Sie sind auf Seite 1von 6

Computers & Fluids 38 (2009) 284289

Contents lists available at ScienceDirect

Computers & Fluids


journal homepage: www.elsevier.com/locate/compuid

A fast numerical method for ow analysis and blade design in centrifugal pump impellers
John S. Anagnostopoulos *
School of Mechanical Engineering/Fluids Section, National Technical University of Athens, 9 Heroon Polytechniou Avenue, Zografou, 15780 Athens, Greece

a r t i c l e

i n f o

a b s t r a c t
A numerical methodology is developed to simulate the turbulent ow in a 2-dimensional centrifugal pump impeller and to compute the characteristic performance curves of the entire pump. The ow domain is discretized with a polar, Cartesian mesh and the Reynolds-averaged NavierStokes (RANS) equations are solved with the control volume approach and the ke turbulence model. Advanced numerical techniques for adaptive grid renement and for treatment of grid cells that do not t the irregular boundaries are implemented in order to achieve a fully automated grid construction for any impeller design, as well as to produce results of adequate precision and accuracy. After estimating the additional hydraulic losses in the casing and the inlet and outlet sections of the pump, the performance of the pump can be predicted using the numerical results from the impeller section only. The regulation of various energy loss coefcients involved in the model is carried out for a commercial pump, for which there are available measurements. The predicted overall efciency curve of the pump was found to agree very well with the corresponding experimental data. Finally, a numerical optimization algorithm based on the unconstrained gradient approach is developed and combined with the evaluation software in order to nd the impeller geometry that maximizes the pump efciency, using as free design variables the blade angles at the leading and the trailing edge. The results veried that the optimization process can converge very fast and to reasonable optimal values. 2008 Elsevier Ltd. All rights reserved.

Article history: Received 23 January 2007 Received in revised form 28 February 2008 Accepted 28 February 2008 Available online 9 July 2008

1. Introduction The numerical simulation of the uid ow for the design in hydraulic turbomachinery has become a requisite tool in order to increase efciency and reduce cavitation. However, in spite of the continuous increase in computing power, the inverse design numerical optimization is still a laborious task, because it needs a large number of ow eld evaluations. Such computations may be very costly, especially when the entire 3D domain in both the impeller/runner and the casing of the machine are simulated, and many design variables are incorporated. For this reason, few real 3D inverse design methods have been developed, as the inverse time marching method [1], the pseudo-stream function method [2], and the Fourier expansion singularity method [3]. These methods are very time-consuming and exhibit some difculties in correlating the design parameters with the blade geometry (the rst two) or convergence problems (the latter). A quasi-3D method is recently proposed [4], which performs a blade-to-blade solution and saves computer time by using only one representative hub-to-shroud surface. All the above models are based on the inviscid simplied assumption. The application of fully 3-dimensional turbulent ow analysis for the impeller/runner design is in* Tel.: +30 2107721080; fax: +30 2107721057. E-mail address: j.anagno@uid.mech.ntua.gr 0045-7930/$ - see front matter 2008 Elsevier Ltd. All rights reserved. doi:10.1016/j.compuid.2008.02.010

creased in the past few years [511], usually with the aid of commercial CFD software. However, the use of NavierStokes validation in inverse design optimization methods is still not a common practice, since in addition to the time-consuming calculations there is a need for automated mesh generation in complex geometries. Some recent numerical works are directed towards that advanced approach [12,13]. The computer time needed by these models depends strongly on the generation cost of the body-tted grid, as well as on the grid quality. An alternative practice in complex domains is the use of Cartesian grids that require a much reduced construction effort. The main drawback of those is the inability of their lines to follow a non-orthogonal or irregular boundary. Several numerical techniques to improve the accuracy in such regions have been published; most of them they can be classied as cell-cut or immersed boundary methods [14,15]. In the present work a cell-cut sharp-interface method developed for the automated generation of Cartesian grids in irregular geometries is incorporated in a computer algorithm for the simulation of a centrifugal pump impeller operation. The shape optimization process with the use of Cartesian grid has been successfully applied for laminar ows in various geometries (eg. [16]). In order to accelerate convergence the computational domain contains only the pump impeller in a 2-dimensional approach, while special modelling is implemented to reproduce the characteristic performance curves

