage continues to rotate to the other sector of the volute. Although the ow patterns inside the impeller passages appeared to be three-dimensional, as shown by the criss-crossing of the streaklines, the two-dimensional numerical approach shows good qualitative agreements with ow visualization results. The results indicate that the stress levels found within the blade passages are generally below the threshold level of 150 N/m2 for extensive erythrocyte damage to occur. There are some localized regions near to the leading edge of the blades where the stress levels are some 60% above the threshold level. However, given such a short residence time (estimated to be less than 0.03 s) for the uid particles to go through the blade passage, their effects seemed to be insignicant. It should be mentioned that the simulations were conducted using a simplied two-dimensional analysis and that the ow could be three-dimensional. Furthermore, the effects of the volute and splitter plate were not accounted for. Although the pump prototype investigated here was not magnetically suspended and the efciency among different designs cannot be compared directly, the results generally suggest that other differences between the two designs investigated are very little. Both can achieve the head required and the shear stress levels are found to be acceptable for both cases. It should be noted that the present experiments were conducted using a shaft whereby any possible meandering effects of the impeller structure were not accounted for. It can be expected that as the ow went around each impeller blade, an aerodynamics lift would be generated, as shown in Fig. 8(c). While the lift generated by the radial blade would be acting tangentially with respect to the blade center, the lift generated by the backward swept blade would be acting at an angle in the outward direction. This may be the likely cause of the impeller meandering effect observed by some of the curved blade designs ([27] private communications). On-going works include the examination of the disk friction loss and gap loss to the overall

S.C.M. Yu et al. / Medical Engineering & Physics 22 (2000) 381393

393

pump performances and their effects on the thrombus formation and hemolysis. It should be mentioned that the centrifugal pump may have several advances over the other pump designs such as the axial (or diagonal) pump. Although the size of the axial pump may be kept smaller than the centrifugal pump investigated here, to achieve the same ow rate at 5 l/min the working speed for the axial pump has to increase. Multi-stage design may have to be considered to achieve an acceptable head. This may inevitably lead to higher shear stresses (or hemolysis level) for the blood transiting the pump. It should be pointed out that the investigations on the simulated interaction of pump with the heart on our centrifugal pump have not yet been carried out although this may be improvised by adjusting the opening of the throttle clamp and elevation of the uid reservoir shown in Fig. 1(a). Furthermore, the pump is basically designed for the left ventricle and future work will include its possible applications with the right ventricle.

[8]

[9]

[10]

[11]

[12]

[13] [14]

Acknowledgements
[15]

The authors are grateful to Professor T. Akamatsu in providing much useful information with regards to the MSCBP design. Financial support to this project from the Academic Research Grant Committee and a research studentship to Mr Bernard T.H. Ng from the School of Mechanical and Production Engineering are gratefully acknowledged.

[16]

[17]

[18]

References
[19] [1] Nose Y. A rotary blood pump: its design and development strategy. Artif Organs 1997;21(4):2634. [2] Hwang NHC, Turitto VT, Yen MRT. Current advances in cardiac assist devices. In: Advances in cardiovascular engineering. New York: Plenum Press, 1992:34561. [3] Treichler J, Rosenow SE, Damm G, Naito K, Ohara Y, Mizuguchi K, Makinouchi K, Takatani S, Nose Y. A uid dynamics analysis of a rotary blood pump for design improvement. Artif Organs 1993;17(9):797808. [4] Nose Y, Koji K, Nakazawa T. Can we develop a nonpulsatile permanent rotary blood pump? Yes. We can. Artif Organs 1996;20(6):46774. [5] Nakata K, Shiono M, Orime Y, Hata M, Sezai A, Saitoh T, Sezai Y. Effect of pulsatile and nonpulsatile assist on heart and kidney microcirculation with cardiogenic shock. Artif Organs 1996;20(6):6814. [6] Sezai A, Shiono M, Orime Y, Nakata K, Hata M, Yamada H, Iida M, Kashiwazaki S, Kinishita J, Nemoto M, Koujima T, Sezai Y, Saitoh T. Renal circulation and cellular metabolism during left ventricular assisted circulation: comparison study of pulsatile and nonpulsatile assists. Artif Organs 1997;21(7):8305. [7] Allen GS, Murray KD, Olsen DB. The importance of pulsatile

[20]

