Proceedings of ASME Turbo Expo 2012

GT2012
June 11-15, 2012, Copenhagen, Denmark

GT2012-69709

PERFORMANCE LIMITS OF AXIAL COMPRESSOR STAGES

D. K. Hall, E. M. Greitzer, C. S. Tan

Gas Turbine Laboratory
Massachusetts Institute of Technology
Cambridge, Massachusetts 02139
Email: [email protected]

ABSTRACT u Velocity
This paper presents a framework for estimating the upper α Absolute flow angle
limit of compressor stage efficiency. Using a compressor stage β Blade-relative flow angle
model with a representative design velocity distribution with tur- η Stage efficiency
bulent boundary layers, losses are calculated as the sum of se- ηpoly Component polytropic efficiency
lected local irreversibilities, rather than from correlations based ηth Cycle thermal efficiency
on data from existing machines. By considering only losses that Λ Stage reaction
cannot be eliminated and optimizing stage design variables for Φ Viscous dissipation rate
minimum loss, an upper bound on stage efficiency can be deter- φ Flow coefficient (= ux /ωr)
mined as a function of a small number of stage design parame- ψ Stage loading coefficient (= ∆ht /(ωr)2 )
ters. The impact of the stage analysis results are evaluated in the
σ Solidity
context of gas turbine cycle performance. The implication from
ρ Density
the results of the stage level and cycle analyses is that compres-
θ ∗ Boundary layer kinetic energy thickness
sor efficiency improvements that result in substantial increases
in cycle thermal efficiency are still to be realized. τ Clearance gap height
ω Rotor angular velocity

NOMENCLATURE Subscripts
AR Aspect ratio e Boundary layer edge quantity
CD Dissipation coefficient t Stagnation quantity
Cs Camber length
D Diffusion factor
h Enthalpy INTRODUCTION
` Streamwise coordinate In this paper, we provide estimates of the limits on axial
ṁ Mass flow rate compressor stage efficiency. The overall context is in the de-
p Pressure velopment of a framework to assess the potential for contribu-
r radius tions to gas turbine engine performance. The estimates are based
Re Reynolds number on physical principles that govern irreversible processes within
s Entropy a turbomachine, rather than extrapolation of available data. Al-
T Temperature though analyses of this type may exist in a proprietary context, to

1 Copyright © 2012 by ASME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 03/20/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 the authors’ knowledge no such estimates exist in the open liter- h
ature. A central aim in the presentation of the work, therefore, is
to provide a forum for discussion of opportunities for increasing
∆s
efficiency in terms of specific processes.
It is useful at the outset to describe the scope and purpose of 2 ht2 − ht2s
the work, because that frames many of the choices about which 2s
effects to include and the level of detail needed in the analyses

