Hidetaka Okui1

Mitsubishi Heavy Industries, LTD, 2-2-1 Shinhama Arai-Cho Takasago, Hyogo, 676-8686 Japan e-mail: hidetaka_okui@mhi.co.jp

Three-Dimensional Design and Optimization of a Transonic Rotor in Axial Flow Compressors

This paper presents a 3-D optimization of a moderately loaded transonic compressor rotor by means of a multiobjective optimization system. The latter makes use of a differential evolutionary algorithm in combination with an Articial Neural Network and a 3D Navier-Stokes solver. Operating it on a cluster of 30 processors enabled the evaluation of the off-design performance and the exploration of a large design space composed of the camber line and spanwise distribution of sweep and chord length. Objectives were an increase of efciency at unchanged stall margin by controlling the shock waves and offdesign performance curve. First designs of single blade rows allowed a better understanding of the impact of the different design parameters. Forward sweep with unchanged camber improved the peak efciency by only 0.3% with the same stall margin. Backward sweep with an optimized S shaped camber line improved the efciency by 0.6% at unchanged stall margin. It is explained how the camber line control can introduce the same effect as forward sweep and compensate the expected negative effects of backward sweep. The best results (0.7% increase in efciency and unchanged stall margin) have been obtained by a stage optimization that allows also a spanwise redistribution of the rotor ow and an increase of loading by extra ow turning. The latter compensates the loading shift induced by the backward sweep in order to reduce the inlet Mach number at the downstream stator hub. [DOI: 10.1115/1.4006668]

Tom Verstraete
e-mail: tom.verstraete@vki.ac.be

R. A. Van den Braembussche

e-mail: vdb@vki.ac.be

Zuheyr Alsalihi
e-mail: alsalihi@vki.ac.be Turbomachinery and Propulsion Department, von Karman Institute for Fluid Dynamics, Waterloose steenweg 72, 1640 Sint-Genesius-Rode, Belgium

Introduction

The recent trend to increase the capacity of modern heavy duty gas turbine compressors requires high performance transonic rotors with increasing tip Mach numbers. It is well known that the highly three-dimensional ow in those rotors interacts with a radially swept shock wave [1,2]. New sophisticated design approaches based on 3-D CFD and rig tests are therefore needed to maximize their performance. Blade sweep is known to be an effective technique to redistribute the radial loading [3,4] and has been widely used to control the shock wave in transonic fans and rotors. Backward sweep was initially investigated by Hah and Wennerstrom [2]. They developed a three-dimensional shock structure model and built an aft-swept blade, rotor6, to strengthen the sweep effect. This rotor showed a signicant improvement of the peak efciency [5,6] but a large decrease of the stall margin. After several investigations, Wadia et al. [7] concluded that the penalty on the stall margin was caused by the local increase of loading at the tip section resulting in a stronger bow shock and an accumulation of the low momentum uid at that section. In addition, no overall efciency gain was achieved with the aft-swept blade in spite of the better performance of the inboard section. Forward sweep on the contrary, demonstrated a signicant improvement in stall margin and a slightly higher peak efciency. Shock waves tend to be normal to the casing wall and in combination with the forward sweep become oblique in the blade to blade surface. This reduces the losses and stabilizes the ow in the tip region. The result of the numerical investigations of Xu and Denton [8] and Blaha et al. [9] supported this conclusion about the advantage of forward sweep, but less successful results have also been reported [10].
1 Corresponding author. Contributed by the International Gas Turbine Institute (IGTI) Division of ASME for publication in the JOURNAL OF TURBOMACHINERY. Manuscript received December 18, 2011; nal manuscript received January 30, 2012; published online March 25, 2013. Editor: David Wisler.