J.S. Anagnostopoulos / Computers & Fluids 38 (2009) 284289

285

of the entire pump. With this methodology the computer cost per evaluation is much reduced compared to a fully 3D simulation of the pump, giving thus a quick and reliable estimation of the optimum values of the free design parameters. The method is expected to be more successful for low specic speed impellers, where the ow is mainly 2-dimensional. 2. The numerical methodology For the simulation of the ow in a 2-dimensional pump impeller the incompressible NavierStokes equations are expressed in polar coordinates and a rotating with the impeller system. Using circular inner and outer boundaries, the governing equations for a horizontal space (no gravity) become:

savings in computer cost than each of the local grid renement or the higher-order discretization methods alone [17]. 2.2. Geometry representation A cell-cut, sharp-interface grid construction method, developed and tested with success in various applied studies in the past (eg. [16,18]), is modied and further improved in order to increase its accuracy near the irregular boundaries. With the new method no cell-merging is performed, but all the grid cells that are totally or partly lled with the uid (Fig. 2) are solved using the same general equation of the following linearized form:

AP cV SP UP AP X
i

X
i

ci Ai Ui cV SU ;
3

~0 Continuity : r w ~ x ~ ~ ~ w ~x ~ rw ~ 2x r Momentum : w

ci Ai ; i E; W ; N; S; U ; D

rp

r ~ s

~ is the uid velocity in the rotating system (relative uid where w velocity), x is the angular rotation speed of the impeller, p and q are the uid pressure and density, respectively, and ~ s is the stress tensor that includes both the viscous and the turbulence viscosity terms. The standard ke turbulence model is adopted, which is suitable for high Re number ows, as the examined here. The system of the averaged form of the above equations is numerically solved with the nite volume approach and a collocated grid arrangement, using a preconditioned bi-conjugate gradient (Bi-CG) solver. Setting cyclic boundary conditions the solution can be restricted to the 1/z part of the impeller, where z is the number of blades. 2.1. Numerical grid As stated in the Introduction, the use of Cartesian grids for numerical design optimization provides the signicant advantage of a fast and automated grid generation process. Other desirable features include the easiness in the construction and control of locally or adaptively rened regions, as well as the capability to use discretization schemes of higher, in general, accuracy, compared to other grid types. All the above features are developed and incorporated in the computation algorithm used for the present study. The numerical technique, more details of which can be found in Anagnostopoulos [17], introduces a multiple stencil that allows the application of second-order discretization schemes to any grid cell, regardless of its renement ratio or local grid topology, therefore it is applicable not only to rened but also to completely unstructured Cartesian grids. Moreover, in spite of the increased accuracy the resulting expressions remain simple and robust [17]. Finally, this method can be easily extended to 3-dimensions. An indicative picture of such a computational grid adaptively rened in two consecutive layers around the blades of a centrifugal impeller is shown in Fig. 1. For a given accuracy of the results, it was found that the above technique achieves considerably greater

where Ai are the coefcients linking the dependent variable UP with its neighbours on the adjacent grid volumes, and SU, SP are additional source terms. The geometric coefcients ci and cv represent the free portion (not blocked by the solid boundary) of the cell-faces and volume, respectively. The uid variables (velocities, pressures, etc) are computed at the centroid of the Cartesian cells, which for a partly lled cell does not coincide with its geometric centre, as shown in Fig. 2. For this reason, special stencils are introduced to compute the cell-face values and the gradients of the ow variables, making a compromise between simplicity and accuracy, which results in cost-effective relations of almost second-order accuracy. The additional terms are included in the coefcients of the general equation (3), whereas all the needed geometric quantities are computed by a pre-processing algorithm. As a result, after dening the geometry of the computational domain, the grid construction process can be performed in a fast and fully automated way. Wall boundary conditions are also set automatically to every boundary cell (e.g. cells P1P4 in Fig. 2). The above partly-lled cells (PFC) method preserves the accuracy of the boundary representation and retains the conservation property. Moreover, it does not affect the stability of the solution algorithm, while its simplicity makes it easily applicable to both 2D and 3D complex geometries. Only the solvable cells are stored and take part in the computations, as shown in the example of Fig. 1. 2.3. Impeller head and power calculations The net energy added to a unit mass of uid by the impeller can be calculated after computing the total energy of the uid at the impeller inlet and outlet (Fig. 3). Hence the uid head H12 is obtained from the ux-weighted relation:

H12 H2 H1

1 Qu

 Z  p2 p1 c2 c2 1 dq 2 qg 2g E

Fig. 1. (a) Indicative structure of the Cartesian grid, and (b) detailed view of the grid at the blade leading edge.