∂h
and computations. Previous work on compressor efficiency can ∆ht ∂s ≈ Tt2
be roughly divided into two types of approach. The first is con-
cerned with estimates of the efficiency of real or proposed stages.
A notable example of this is given by the several papers from
Smith and his co-workers [1] [2] [3]. Another recent example is 1
by Wright and Miller [4]. In these treatments, analyses and corre- s
lations are developed for stage performance, with modifications
made based on data from test rigs and engines. The resulting FIGURE 1:IRREVERSIBLE COMPRESSION PROCESS
methodology gives designers an excellent assessment of the po- THERMODYNAMICS
tential performance for state-of-the-art machines, but it does not
provide information on the upper bounds of compressor perfor-
mance. APPROACH
The second approach focuses on the fundamental irre- The issue of determining a meaningful upper bound on stage
versible flow phenomena responsible for losses in efficiency. A performance is framed by the metrics used to define perfor-
landmark in this category of analysis is Denton’s [5] comprehen- mance, the losses included in the analysis, and the assumptions
sive treatment of individual loss mechanisms, which lays out pro- made about the flows through the machine.
cedures for estimating various losses using first-principles fluid A key concept is the link between efficiency and entropy
dynamic analysis. Other analyses of this type, by Storer and generation. For small pressure rises compared to the inlet level,
Cumpsty [6] and Yaras and Sjolander [7], present tip leakage the stage efficiency can be expressed in terms in the entropy gen-
loss models for compressor and turbine stages, respectively, and eration and the stagnation enthalpy rise (see Figure 1):
Young and Wilcock [8] have provided a model for losses due to
turbine cooling flows and other flow injection processes. The
ht2s − ht1 Tt2 ∆s
common thread in the quantification of losses in these treatments η= ≈ 1− . (1)
is control volume analyses of the phenomena of interest, enabling ht2 − ht1 ∆ht
loss estimates based on models of irreversible processes that do
not rely on information about flow details. Equation (1) shows that entropy increases through irreversibility,
The present work fits into this second type of approach. We or “lost work,” are linked directly to decreased efficiency. We
seek to determine the upper limit of stage efficiency through es- thus focus on the mechanisms for entropy generation.
timates of the minimum loss arising from irreversible flow phe- The following sources of entropy generation represent irre-
nomena that cannot be eliminated. The paper is organized as versibilities inherent in flow through compressors; while they can
follows. We first provide an overview of the ground rules un- be reduced, they cannot be eliminated:
der which this study has been carried out, i.e., definition of the
1. viscous dissipation near solid surfaces (within boundary lay-
losses considered and assumptions included in the model. It will
ers) on blades, vanes, and end walls,
be seen that there are a number of choices to be made concern-
2. mixing of wakes downstream of blade rows,
ing what mechanisms should be included and what should be
3. mixing of the leakage flow over the rotor tip gap with the
omitted. We present the underlying views that have led to the
main flow.
choices made because, in a very real sense, this is the crux of the
approach and the most difficult part. Second, we present a com- These are the only loss sources considered here. The losses that
pressor stage model, consisting of a generic geometry and veloc- are not included, most notably losses due to shocks and three-
ity distributions. This in employed with the goal of developing dimensional effects such as corner separation and stator fixing
useful estimates of stage efficiency as a function of a manage- configuration, are neglected. This is part of the basic choice al-
able number of stage input parameters. Finally, efficiencies for luded to previously, and thus deserves some comment. Some of
optimized stages representing a range of aero engine compres- these losses are dependent on the detailed design of the stage;
sor stages are presented, and the overall benefit in terms of gas for example, shock losses can be modeled as a function of Mach
turbine thermal efficiency is evaluated. number, but depending on the design, this loss may be reduced or

2 Copyright © 2012 by ASME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 03/20/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 eliminated through selection of blade Mach number. For others
the complexity arising from modeling them was deemed unjusti- s
fied compared to the gain in fidelity they might provide. Where β1 β2
ux ux ux
possible, however, estimates of the magnitudes of these “miss- α1 α2 α3
ing” losses are given in the subsequent discussion.
The loss magnitudes for the stages examined are deter- ωr c
ωr
mined from a combination of two-dimensional models. Profile
(a) Mean-line profile geometry
losses over the span of rotor blades and stator vanes are assumed
to be captured by a mean-line analysis. Tip leakage and end τ
wall losses are determined from two-dimensional control volume ro
analyses. Three-dimensional flow features (e.g., passage and tip
leakage vortices) are not modeled, consistent with the choices Rotor Stator
made about which losses to include.1
The flow is taken as incompressible. Compressibility ef- ri
fects can decrease performance through increased dissipation in
boundary layers or through the presence of shocks [9] [10]. Cor-
relations based on test data by Koch and Smith [1], however, Centerline
indicate that increased boundary layer losses at stage inlet Mach (b) Meridional geometry
numbers as high as 1.5 result in losses in efficiency of less than
one percent; shock losses were also shown to be small relative FIGURE 2: COMPRESSOR STAGE GEOMTRY
to profile losses at Mach numbers up to 1.2. Because the magni-
tudes of these losses are not large, may depend on the detailed
design of the stage (e.g., airfoil loading distribution), and are ratio ri /ro , both of which are assumed constant. Axial spacing
generally most important for the front few stages of a machine, of the blade rows is not considered, since it may be set based on
neglecting them is consistent with our approach to estimating stability or mechanical constraints; the result is an underestima-
maximum efficiency; the incompressible case provides an up- tion of the end wall loss approximately equal to the ratio of end
per bound on the attainable efficiency while eliminating Mach wall area in the gap to end wall area in the blade passage. The
number as a design parameter. gap between rotor blade tip and the engine casing is character-
Boundary layers are taken to be steady and turbulent. In ized by the gap-to-height ratio τ/(ro − ri ). No attempt is made
practice, blade boundary layers are unsteady with a portion of to model stator clearance loss, leading to an optimistic result for
laminar flow [9] [11] [12]. Further, the possible unsteady recov- stage clearance losses.
ery in wake mixing can result in decreased wake loss [13]. We The compressor is taken to have repeating stages, and
have neglected these effects; the result is an underestimation of as a result, the mean-line velocity triangles are characterized
the efficiency limit, but this is deemed acceptable in view of the by the flow coefficient φ = ux /ωr, stage loading coefficient
number of parameters needed to describe the unsteady transition ψ = ∆ht /(ωr)2 , and specification of either the absolute rotor in-
process and wake recovery. The analysis can thus be regarded as let flow angle α1 or stage reaction Λ. The velocity triangles in
conservative, but it will be seen that even with these assumptions, Figure 2a can be described in terms of these parameters using the
there is substantial potential for increases in efficiency. Euler turbine equation.