Ji [11] reviewed the previous studies in 2005, and concluded that sweep is a degree of freedom (DOF) that controls the spanwise matching of the blade to blade ows. It is therefore incomplete to discuss the performance and stability according to the type of sweep only. The impact of sweep depends also on the original loading distribution, which is affected by other design parameters. Several researchers indicated that the thickness distribution [6], camber distribution [12] and solidity [13] have an equally large inuence on the performance of transonic rotors as sweep and that they all need to be optimized simultaneously. Another aspect of sweep is that it often disturbs the matching with the following stage due to a radial redistribution of the loading and mass ow [7,9]. Inverse methods seem to be suited for the design of transonic fans and rotors. The loading or pressure distribution can be explicitly specied and the geometry, which satises the given design intend almost exactly, is quickly found [14]. However, the method experiences some practical difculties in establishing criteria for an optimal loading distribution and in imposing the mechanical constraints. Watanabe and Zangeneh [15] used an inverse method to design a swept transonic fan aiming to compensate the change in specic work and pressure ratio caused by the given forward sweep. They reasonably recovered the speed line and achieved an efciency improvement of 0.5% by specifying a very smooth loading distribution at the tip section. The method, however, did not include the tip clearance ow and the sweep was not optimized for the given loading distribution but explicitly specied. In this study, an efcient optimizing design system developed at the von Karman Institute [16,17] has been applied to a moderately loaded transonic rotor for heavy-duty gas turbine compressors. A multiobjective optimization for the design and off-design operation was carried out, in which the sweep, chord length and the camber distributions are controlled. By tailoring these parameters simultaneously, the loading distribution could be optimized so that both the stability and the performance were improved. A constant speed line calculation evaluated the off-design performance. MAY 2013, Vol. 135 / 031009-1

Journal of Turbomachinery

C 2013 by ASME Copyright V

Downloaded From: http://turbomachinery.asmedigitalcollection.asme.org/ on 04/04/2013 Terms of Use: http://asme.org/terms

Table 1

Specication of the baseline blade design 1.35 180.6 22 3600 1.20 0.34 0.3 kg/s/m2 rpm % height

Pressure ratio Specic ow Number of blades Corrected speed Tip relative Mach number Mid Diffusion Factor Tip clearance

Three studies have been carried out. The sweep and the chord length were optimized in the rst one. The camber distribution was added as an additional variable in the second one. The purpose of these rst cases was to validate the design method and to clarify the effect of the camber line control. Finally, a stage optimization was made to evaluate the possible impact of previous design parameters on the stage performance.

Fig. 1 Optimization algorithm

Baseline Blade Geometry

The specications of the original blade are summarized in Table 1. The baseline blade, specially designed for this study, is a typical front stage transonic rotor for heavy duty gas turbines. No lean or sweep was introduced yet. The controlled diffusion airfoil (CDA) design concept was used to dene the multiple circular arcs (MCA) blade sections at ve representative positions (10, 30, 50, 70, and 90% span). The other sections were obtained by interpolation in the radial direction. The shock wave was appropriately controlled at the tip section so that it attached to the leading edge at the design point. As a result, the adiabatic efciency of this baseline rotor, predicted by CFD, was already high. To achieve a further efciency gain by optimization was therefore a quite challenging task.

Optimization Method

Figure 1 shows the algorithm of the optimization system used for the present study. One of its advantages is the use of a metamodel, based on an articial neural network (ANN) as an interpolation tool by which the performance corresponding to the given design parameters are predicted. It reduces the large computational cost of evaluating the blade performance by 3D-CFD and; consequently, enables an increase of the number of DOF for the optimization. These advantages have been demonstrated by many complicated 3D industrial design problems [1821]. The algorithm consists of two steps. The rst one is an approximated loop with ANN predictions based on the information stored in the database. The differential evolutionary (DE) algorithm selects promising designs that are then veried in an accurate loop by means of a full 3D Navier-Stokes solver. The results of these evaluations are added to the database which allows improvements to the accuracy of the next ANN predictions. This cycle is repeated until the CFD conrms that the optimization is based on accurate ANN predictions. A more detailed description of the optimization method is given in Refs. [16] and [17]. The following subsections summarize some components and their application. 3.1 Navier-Stokes Evaluation. The performance of new blades is evaluated by means of the Reynolds averaged full 3D Navier-Stokes solver TRAF3D, which has been validated with the experimental data of Rotor67 [22]. The convection term is discretized by a central differential scheme and stabilized by Jamesons articial dissipation. Turbulence closure is by the Baldwin-Lomax model. The boundary layer on the blade surface is assumed to be fully turbulent from the leading edge. The accuracy was veried by comparing predictions of several similar geometries with MHI 031009-2 / Vol. 135, MAY 2013