286

J.S. Anagnostopoulos / Computers & Fluids 38 (2009) 284289

P2
E

fluid P1
B

P4
C

free cell partly filled blocked

where k is the friction coefcient, wr and wu are the ow radial and (relative) tangential velocity components, and Dh  2b is the hydraulic diameter, with b(r) the impeller width. At the exit of the real pump impeller the ow enters into the spiral volute and decelerates. The sudden expansion losses are proportional to the absolute exit velocity of the uid, namely:

solid

dh1

k1 Qu

Z
E

c2 2 dq

P3
A

Away from the design ow rate conditions additional losses appear at the suction side, due to the impeller incidence and the inlet pipe recirculation, as well as at the volute tongue. All these losses can be included in an approximate expression of the type:

Fig. 2. Treatment of partly lled grid cells.

 2 Q dh2 k2 1 Q0

s id e

c2 r 2 wr wu w2 w n

s su re

r2
io n suct
Nu x Mu x

where Q0 is the design ow rate. The rest hydraulic losses in the inlet and outlet sections of the pump, as well as in the spiral volute, can be taken proportional to the square of the pump ow rate Q:

pre

r1 c1 w 1
r1

dh3 k3 Q 2

10

Using the above relations, the pump head H can be estimated by abstracting from the computed uid head in the impeller H12 (Eq. (4)) all the above losses:

Fig. 3. Sketch of a centrifugal pump impeller.

where c is the absolute velocity of the uid, Qu the volume ow rate through the impeller and g the gravity acceleration, while the subscripts 1 and 2 denote conditions at the impeller inlet and exit radius, respectively (Fig. 3). The integration is approximated by a summation over the radial ow rates dq at all grid cells facing the inlet or the outlet periphery of the impeller with cumulative area E. On the other hand, the power absorbed from the impeller, Nu, can be calculated from the torque Mu developed on the blades:

where ~ n is the unit vector normal to the blade surface, ~ sw the wall shear stress, b the blade angle and b the impeller width, whereas the integration covers both the pressure side and the suction side of the blade (Fig. 3). The impeller width b is a function of radius r, according to the real impeller geometry. Finally, the theoretical (Euler) head of the impeller can be obtained as:

si d e
1

r 1

H H12 dhf dh1 dh2 dh3

11

The mechanical losses at the shaft bearings, along with the disk friction power, are taken into account introducing the mechanical efciency factor, gm, hence the pump power N is:

N Nu =gm

12

Also, the volumetric losses due to the leakage ow are expressed by the volumetric efciency factor:

gq Q =Q u

13

and the overall pump efciency can be nally computed from the relation:

qgHQ
N

H g g Hu q m

14

r2

~ sw cot b b dr r ~ n p ~ r ~

r1

Hu

Nu

q g Qu

In the above expressions the values of the coefcients k1k3 and of the efciency factors gq and gm were regulated with the aid of available experimental data of the examined impeller, as will be described in chapter 3.2 below. These coefcients need to be calibrated for every new impeller, and this constitutes a drawback of the methodology. However, there are several alternative ways to perform this calibration even without measurements, like by the use of statistical data or of theoretical and empirical expressions for centrifugal pumps [19]. Also, a set of numerical results for the ow in the actual impeller geometry, obtained by applying a more accurate but costly 3D solver, can be used instead of experiments. 2.5. Numerical optimization

2.4. Pump characteristic curves Although the simulation is restricted to the impeller section, the numerical results can be used as a basis to estimate the performance of the entire pump. To achieve this, the additional hydraulic losses are properly expressed and abstracted from the head results, according to the following analysis (Fig. 3). The effect of the impeller shroud and hub surfaces can be computed using Darcys law, from the relation: In order to nd the combination of the impeller design variables that maximizes the objective value, an optimization algorithm is developed based on the unconstrained gradient approach. This method is selected after some preliminary numerical tests, which showed that the cost function (here the pump efciency) does not exhibit local maxima outside a global maximum region. However, the pump efciency is not analytic function hence a problem of non-continuity and scattering arises. The algorithm is specially designed to operate even for such discrete data, using an adjustable with trial-and-error step size along the gradient direction. Also, the gradients are computed using forward nite differences

dhf

1 Qu

Z Z
E

r2

k
r1

dr w2 w2 u r dr dq Dh 2g

J.S. Anagnostopoulos / Computers & Fluids 38 (2009) 284289

287

Nu (KW)