STAGE DESCRIPTION ψ = φ (tan α2 − tan α1 ) (2)

A stage consists of a row of rotating blades (rotor) followed
by a row of stationary vanes (stator). The nomenclature is given In determining the dissipation on the blades and vanes,
in Figure 2. For the purposes of determining blade surface area, we make use of the velocity distribution used by Dickens and
airfoils are modeled as having negligible thickness and circular Day [14], with linear changes in velocity from leading edge to
arc camber lines. The blade spacing is characterized by the so- trailing edge on both sides of the airfoil. The initial velocity
lidity σ = c/s, where c is the blade chord, and s is the spacing near the leading edge (i.e., following a rapid acceleration after
between blades at the mean line radius. The annular geometry is the leading edge stagnation point) is taken as equal to the incom-
characterized by the aspect ratio AR = (ro − ri )/c and hub-to-tip ing blade-relative velocity uin plus or minus some increment ∆u,
and the velocity at the trailing edge is equal (on both sides) to the
blade-relative velocity just downstream. The inlet and exit ve-
1 One view of this assumption is that future designs will mitigate, or perhaps locities are set by the velocity triangles, and ∆u is determined by
eliminate, losses due to three-dimensional effects.

3 Copyright © 2012 by ASME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 03/20/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 1.5

Mixing
1.0
ue /u1

0.5

Linear velocity distribution

Computed velocity distribution FIGURE 4: WAKE MIXING DOWNSTREAM OF A BLADE
0 ROW
0 0.5 1.0 1.5
ℓ/Cs
Boundary layer dissipation
The viscous dissipation is estimated from two-dimensional
FIGURE 3: MODEL ASSUMED VELOCITY DISTRIBU- turbulent boundary layer theory as the product of local dissipa-
TION COMPARED TO CALCULATED CASCADE VELOC- tion per unit area and blade wetted area. This approach has been
ITY DISTRIBUTION used by Denton to good effect (see reference [5] for additional
comments).
The dissipation per unit depth, Φ0 , for a two-dimensional
satisfying conservation of angular momentum through the blade
boundary layer can be expressed in terms of integral boundary
row given the blade spacing.
layer properties at a streamwise location, ` [16].
Figure 3 shows a comparison of this velocity distribution
to a velocity distribution computed using a 2D viscous-inviscid Z `
1
cascade flow solver [15] for a representative compressor airfoil Φ0 (`) = ρe u3e CD d` = ρe (`)ue (`)3 θ ∗ (`), (4)
cascade. Although the linear velocity distribution is a simple es- 0 2
timate, it captures the peak overspeed and the slope of the suction
In Equation (4), CD is the dissipation coefficient, and θ ∗ is the
side deceleration, which have the largest effect on profile losses.
boundary layer kinetic energy thickness. If the surface velocity
The end wall velocity is assumed to increase linearly from
distribution ue (`) is known, an integral boundary layer method
suction side to pressure side, with the rotor casing velocity taken
provides a rapid and useful way to determine the evolution of
as the vector sum of the velocity in the blade-relative frame and
boundary layer quantities and thus loss. For this, closure correla-
the rotor tip speed.
tions as described in [17] are needed to relate CD to the momen-
tum thickness θ and boundary layer shape factor H. The use of
LOSS MECHANISMS the dissipation coefficient is advantageous because its variation
As in Equation (1), the efficiency is directly related to the with shape factor is much less than the skin friction coefficient.
irreversible increase of entropy. For incompressible, adiabatic Over the range 300 < Reθ < 1000 and 1 < H < 2, the dissipation
flow, the entropy rise is related to the mechanical dissipation, Φ, coefficient can be approximated as constant, and the loss per unit
as surface area is proportional to the cube of the velocity [5] [17].
ZZZ ZZ