in-house data. In addition, all of the efciency gains and the stall margin achieved in each optimization study have been conrmed by the ANSYS CFX solver with SST turbulence model. The stall limit was dened as the point where the ow eld becomes numerically unstable. A threshold of the residual was dened in advance, based on several numerical tests. In order to verify the choking, stall margin and peak efciency, calculations were done at seven points on a constant speed line using 30 CPUs of the VKI parallel computing system. Figure 2 shows the computational mesh used for the analysis. The structured H-type mesh has 220 53 53 points in, respectively, the stream, pitch and spanwise directions. In order to avoid the grid dependency on the shock wave, more than 100 grid lines were clustered in the front passage area where the bow shock is expected. The minimum grid spacing from the wall was adjusted to y < 5 in accordance with the requirements of the Baldwin Lomax turbulence model. The downstream grid is parallel to the expected exit ow angle to avoid articial dissipation of the wake. The dependency of the performance on these mesh parameters was carefully evaluated in advance by the verication tests. The spanwise distributions of the absolute total pressure and temperature, swirl and pitch angle were specied as inlet boundary conditions. The average static pressure was imposed at the exit and the spanwise variation is obtained from radial equilibrium. A constant speed line is obtained by modifying the back pressure, leaving the inlet conditions unchanged.

Fig. 2

Computational mesh

Transactions of the ASME

Downloaded From: http://turbomachinery.asmedigitalcollection.asme.org/ on 04/04/2013 Terms of Use: http://asme.org/terms

The spanwise variation of chord length and stacking line is parameterized by B-Spline curves to create smooth and reasonable distributions (Fig. 4). They are linked to each other by the same spanwise positions to reduce the number of variables. Sweep is dened by six parameters. S1, S2, A1 and A2 can be positive or negative corresponding to forward or backward sweep. The chord length is specied by four parameters as a fraction of the local chord length of the original blade. It was veried that this parameterization could accurately reproduce the baseline airfoil. 3.3 Problem Definition. The optimization requires the transformation of the given engineering problem into a mathematical one by dening objective and constraint functions [17]. A multiobjective optimization aiming for higher performance and stability was carried out by minimizing the following objective functions. Obj1 1:0 Peak Efficiency Obj2 0:3 Throttle Margin
Fig. 3 Parameterization of camberline

(1) (2)

3.2 Geometry Definition. New blades are derived from nondimensional 2D proles dened at the ve previously specied representative positions. They differ by the thickness and camberline distributions. In the present study, the thickness distribution has been maintained at its original value to eliminate the effect on throat area [23] and to avoid mechanical problems. Figure 3 shows the parameterization of the camber line. It is dened by the local slope angle whose change directly relates to the local loading. The current denition requires only the four variables listed on Fig. 3. The two additional control points are displaced together with the leading, trailing edge and the central points. They intend to keep the curvature in the front and rear parts unchanged in order to avoid undesirable local accelerations or decelerations. The advantages of this second order expression of the camber line are its relative simplicity, smoothness and the capacity to represent relatively complicated camber lines. The trailing edge metal angle was kept constant in the single row optimizations, to conserve the stage matching. It was a variable in the stage optimization, because the matching was then implicitly guaranteed by the stage performance.

where the throttle margin is dened as a function of the exit corrected mass ow at stall and design operating point ( mout_design/mout_stall 1). The latter is a design target and thus, explicitly specied. The optimization is governed by following constraints. Cons1 min
chok jbaseline

min

chok jdesign

0

(3)

where min is the inlet corrected mass ow. Equation (3) requires that the choking mass ow is larger than or equal to the baseline one. Cons2 mout design mout stall mout choke mout design  0 (4) guarantees that a new speed line includes the design throttle point. These denitions allow a shift of the speed line due to a small change of the mass ow induced by the swept stacking, which accelerates the optimization. After the convergence, the optimization provides a so-called Pareto-front [17]. The best result is the one that respects the stall margin (Obj2 < Obj2_baseline) with the highest peak efciency.

Fig. 4

Parameterization of sweep and chord length

Journal of Turbomachinery

MAY 2013, Vol. 135 / 031009-3

Downloaded From: http://turbomachinery.asmedigitalcollection.asme.org/ on 04/04/2013 Terms of Use: http://asme.org/terms

Fig. 5 (a) Resultant geometry of the rst optimization: spanwise variation of sweep. (b) Resultant geometry of the rst optimization: spanwise variation of chord length.