at the beginning, and central differences when the cost function approaches maximum, and with a variable step size. The algorithm converges very fast within the region where the cost function maximizes, although due to scattering it cannot always nd the absolute maximum. However, the fast performance allows repeating the calculations from different starting values, in order to verify the close approximation of the optimum. 3. Results 3.1. Accuracy and precision checks The accuracy of the representation of the blade geometry with the PFC method was tested at rst, along with the grid-dependency of the ow eld results. These numerical test were performed using the base-case geometry of Fig. 4, which is a periodically symmetric section of the centrifugal impeller (z = 9 blades). Then, the impeller is rotated step-by-step at small fractions (1/20) of the tangential resolution of the grid, producing a number of different grid topologies around the blade boundary line. Consequently, the scattering in the corresponding results is due to the numerical error introduced by the PFC method. The computed ow led drawn in Fig. 4 at the best efciency point (BEP) of the pump is similar for all the examined cases, however the values for the uid head and the impeller power (Eqs. (4) and (5)) are not exactly the same. The latter are concentrated in Fig. 5, for two grids: a coarse (8000 nodes) and a ne (27,000 nodes) that has two rened layers around the blade, as in the example of Fig. 1. The precision of the rened grid results is satisfactory, since the mean and maximum deviation from the mean values are of the order of 0.5% and 1.5%, respectively, for both the Nu and the H12. These deviations are roughly one-fourth of the corresponding ones with the coarse grid (Fig. 5), conrming that the PFC method preserves the accuracy of the discretization scheme. On the other hand, the differences in the mean values between the two grids represent the grid-dependency of the results, and they are again of the order of 1%, which is an adequate accuracy. Consequently, the rened grid is selected to be used for the rest calculations. 3.2. Regulation of the model The adjustable coefcients ki involved in the model Eqs. (7)(14) are regulated using the characteristic curves of a commercial centrifugal pump operating with a new impeller, constructed in the Laboratory. The impeller has nine two-dimensional (non-twisted) blades with inlet and exit diameter D1 = 70 mm, D2 = 190 mm, exit

11

10.8

10.6

10.4

Coarse grid Refined grid


50 52 54 56

10.2

H12 (m)
Fig. 5. Predicted uid head and corresponding impeller power values.

100

Head (m)

80

predictions - Hu predictions - H12 measured head H fitted head H

60

40

20 0 20 40 60 80 100 120 140

Q (m3/h)
100 measurements predictions

Pump efficiency, (%)

80

60

40

20 0 20 40 60 80 100 120

Q (m3/h)
Fig. 6. Measured and computed characteristic curves: (a) Fluid, impeller and pump head; (b) pump overall efciency.

Fig. 4. Pressure contours and ow streamlines in the standard impeller (b1 = 26, b2 = 49).

width b2 = 9 mm, and inlet and exit angle b1 = 26, b2 = 49. The blade shape is a simple circular arc of constant thickness 5 mm, and with both ends rounded (Fig. 4). The measured operation characteristic curve H Q for this pump is depicted in Fig. 6a, along with the numerical results for the uid head H12 and the impeller head Hu, obtained by Eqs. (4) and (6), respectively. The head H12 that the water acquires in the impeller is, as expected, higher than the measured pump head H,