ṁTt2 ∆s = Φ = τ · ∇ · udV (3) Φ ≈ 0.002 ρe u3e dA (5)

Equation (5) is used for estimates of dissipation in the hub and

In Equation (3), the term τ is the stress tensor, and u is the ve-
casing boundary layers.
locity vector. The dissipation is the rate of change of mechanical
energy to heat due to the action of viscous stresses. In the analy-
sis, it is not necessary to perform the volume integral of Equation Wake mixing dissipation
(3) because, if a control volume is appropriately chosen, the dis- The downstream wake mixing losses can be evaluated using
sipation can be determined in terms of the mechanical energy the control volume shown in Figure 4, which is bounded by pe-
flux at the control volume boundaries [16]. riodic streamlines and extends from the blade row trailing edge

4 Copyright © 2012 by ASME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 03/20/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 us TABLE 1: STAGE EFFICIENCY CALCULATION INPUTS

Flow coefficient φ = ux /ωr independent

us Stage loading coefficient ψ = ∆ht /(ωr)2 independent
ζ
Reynolds number Re= ux c/ν design parameter
Aspect ratio AR = (ro − ri )/c design parameter
d`
Hub-to-tip ratio ri /ro design parameter
up
Gap-to-height ratio τ/(ro − ri ) design parameter
Inter-stage swirl α1 design variable
Solidity σ = c/s design variable
FIGURE 5: TIP LEAKAGE FLOW OVER SMALL CHORD
LENGTH d`
STAGE EFFICIENCY
Parametric dependence and optimization
to a downstream location where the flow properties are uniform. Combining Equations (1) and (3), the loss in stage efficiency
Given the conditions at the trailing edge, the mixing dissipation is expressed as the sum of the dissipation due to the different loss
can be found by applying conservation of mass and momentum. sources divided by the stage work.

∑Φ
Leakage flow dissipation η = 1− (7)
ṁ∆ht
The third loss source considered is that due to mixing of tip
leakage flow with the main flow on the suction side of the rotor
blade tip. Consistent with the ideas expressed above, we use a The dissipation terms can be found in terms of the input param-
control volume analysis [5] based on the following assumptions: eters listed in Table 1. The details of the calculation are given in
Appendix A, but we can state the parametric dependence as
1. the flow through the clearance gap prior to mixing is isen-
tropic and driven by the static pressure difference across the
η = F (φ , ψ, Re, AR, ri /ro , τ/(ro − ri ), α1 , σ ) . (8)
blade, and
2. the clearance flow is small relative to the main flow and mix-
ing loss occurs at the gap exit conditions.2 Calculations of efficiency over a range of inputs show α1 and
σ take on values that maximize efficiency given values for the
Figure 5 shows the leakage flow over a portion of the blade tip d`. other inputs. These parameters are thus considered as design
The loss is modeled as mixing of an incremental mass injection, variables, which can be solved for, rather than specified, by op-
d ṁ, with the main flow, both having velocity us and a mixing timization for maximum η. The parameters Re, AR, ri /ro , and
angle ζ . The mixing loss is equal to the kinetic energy lost by τ/(ro − ri ) will, in general, be fixed by constraints additional to
the leakage flow as it equilibrates with the suction side flow. aerodynamic performance. They can therefore be considered as
the design parameters, which are fixed inputs for a given design
1 (i.e., for a certain stage of a machine operating at a specific set
dΦ = u2s [sin2 ζ + (1 − cos ζ )2 ]d ṁ (6) of conditions). The remaining inputs, φ and ψ, are independent
2
variables. Equation (8) can then be rewritten as
The total leakage loss is obtained by integrating Equation (6)
over the chord length, with the local distributions of velocity, [ηmax , α1∗ , σ ∗ ] = F ∗ (φ , ψ; [Re, AR, ri /ro , τ/(ro − ri )]) , (9)
u(`), mixing angle, ζ (`), and leakage mass flow, d ṁ(`), deter-
mined by the blade pressure and suction side velocity distribu- with results expressed as contours of maximum efficiency and
tions. optimal design variables as a function of φ and ψ for a given
combination of the bracketed terms.
While conceptually, the optimization implied in Equation
2 This assumption appears to hold well for compressors, but not for turbine (9) is straightforward, practical implementation is more challeng-
leakage flows in which the vortex dynamics play a larger role [18] ing. For moderate to high stage loadings, at least one blade row