Results and Discussion

4.1 The First Optimization. The rst optimization allows only a spanwise variation of sweep and chord length. The optimized geometry, specied on Fig. 5, shows a combination of a strong forward sweep and varying chord length. The latter is reduced by 5% at 80% span but is almost back to its original value at the tip. The resulting performance curve (Fig. 6) has lower Obj1 values and respects the design requirements specied by the constraint functions. The new pressure rise curve conserves the original characteristics except for a small decrease near stall point (Fig. 6(a)). Figure 6(b) conrms the efciency gain attributed to forward sweep by previous researchers [7,8]. One observes a substantial improvement (0.3%) of the peak efciency and a small efciency drop near stall point. The latter explains the small decrease of the pressure ratio near stall point. The design point ow pattern of the new blade is compared with the one of the baseline blade on Figs. 710. These gures also include the results of the backward swept blade, which is the outcome of a second optimization and explained in the following subsection. The same color scales are used for each group of gures, so that one can directly compare the strength of the shock

wave, the level of the Mach number and the magnitude of the losses. Figure 7 is the meridional view of the static pressure contours on the blade suction surface. Since the shock wave has to be perpendicular to the outer wall, he does not follow the leading edge above the 80% span of the optimized blade (Fig. 7(b)). The strong forward swept blade cuts the bow shock into a very weak compression shock on the suction side and a weaker passage shock. This is also visible on the static pressure contours in the blade to blade surface (Fig. 8). The surface of reference is indicated by the red line on the small meridional contours in the right lower corners. At 95%Ht one observes again a passage shock in combination with a weak shock type deceleration impinging on the suction side at 25% chord length. The smaller acceleration on the suction side results in a lower Mach number approaching the leading edge. The static pressure recovery in the passage area is by a multiple shock system consisting of a weak oblique shock attached to the leading edge, followed by the passage normal shock. This multiple shock pattern reduces the local shock loss and the interaction loss with the suction side boundary layer. The ow at 70% span is almost identical to the one of the baseline blade (Fig. 8(e)).

Fig. 6

(a) Compressor map: speed line. (b) Compressor map: efciency.

031009-4 / Vol. 135, MAY 2013

Transactions of the ASME

Downloaded From: http://turbomachinery.asmedigitalcollection.asme.org/ on 04/04/2013 Terms of Use: http://asme.org/terms

Fig. 7

Meridional view of static pressure contour on suction surface

Fig. 8 Static pressure contour on blade to blade surface

Journal of Turbomachinery

MAY 2013, Vol. 135 / 031009-5

Downloaded From: http://turbomachinery.asmedigitalcollection.asme.org/ on 04/04/2013 Terms of Use: http://asme.org/terms

Fig. 9

Comparison of entropy contours at 99% height section

Fig. 10 (a) Spanwise distribution of inlet axial velocity. (b) Spanwise distribution of efciency at rotor exit.

The lower pressure difference between the suction and pressure side indicates a lower tip loading. The effect on the tip leakage vortex is illustrated by the entropy contours at 99% height (Fig. 9). The high entropy region, marked by yellow and red, indicates the tip vortex. It shows that the leakage vortex is weakened and turned in the downstream direction. The vortex no longer impinges on the adjacent blade as it does on the baseline blade. In the latter the low momentum uid perturbs the pressure side ow and leaks again to the suction side of the adjacent blade. This double leakage [24,25] contributes also to the thicker wake of the baseline blade. The spanwise distribution of the inlet axial velocity and adiabatic efciency are presented in Fig. 10. It shows that the axial velocity increases at the tip section and decreases below 50% span. The velocity increase at the tip section reduces the incidence. As a result, the shock wave moves into the passage as previously explained and the loading at the tip section is reduced. On the contrary, the decreased velocity below mid section increases the incidence and ow turning and thus, the loading in that region. 031009-6 / Vol. 135, MAY 2013

This spanwise loading redistribution is typical of forward swept transonic blades. It persists when increasing the pressure ratio, and causes a larger separation in the hub to mid region, which is responsible for the small efciency drop near stall point (Fig. 6). The reduced shock strength and the smaller leakage ow discussed in previous paragraphs, result in a considerable gain in adiabatic efciency in the tip region with only a slight decrease in the hub region. These mechanisms are in agreement with previous observations [7,8] and therefore verify the validity of present optimization system. 4.2 The Second Optimization. The second optimization allows a modication of the camber line in addition to sweep and chord length. Figure 11 shows that the optimized geometry has a relatively strong backward sweep and a slightly barreling chord (Fig. 7(c)). The later helps to reduce the mid-span loading and weakens the local shock (Fig. 8(c)). The chord length in the outward region is now increased by 3% of its original value while in the forward swept blade it was reduced with a minimum at 80% span. The Transactions of the ASME