[21] [22]

[23]

[24] [25] [26]

[27]

and nonpulsatile ow in the design of blood pumps. Artif Organs 1997;21(8):9228. Yada I, Golding L, Harasaki H, Jacobs G, Koike S, Yozu R, Sato N, Fujimoto L, Snow J, Olsen D, Murabayashi S, Venkatesen V, Kiraly R, Nose Y. Physiopathological studies of non-pulsatile blood ow in chronic models. ASAIO J 1983;29:5205. Minami K, El-Banayosy A, Posival H, Seggewi H, Murray E, Korner M, Korfer R. Improvement of survival rate in patients with cardiogenic shock by using nonpulsatile and pulsatile ventricular assist device. Int J Artif Organs 1992;15(12):71521. Minami K, Posival H, El-Banayosy A, Korner M, Schrofel H, Murray E, Korfer R. Mechanical ventricular support usig pulsatile abiomed BVS 5000 and centrifugal biomedicus-pump in postcardiotomy shock. Int J Artif Organs 1994;17(9):4928. Noda H, Takano H, Taenaka Y, Nakatani T, Umezu M, Kinoshita M, Tatsumi E, Yagura A, Sekii H, Kito Y, Ohara K, Tanaka K, Kumon K, Hiramori K, Yutani C, Beppu S, Fujita T, Akutsu T, Manabe H. Treatment of acute myocardial infarction with cardiogenic shock using left ventricular assist device. International Journal of Articial Organs 1989;12(3):1759. Deleuze PH, Liu Y, Okude J, Leandri J, Gouault M, Galacteros F, Cachera JP, Loisance D. Centrifugal pump or pneumatic ventricle for short-term mechanical circulatory support? Int J Artif Organs 1989;12(5):32732. Golding LAR, Smith WA. Centrifugal pumps now and the future. Assist Circulat 1995;4:14251. Hart RM, Filipenco VG, Kung RTV. A magnetically suspended and hydrostatically stabilized centrifugal blood pump. Artif Organs 1996;20(6):5916. Khanwilkar P, Olsen D, Bearnson G, Allaire P, Maslen E, Flack R, Long J. Using hybrid magnetic bearings to completely suspend the impeller of a ventricular assist device. Artif Organs 1996;20(6):597604. Akamatsu T, Nakazeki T, Itoh H. Centrifugal blood pump with a magnetically suspended impeller. Artif Organs 1992;16(3):3058. Park C, Nishimura K, Akamatsu T, Tsukiya T, Matsuda K, Ban T. A new magnetically suspended centrifugal pump: in-vitro and preliminary in-vivo assessment. Artif Organs 1996;20(2):12831. Nishimura K, Kono S, Nishina T, Akamatsu T, Tsukiya T, Nojiri C, Ozaki T, Komeda M. Results of chronic animal experiments with a new version of a magnetically suspended centrifugal pump. ASAIO J 1998;44:7257. Yamada T, Nishimura K, Akamatsu T, Tsukiya T, Park C, Kono S, Matsuda K, Ban T. In-vivo testing of a magnetically suspended centrifugal pump designed for long-term use. Int J Artif Organs 1997;20(10):5629. Akamatsu T, Tsukiya T. Development of the centrifugal blood pump with magnetically suspended impeller and the related uid mechanical problems. In: The Seventh Asian Congress of Fluid Mechanics, 1997:710. FLUENT Version 4.3 Users Guide. Fluent Inc, 1995. Li HD. Inverse design and optimization of blood pump impeller using computational uid dynamics. In: Final Research Report 199799. Nanyang Technological University, 1999. Ohki H, Yoshinaga Y, Tsutsumi Y. Visualization of relative ow patterns in centrifugal impellers. In: Proceedings of the 3rd International Symposium on Flow Visualization, 1983:7237. Nakamura S, Yano K. Computational simulation of ows in an entire centrifugal heart pump. Artif Organs 1999;23(6):5725. Leverett LB, Hellums JD, Alfrey CP, Lynch EC. Red blood cell damage by shear stress. Biophys J 1972;12:25772. Sutera S, Croce P, Mehrjardi M. Hemolysis and subhemolytic alterations of human RBC induced by turbulent shear ow. Trans Am Soc Artif Intern Organs 1972;18:5839. Akamatsu T. Private communication, 1999.