5 Copyright © 2012 by ASME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 03/20/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 TABLE 2: MAGNITUDES OF LOSSES AT PEAK EFFI-
0.8 CIENCY
Stage loading coefficient ψ

0.7 Loss source ∆η (%)

0.6 Rotor blade boundary layer 1.00

0.5 Rotor wake mixing 0.23

Rotor end wall boundary layer 0.96
0.4
Rotor tip clearance 0.41
0.3
Stator vane boundary layer 1.00
0.2 Stator wake mixing 0.23
0.1 Stator end walls 0.73
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
Total 4.55
Flow coefficient φ

FIGURE 6: EFFICIENCY OF REPRESENTATIVE COMPRES-

SOR STAGE; CONTOUR INCREMENT: 1%, PEAK CON- Figure 6 shows contours of efficiency versus flow coefficient
TOUR: 95% and stage loading coefficient, and we make three observations
about the results. First, the peak efficiency is 95.4%. This can
be compared to stage efficiencies of 92% quoted in the open lit-
of the optimal stage is operating near or at the separation limit erature [20], although we recognize that much of the relevant
(i.e., increases in the turning or blade spacing result in sepa- information on this point resides within industry. Second, there
ration). The functional dependence of efficiency on the design exists a region of high loading and low flow coefficient (the blank
variables at this limit is such that multiple local maxima in effi- region in the upper left-hand corner of Figure 6) where attached
ciency occur. However, because the efficiency is not sensitive to flow cannot be maintained over the entire blade profile. Even
changes in blade spacing and stage reaction near the separation with the solidity set by a fixed diffusion factor, beyond a certain
limit, we can cast the optimization in terms of stages with speci- loading, the suction side deceleration is too large, and the flow
fied diffusion factor, D, and stage reaction, Λ. More specifically, separates. Third, the peak efficiency occurs with flow coefficient
for the blade velocity profile distribution used, a reaction of 0.5 in the range of 0.4-0.6 and stage loading coefficient between 0.2
minimizes the overspeeds and suction side deceleration and cor- and 0.3. In this range, values of de Haller number (uout /uin ) are
responding profile losses in both blade rows. Detailed computa- near 0.75, in agreement with observations from cascade theory.
tions (which are the principle justifcation for the approximation At higher loadings, losses increase more rapidly than the stage
made) show that stages with diffusion factor of 0.45 and with work. At loadings below the range of peak efficiency, losses re-
reaction of 0.5 provide an excellent estimate for the optimum, main nearly constant as the work decreases, resulting in lower
since the change in efficiency from these values to the optimized efficiencies. Similar trends with stage loading and flow coeffi-
case is on the order of 0.1% [19]. cients are seen in existing data for compressor stage efficiency,
although, in line with the aim of the paper, the levels in Figure 6
are higher by several points than quoted results [4]. The trends
A representative stage also agree with findings of Dickens and Day [14], who concluded
To show the procedure in a more concrete fashion, we that increasing stage loading to values above that of conventional
present results for stage efficiency for a representative aero en- designs (0.2 to 0.4) led to decreased stage efficiency.
gine high pressure compressor with the following design param- Table 2 shows the breakdown of loss sources in the rotor
eters. and stator. The even split between rotor and stator profile losses
Re 500,000 is due to the 50% stage reaction. The end wall boundary layer
AR 2.25 losses in the rotor are larger than in the stator because the ve-
ri /ro 0.65 locity on the rotor casing (the vector sum of the blade-relative
τ/(ro − ri ) 0.01 velocity distribution and the rotor tip speed) is higher than sta-
D 0.45 tor casing velocity (which for the 50% reaction case is the same
Λ 0.5 as the rotor blade-relative velocity distribution). At the design