Downloaded From: http://turbomachinery.asmedigitalcollection.asme.org/ on 04/04/2013 Terms of Use: http://asme.org/terms

Fig. 11 (a) Resultant geometry of the second optimization: spanwise variation of sweep. (b) Resultant geometry of the second optimization: spanwise variation of chord length.

Fig. 12 (a) Chordwise camber line distribution (90%Ht). (b) Difference of peak camber location.

comparison of both results indicates that the camber line control has a signicant inuence on the optimum stacking line. Figure 12(a) shows a comparison of the new camber line at 90% height with the original one. It is noted that the optimized blade has an S-shaped camber line with negative camber in the front chord area and the peak camber position moved 18% upstream. This position roughly corresponds to the throat, which is just upstream of the location where the passage shock usually occurs. Figure 12(b) shows the spanwise distribution of the shift in peak camber position between the baseline and the optimized blade. One observes a gradual change from the 18% forward shift (neg. abscissa) at the blade tip, to a small backward shift (pos. abscissa) below 50% span. Its effect on the performance of the new blade is discussed later. The corresponding performance map is compared to the previous one in Fig. 13. It includes also the results of the rst optimization for comparison. The shape of the original speed line, which is important for the stage matching [12], is better conserved. The 0.6% increase in peak efciency is larger than in the rst optimization and sustained over the entire operation range in spite of the strong backward sweep. Contrarily to the previous case, this blade is tip critical for stability because of the backward sweep. The conservation of the original stall margin must therefore be attributed to the camber line control. This result shows that the optimum stacking distribution depends on the loading distribution and that the optimizer can nd it to achieve the best combination of performance and stability. The tip Mach number distributions at the design operation point are shown on Fig. 14 with the blade center of gravity superimposed. They illustrate the impact of the camber-line on Journal of Turbomachinery

stability. Figure 14(a) shows a comparison between the baseline blade and the second optimized one. The leading edge position of each blade is indicated by an arrow. It shows that the bow shock of the optimized blade is attached at the leading edge and that the passage shock is more oblique and incidents on the suction side at the same position as the baseline blade. Figure 14(b) shows the comparison between the baseline blade and one with the optimized backward sweep and chord length, but with the original camber line distribution. One observes that, in the latter case, the bow shock is completely detached and normal to the suction side. This normal shock moves further upstream with increasing back pressure and stall occurs sooner due to too high positive incidence. This explains the reduction in operating range often observed with back swept blades. The backward swept blade with the original camber distribution does not show any performance improvement. This is different from what occurs with the optimized camber where, at increasing backpressure, the oblique passage shock rst increases strength by increasing the shock angle, before it detaches from the leading edge. Therefore, it is obvious that the camber-line control stabilizes the blade by compensating for the forward shift of loading by backward sweep. The experimental result of Low and Wadia [6] showed a similar change of shock position and stall margin by a spanwise shift of the maximum thickness location. In present study, a similar change in suction side shape is obtained by modifying the camberline (Fig. 12(b)). This has a stronger inuence on the shock position without the negative effect on efciency due to the increase of the wedge angle. MAY 2013, Vol. 135 / 031009-7

Downloaded From: http://turbomachinery.asmedigitalcollection.asme.org/ on 04/04/2013 Terms of Use: http://asme.org/terms

Fig. 13 (a) Compressor map: speed line. (b) Compressor map: efciency.