288

J.S. Anagnostopoulos / Computers & Fluids 38 (2009) 284289

because the former does not include the additional losses in the rest pump sections. On the other hand, it is less than the theoretical head Hu, due to the mechanical losses of the ow in the impeller. Next, the efciency coefcients gq and gm were determined from statistical data, whereas using least-squares regression analysis the values of the adjustable coefcients k1, k2 and k3 were regulated so as the pump head H computed from Eq. (11) approximates well the corresponding experimental curve, as shown in Fig. 6a. Finally, the overall efciency of the pump is calculated from Eq. (14) and the resulting characteristic curve is plotted in Fig. 6b. Although that curve is not produced by tting, the agreement with the corresponding gQ measurements is very good, and this veries the consistency of the followed modelling strategy and the validity of the adjusted coefcients for this particular pump. Moreover some 3D ow effects, which become even stronger at off-design operation of the pump, are also included in the values of the above coefcients. 3.3. Optimal blade design The objective here is to maximize the best efciency value of the pump, using as free design variables the inlet and the exit blade angles. However, the exact location of the best efciency point (BEP) of the pump depends on the blade design therefore it has to be determined for each examined set of blade angles. This would need the construction of the gQ characteristic curve of the pump, by computing several points on it after corresponding runs of the evaluation algorithm. In order to accelerate the computations an alternative and much faster technique is implemented, according to which the unknown volume ow rate at the BEP is treated as an additional free design variable, and its value is obtained after the convergence of the optimization procedure. During the optimization the evaluation algorithm is capable to generate the grid and solve the ow equations for a wide range of different blade congurations without user interference. Two extreme geometry examples are plotted in Fig. 7, along with the corresponding ow eld results. The use of large blade angles in Fig. 7a results in a short and almost straight blade, which however cannot lead properly the ow and a large recirculation zone is formed at the pressure side in agreement with the theory, since the volume ow rate is much smaller than the optimal for b1 = 35. On the other hand, the blade with small angles in Fig. 7b seems to perform more smoothly, however the ow passage is much longer and narrower and consequently the mechanical losses are increased. The convergence rate of the optimization algorithm described previously in chapter 2.5 is shown in Fig. 8, for two different sets

76

Pump efficiency (%)

72

68

1 (deg), 2

Starting values (deg), Q (m3/h)

64

35, 45, 45 35, 60, 60


0 20 40 60 80

60

Evaluations
Fig. 8. Convergence history of the optimization algorithm.

of starting values. Although the latter are selected far from the optimal region, the algorithm reaches there in less than 20 evaluations, whereas nal convergence occurs in about 6080 evaluations. The whole procedure takes about 1520 h in a P4 PC. The corresponding variation of the blade angles during optimization is depicted in Fig. 9, where contour lines of constant efciency are also plotted, computed for the optimal ow rates. The two paths converge to the same region, although not exactly to the same point, due to non-continuity effects. The optimal inlet blade angle is about 21, which is more consistent with the present rotation speed (3000 rpm) and the impeller inlet diameter (the real laboratory blade had been constructed with an inlet angle of 26, in order to operate the pump effectively as turbine too). On the other hand, the exit blade angle exhibits a wider optimal region, ranging between 50 and 58, which is in agreement with theoretical and statistical data. For example, the classic blade number selection relation of Peiderer [20]:

z 6: 5

  D2 D1 b b2 sin 1 D2 D1 2

15

for b2 = 54 gives z = 8.6 ? 9 blades. The calculated characteristic curves of the pump with the optimized impeller are drawn in Fig. 10 along with the corresponding initial blade curves (shown also in Fig. 6). The maximum efciency of the pump with the optimal blade shape is about 3% higher (from 70.5% to about 73.5%). However, the BEP is shifted to smaller ow rates compared to the standard design (52 m3/h compared to 62 m3/h) because b1 is smaller, and the same is valid for the maximum head. The latter is now reasonably higher because of the larger exit angle b2 = 54 (standard b2 = 49).

Fig. 7. Computed ow eld for different blade shapes: (a) b1 = 35, b2 = 72; (b) b1 = 21, b2 = 22.

J.S. Anagnostopoulos / Computers & Fluids 38 (2009) 284289

289

60

Exit blade angle (deg)

and thus to accelerate the numerical design process. Although the suitability of the method for high specic speed pumps needs to be evaluated, its accuracy can be enhanced by introducing additional adjustable coefcients to account for the 3D effects. Acknowledgements

55

50

45

The project is co-funded by the European Social Fund (75%) and National Resources (25%) Operational Program for Educational and Vocational Training II and particularly the Program Pythagoras.