6 Copyright © 2012 by ASME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 03/20/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 point considered, the end wall area is approximately equal to the TABLE 3:CALCULATED MAXIMUM EFFICIENCY OF
blade and vane surface area, and the boundary layer losses of STAGES REPRESENTING DIFFERENT AERO ENGINE
each are roughly equal as well; this would suggest that, under COMPRESSOR STAGE TYPES
the assumptions of the model, for well-designed stages, all skin
friction losses are proportional to the wetted surface area. The Re AR ri /ro τ/(ro − ri ) η (%)
downstream wake mixing loss is approximately 20% of the total
profile loss for rotor and stator. The tip clearance loss is about First LPC 1.0 × 106 2.5 0.8 0.01 96.1
10% of the total stage loss, corresponding to a loss of a half per- First HPC 0.5 × 106 2.25 0.65 0.01 95.4
cent in stage efficiency for one percent non-dimensional blade
height. Last HPC 1.0 × 106 1.25 0.9 0.05 93.1

Impact of design parameters mercial airliner. It is worth stating the assumptions and losses
The sensitivity of stage efficiency to design parameters can included in the model will have varying levels of applicability
be evaluated by considering the dependence of each in the com- depending on the stage. For instance, a compressor first stage
ponents of the efficiency formulation (given in Appendix A). can have incoming relative Mach numbers that lead to shocks
Numerical determination of these sensitivities is complicated and increased profile losses, and will likely have a higher reac-
(e.g., the dependence on Reynolds number, since the profile tion than 50%. The result here is thus an optimistic estimate of
losses are determined using an integral boundary layer method the efficiency in such a stage. When compared to other similar
with Reynolds-dependence closure relations) and varies with the stages, however, the trends can give useful insight into the impact
choice of inputs. Useful information can be obtained, however, of factors that are included in the model (e.g., choice of annular
from considering the approximate scalings of efficiency with re- geometry).
spect to each individual design parameter. For example, because Some of the trends described above are apparent, such as
the profile boundary layers are taken as turbulent, the boundary the decrease in efficiency with tip clearance height (here, the
layer thicknesses and corresponding boundary layer and wake non-dimensional clearance height has been estimated assuming
losses are proportional to Re−1/5 , a scaling inherent in the turbu- a fixed dimensional clearance height, leading to the large clear-
lent closure relations of the integral boundary layer method. ance in the final stage of the high pressure compressor). The be-
Aspect ratio and hub-to-tip radius ratio, the parameters that havior of efficiency with loading and flow coefficients for these
characterize the annular geometry, determine the end wall sur- designs have been seen to be similar to that seen in Figure 6. For
face area per unit mass flow. End wall loss is inversely propor- design parameters spanning the ranges in modern aero engine
tional to aspect ratio, since, if all other variables are held con- compressors, the peak efficiency is not particularly sensitive to
stant, an increase in aspect ratio shrinks the axial extent and thus Reynolds number or the annular stage geometry, which account
the relative end wall area of the blade rows. High aspect ratios for changes in efficiency of less than 1% over the range of de-
have been observed, however, to be susceptible to other perfor- signs presented in Table 3.
mance issues [21], we thus confine our examination to the pa-
rameter space of modern compressor aspect ratios. The hub-to-
tip radius ratio has a small effect on the relative end wall area, IMPLICATIONS FOR CYCLE PERFORMANCE
since the changes in relative hub and casing area roughly off- To provide context for the stage results, we present the sen-
set. Hub-to-tip ratio can, however, have a strong effect on rotor sitivity of engine performance to compressor efficiency. We con-
casing loss, since changes in ri /ro can increase or decrease the sider the thermal efficiency of an uncooled3 Brayton cycle, which
vector sum of the blade-relative velocity and the tip speed of the can serve as a surrogate for the gas generator of an aero engine4
rotor. or other gas turbine application.
The tip clearance loss model neglects interactions of the end
wall and clearance flows, and thus does not capture the increase
3 From a theoretical cycle performance standpoint, it has been shown that the
in losses at very small clearances (below roughly 0.5% of height)
and the resulting optimal gap height [9]; outside this range of presence of cooling flow has no effect on thermal efficiency [23]. In practice,
cooling flows may introduce additional losses in the engine [8], and an uncooled
clearances, however, the linear trend with gap height is in accord cycle can be thought of as representing a cooled cycle with a given level of cool-
with experience, though at a smaller rate [22] (see Appendix A ing and material technology.
for a fuller description of the leakage mixing model). 4 In this context, the gas generator consists of the inner annulus of the fan