In addition, the S-shaped tip section has smaller ow acceleration upstream of the bow shock (Fig. 8(c)). This reduces the leading edge relative Mach number and results in a weaker shock. It requires a larger subsonic diffusion in the downstream part but the corresponding diffusion losses are normally lower than the equivalent shock losses. Consequently, the losses at the tip section of the second optimized blade are also reduced in spite of the backward sweep (Fig. 10). As one can see from the pressure difference in the Fig. 8(c), the optimized S-shaped camberline reduces the loading in the front part. It displaces the leakage ow to downstream (Fig. 9(c)) which has favorable impact on both the stability and performance. It is noted that the S-shaped blade has a reacceleration downstream of passage shock (Fig. 8(c)), resulting from the smaller throat section than the original cambered blades. A local decrease of the throat section is usually acceptable if it can be compensated by an increase at the other spanwise positions. In present optimization the overall choking mass ow is automatically guaranteed by the constrain (Eq. (3)). In addition to the improvement of the tip region ow, the shock front is inclined at mid span due to the backward sweep and the extended chord locally reduces the peak Mach number. As a result, the passage shock is signicantly weakened and has partly disappeared in the mid span region (Fig. 8(f)).

The optimized blade therefore, prots from the positive effect of the backward sweep without the disadvantages and the efciency gain is achieved over almost the entire span (Fig. 10). 4.3 The Stage Optimization. Before carrying out the last optimization, stage calculations have been made with the rotor geometries obtained by the previous optimizations. The performance maps are shown on Fig. 15 together with the full stage optimization results, explained later. The two optimized rotors still demonstrate certain advantages over the baseline stage and the trend of the pressure rise curves and the stall margins are roughly consistent with the rotor only results. However the efciency gain achieved with the Opt. 2 rotor (back swept blade) is now 30% lower than its rotor only value. Table 2 summarizes the performance changes at the rotor peak efciency point. It shows that the single row efciency gain of the Opt. 2 (backward swept rotor) is still the highest but that the stage efciency is badly hurt by the increased stator losses. The radial shift of mass ow at the rotor peak efciency point due to the backward sweep is considerably larger than the one with forward sweep (Fig. 10). As a result, a rotor-stator mismatching occurs and the efciency gain is reduced. Figure 16 shows the design point absolute Mach number distribution at the stator inlet and exit. It is found that the backward

Fig. 14 (a) Superimposed tip Mach number contours (baseline versus backward sweep with optimized camber). (b) Superimposed tip Mach number contours (baseline versus backward sweep with baseline camber).

031009-8 / Vol. 135, MAY 2013

Transactions of the ASME

Downloaded From: http://turbomachinery.asmedigitalcollection.asme.org/ on 04/04/2013 Terms of Use: http://asme.org/terms

Fig. 15 (a) Compressor map: speed line. (b) Compressor map: efciency. Table 2 Performance at the rotor peak efciency point (difference against the baseline blade case) Opt.1 DRotor Efciency DStage Efciency DStator Loss 1 0.37%. 0.37% 0.01% Opt.2 0.52% 0.18% 0.19%

Fig. 17 Difference of rotor exit ow angle

Fig. 16 Stator inlet and outlet Mach number (comparison of single row optimization geometries)

swept blade has the highest inlet Mach number in the hub region. The latter results from the increased loading in that region by the swept stacking. The higher inlet Mach number causes the higher total pressure loss on the stator suction side, which is responsible for the smaller gain in stage efciency. In order to avoid such a matching problem, the rotor geometry has been optimized again, considering the full stage performance and allowing for a variation of the rotor exit metal angle. The steady multirow calculations have been made with the mixing plane technique at the rotor-stator interface. Details of the newly Journal of Turbomachinery

optimized blade are not shown because, apart from a change of the rotor exit metal angle, the resulting geometry is almost identical to the S-shaped backward swept blade (Fig. 11). However, the stage efciency showed a large (0.7%) improvement against the baseline blade while maintaining the stall margin (Fig. 15). Figure 17 shows the difference between the exit ow angle of the optimized and the baseline blade. Positive values indicate more turning; hence a decrease of the exit ow angle. One observes that the exit ow angle is decreased over the entire span. This does not mean an increase of stator loading or mass ow at the design point because the operating point is dened by the corrected mass ow at the stage exit (see the design throttle line in Fig. 15(a)). This implies that, without a signicant change in the stator losses, the corrected mass ow and thus also the averaged Mach number at the stator inlet are approximately conserved even when the speed line is signicantly shifted. However, the optimized change of rotor exit angle modies the spanwise loading distribution. The variation of the outlet ow angle for a constant increase of enthalpy rise along the span is superposed on Fig. 17 for comparison. The difference between these curves characterizes the spanwise redistribution of the loading with the rotor exit ow angle. It is found that the optimized MAY 2013, Vol. 135 / 031009-9