B
40 15 20 25 30 35
References
[1] Zangeneh M, Goto A, Takemura T. Suppression of secondary ows in a mixedow pump impeller by application of three-dimensional inverse design method: Part 1 design and numerical validation. ASME Trans J Turbomach 1996;118:53643. [2] Xu JZ, Gu CW. A numerical procedure of three-dimensional design problem in turbomachinery. ASME Trans J Turbomach 1992;114:54882. [3] Borges JE. A three-dimensional inverse method for turbomachinery: Part-1 Theory. ASME Trans J Turbomach 1994;112:34654. [4] Peng G, Cao S, Ishizuka M, Hayama S. Design optimization of axial ow hydraulic turbine runner: Part I an improved Q3D inverse method. Intl J Num Meth Fluids 2002;39:51731. [5] Benra FK. Economic development of efcient centrifugal pump impellers by numerical methods. World Pumps 2001(May):4853. [6] Yulin W, Jianming Y, Shuliang C. Advanced design for large hydraulic turbine runner. In: ASME/JSME FEDSM 99, San Francisco, California, Paper S-287, July 1823, 1999. [7] Garrison LA, Richard KF, Sale MJ, Cada G. Application of biological design criteria and computational uid dynamics to investigate sh survival in Kaplan turbines. In: HydroVision 2002, Portland, Oregon, July 29August 2, 2002. [8] Cao S, Peng G, Yu Z. Hydrodynamic design of rotodynamic pump impeller for multiphase pumping by combined approach of inverse design and CFD analysis. ASME Trans J Fluids Engin 2005;127:3308. [9] Asuaje M, Bakir F, Kouidri S, Rey R. Inverse design method for centrifugal impellers and comparison with numerical simulation tools. Int J Comput Fluid Dynamics 2004;18(2):10110. [10] Gehrer A, Schmidl R, Sadnik D. Kaplan turbine runner optimization by numerical ow simulation (CFD) and an evolutionary algorithm. In: 23rd IAHR symposium on hydraulic machinery and systems, Yokohama, Japan. Paper F125; 2006. [11] Miyauchi S, Kasai N, Fukutomi J. Optimization of meridional shape design of pump impeller. In: 23rd IAHR symposium on hydraulic machinery and systems, Yokohama, Japan. Paper F310; 2006. [12] Mazzouji F, Couston M, Ferrando L, Garcia F, Debeissat F. Multicriteria optimization: viscous uid analysis mechanical analysis. In: 22nd IAHR symposium on hydraulic machinery and systems, Stockholm, Sweden, June 29 July 2. Paper A04-1; 2004. [13] Kueny JL, Lestriez R, Helali A, Dmeulenaere A, Hirsch C. Optimal design of a small hydraulic turbine. In: 22nd IAHR symposium on hydraulic machinery and systems, Stockholm, Sweden, Paper A02-2, 2004. [14] Ye T, Mittal R, Udaykumar HS, Shyy W. An accurate Cartesian grid method for viscous incompressible ows with complex immersed boundaries. J Comput Phys 1999;156:20940. [15] Fadlun EA, Verzicco R, Orlandi P, Mohd-Yusof J. Combined immersedboundary nite-difference methods for three-dimensional complex ow simulations. J Comput Phys 2000;161:3560. [16] Anagnostopoulos J, Mathioulakis D. Numerical simulation and hydrodynamic design optimization of a Tesla-type valve for micropumps. IASME Trans 2005;2(6):184652. [17] Anagnostopoulos J. Discretization of transport equations on 2D Cartesian unstructured grids using data from remote cells for the convection terms. Intl J Num Meth Fluids 2003;42:297321. [18] Anagnostopoulos J, Mathioulakis D. A ow study around a time-dependent 3-D asymmetric constriction. J Fluids Struct 2004;19:4962. [19] Neumann B. The interaction between geometry and performance of a centrifugal pump. London: Mechanical Engineering Publications; 1991. [20] Peiderer C. Die Kreiselpumpen. 5th ed. Berlin: Springer-Verlag; 1961.

Inlet blade angle (deg)


Fig. 9. Variation of the design variables during optimization.

80 70

H (m), (%)

60 50 40 30 20 0 20 40 60 80 100 120
Initial blade design Optimal blade design

Q (m3/h)
Fig. 10. Pump characteristic curves with the standard and the optimal impeller.

4. Conclusions A numerical methodology for the calculation of the ow eld in a centrifugal pump impeller and the prediction of the pump performance curves is developed, regulated, and tested against experimental and statistical data, with encouraging results. The proposed ow simulation algorithm is suitable for hydrodynamic design in hydraulic pumps and turbines, thanks to the fast and fully automated generation of the Cartesian grid and the increased precision in representing irregular boundaries. The main advantage validated in the present study is that the methodology provides the ability to localize the optimal range of the free design variables at low computer cost. A more elaborate regulation and identication of the optimal design would require the simulation of the ow in the 3D impeller/runner geometry as well as in the inlet section and the casing of the machine, by performing costly numerical solutions of NavierStokes equations in three dimensions. Consequently, the much faster modelling approach proposed here can be used as a starting optimization tool in order to locate the region of maxima,

Das könnte Ihnen auch gefallen