Downloaded From: http://turbomachinery.asmedigitalcollection.asme.org/ on 04/04/2013 Terms of Use: http://asme.org/terms

Conclusions

It has been shown that a three-dimensional optimization method is a powerful tool to nd out what changes of the camber line, chord length and stacking distributions will result in performance improvements and to understand the mechanisms that govern the complex ow pattern in transonic compressors. The following conclusions about those mechanisms could be drawn: (1) The best results in terms of rotor efciency and stall margin have been obtained with a strong backward swept stacking line in combination with an S-shaped camber line and a barreling chord length. The 0.6% improvement in the rotor peak efciency with unchanged stall margin results from a decrease of losses over the outer part of the blade span. (2) The well-known negative effects of backward sweep on stability and efciency at the tip section can be compensated by an optimum control of the chord wise blade loading. A S-shaped camber line reduces the suction side Mach number, makes the shock at the tip oblique in the blade to blade surface and shifts the loading towards the aft part. This reduces the shock and tip leakage losses and enhances the stability. A forward shift of the spanwise peak camber position near the tip section introduces a forward sweep like effect which makes the shock at the tip attached at the leading edge. (3) The strong backward sweep in combination with the barreling chord length reduces the loading in the mid span region and weakens the leading edge shock. (4) The stage with the backward swept blade, obtained by the rotor only optimization, did not result into an equivalent improvement of the stage efciency because of the increased stator losses due to an off design operation resulting from a strong radial shift of mass ow induced by the backward sweep. (5) The latter can be avoided by a full stage optimization allowing also a change of rotor exit ow angle. The corresponding modication of the spanwise loading reduces the stator loss and results in a gain of stage efciency that exceeds the one for the rotor only at unchanged stall margin.

Fig. 18 Stator inlet and outlet Mach number (comparison of single row and stage optimization geometries)

References
Fig. 19 Static pressure contours at 90% height
[1] Prince, D. C., 1980, Three-Dimensional Shock Structures for Transonic/Supersonic Compressor Rotors, J. Aircr., 17, pp. 2837. [2] Hah, C., and Wennerstrom, A. J., 1991, Three-Dimensional Flow Fields inside a Transonic Compressor with Swept Blades, ASME J. Turbomach., 113, pp. 241251. [3] Yamaguchi, N., Tominaga, T., and Hattori, S., 1991, Secondary-Loss Reduction by Forward-Skewing of Axial Compressor Rotor Blading, International Gas Turbine Congress in Yokohama. [4] Sasaki, T., and Breugelmans, F., 1997, Comparison of Sweep and Dihedral Effects on Compressor Cascade Performance, ASME Paper No. 97-GT-2. [5] Law, C. H., and Wadia, A. R., 1993, Low Aspect Ratio Transonic Rotors: Part 1: Baseline Design and Performance, ASME J. Turbomach., 115, pp. 218225. [6] Law, C. H., and Wadia, A. R., 1993, Low Aspect Ratio Transonic Rotors: Part2: Inuence of Location of Maximum Thickness on Transonic Compressor Performance, ASME J. Turbomach., 115, pp. 226239. [7] Wadia, A. R., Szucs, P. N., and Crall, D. W., 1998, Inner Workings of Aerodynamics Sweep, ASME J. Turbomach., 120, pp. 671682. [8] Denton, J. D., and Xu, L., 2002, The Effects of Lean and Sweep on Transonic Fan Performance, ASME Paper No. GT2002-30327. [9] Blaha, C., Kablitz, S., Hennecke, D. K., Schmidt-Eisenlohr, U., Pirker, K., and Haselhoff, S., 2000, Numerical Investigation of the Flow in an Aft-Swept Transonic Compressor Rotor, ASME Paper No. 2000-GT-0490. [10] Beneini, E., and Biollo, R., 2006, On the Aerodynamics of Swept and Leaned Transonic Compressor Rotors, ASME Paper No. GT2006-90547. [11] Ji, L., Chen, J., and Lin, F., 2005, Review and Understanding on Sweep in Axial Compressor Design, ASME Paper No. GT2005-68473. [12] Medd, A. J., Dang, T. Q., and Larosiliere, L. M., 2003, 3D Inverse Design Loading Strategy for Transonic Axial Compressor Blading, ASME Paper No. GT2003-38501. [13] Beneini, E., 2004, Three-Dimensional Multi-Objective Design Optimization of a Transonic Compressor Rotor, J. Propul. Power, 20(3), pp. 559565.

blade decreases the work at the hub relative to the one at the mid span and tip. Although the total pressure globally increases, it locally decreases at the hub and hence also the Mach number. In addition, the increased work in the mid region pulls more ow into the center of the blade, which further reduces the axial velocity in hub region. This is conrmed on Fig. 18 showing that the stage optimized blade indeed has a lower inlet Mach number at the hub; hence decreases the stator total pressure loss and improves the stage efciency (Fig. 15). It can therefore, be concluded that the stage optimization controlled the rotor exit ow angle to compensate the negative effect of the backward swept blade on the stage matching. In addition, since the pressure recovery in the tip area is achieved by the combination of a shock wave followed by a subsonic diffusion, the reduction of the exit metal angle increases the rear camber and the extra subsonic diffusion further weakens the leading edge shock. This is conrmed by the blade to blade static pressure contours at the tip of the stage optimized blade, compared in Fig. 19 with the ones obtained by the single row optimized S-shaped blade (in stage operation). It appears that the bow shock became even more oblique. 031009-10 / Vol. 135, MAY 2013

Transactions of the ASME

Downloaded From: http://turbomachinery.asmedigitalcollection.asme.org/ on 04/04/2013 Terms of Use: http://asme.org/terms

[14] Demeulenaere, A., and Van den Braembussche, R. A., 1998, Three-dimensional Inverse Method for Turbomachinery Blading Design, ASME J. Turbomach., 120(2), pp. 247255. [15] Watanabe, H., and Zangeneh, M., 2003, Design of the Blade Geometry of Swept Transonic Fans by 3D Inverse Design, ASME Paper No. GT2003-38770. [16] Pierret, S., 1999, Design Turbomachinery Blade by Means of the Function Approximation Concept Based on Articial Neural Network, Genetic Algorithm and the Navier-Stokes Equations, Ph.D. Thesis, von Karman Institute, Belgium. [17] Verstraete, T., 2008, Multidisciplinary Turbomachinery Component Optimization Considering Performance, Stress and Internal Heat Transfer, Ph.D. thesis, von Karman Institute, Belgium. [18] Verstraete, T., Alsalihi, Z., and Van den Braembussche, R. A., 2010, Multidisciplinary Optimization of a Radial Compressor for Microgas Turbine Applications, ASME J. Turbomach., 132(3), p. 031004. [19] Amaral, S., Verstraete, T., Van den Braembussche, R. A., and Arts, T., 2010, Design and Optimization of the Internal Cooling Channels of a HP Turbine Blade Part1: Methodology, ASME J. Turbomach., 132(2), p. 021013.

[20] Verstraete, T., Amaral, S., Van den Braembussche, R. A., and Arts, T., 2010, Design and Optimization of the Internal Cooling Channels of a HP Turbine Blade Part 2: Optimization, ASME J. Turbomach., 132(2), p. 021014. [21] Joly, M., Verstraete, T., and Paniagua, G., 2010, Attenuation of Vane Distortion in a Transonic Turbine using Optimization Strategies, Part I Methodology, Part II Optimization, ASME Paper Nos. GT2010-22370 and GT201022371. [22] Arnone, A., 1994, Viscous Analysis of Three-Dimensional Rotor Flow Using a Multigrid Method, ASME J. Turbomach., 116, pp. 435445. [23] Wadia, A. R., and Copenhaver, W. W., 1996, An Investigation of the Effect of Cascade Area Ratios on Transonic Compressor Performance, ASME J. Turbomach., 118, pp. 70770. [24] Sirakob, B. T., and Tan, C. S., 2002, Effect of Upstream Unsteady Flow on Rotor Tip Leakage Flow, ASME Paper No. GT2002-358. [25] McNulty, G. S., Decker, J. J., Beacher, B. F., and Khalid, S. A., 2003, The Impact of Forward Swept Rotors on Tip-Limited Low-Speed Axial Compressors, ASME Paper No. GT2003-38827.

Journal of Turbomachinery

MAY 2013, Vol. 135 / 031009-11

Downloaded From: http://turbomachinery.asmedigitalcollection.asme.org/ on 04/04/2013 Terms of Use: http://asme.org/terms