Table 3 shows the peak efficiency for a few representative through which the core flow flows, the compressors, combustor, and the por-
tion of the turbine providing work for the compressors and inner annulus of the
stages. The design parameters are meant to represent specific
fan. This definition allows a separation of the gas generator thermal efficiency
stages of a moderate bypass ratio engine for a single-aisle com- (which depends on component efficiencies) from the propulsive efficiency (which
is driven by the fan pressure ratio).

7 Copyright © 2012 by ASME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 03/20/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 80 pressure ratios for the advanced cycles are as high as 100, well
Cycle thermal efficiency ηth (%)

ηturb = 92% beyond those of conventional machines. Two major challenges

ηturb = ηcomp arise in these types of machines: the first is increased compres-
70
sor exit temperature, the second is maintaining a high level of
compressor efficiency as the compressor exit corrected flow and
60 compressor blade heights decrease at the back of the machine.

50 SUMMARY AND CONCLUSIONS

A framework has been presented for estimating the limits of
axial compressor aerodynamic performance. In the analysis, a
40 rotor-stator geometry and an assumed surface velocity distribu-
90 92 94 96 98 tion are used to represent a generic compressor stage. Sources
Compressor polytropic efficiency ηpoly (%) of irreversibility are assessed using basic control volume and
boundary layer analyses, rather than by extrapolating data from
FIGURE 7: GAS GENERATOR THERMAL EFFICIENCY existing machines.
VERSUS COMPRESSOR EFFICIENCY The aim of the study was to determine an upper bound
on compressor stage efficiency, and this has been accomplished
through selection of the loss mechanisms that are included. By
For a given cycle temperature ratio, θt , and component ef- considering different loss sources under different assumptions,
ficiencies, there exists an optimal pressure ratio that maximizes however, the analytical framework can be adjusted to determine
thermal efficiency. The cycle thermal efficiency is thus presented possible avenues for performance improvement in different types
as a function of component efficiency with θt = 5.5, representing of stages.
a contemporary commercial aero engine at cruise [24]. Figure 7 Table 3 gives estimates of the upper bound on efficiency for
shows cycle efficiency versus compressor polytropic efficiency a range of stage designs. Although achieving these levels in prac-
for a fixed turbine efficiency of 92%. Turbine stage analysis sim- tice is indeed a stretch goal, the results show what advances are
ilar to the analysis presented in this paper suggests that simi- possible. The trend in efficiency with stage loading and flow co-
lar advances in turbine efficiency may also be possible [19], so efficients in Figure 6 is similar over the design parameters con-
cycle efficiency for equal compressor and turbine efficiencies is sidered, with peak efficiency occurring at stage loading coeffi-
also shown. The cycle efficiency is calculated assuming the flow cients between 0.2 and 0.4, where the stage profiles have prop-
through the engine is a perfect gas with constant specific heat, erties in ranges known to be efficient. This observation, coupled
starting at atmospheric conditions. At each point, the cycle pres- with the similarities in the shape of the “Smith chart” to that de-
sure ratio has been optimized for maximum thermal efficiency. veloped using empirical information [4] seems to indicate that
Both curves exhibit a more than linear increase in thermal effi- the losses included are dominant in determining the efficiency
ciency with polytropic efficiency. For constant turbine efficiency, trends with respect to basic stage design parameters.
a one percent increase in compressor efficiency results in an in- Figure 7 shows that component performance increases that
crease in cycle thermal efficiency of about 0.5%. If turbine ef- approach the estimated limits of component efficiency result in
ficiency is assumed to increase comparable to compressor effi- substantial increases in overall engine thermal efficiency (51.6%
ciency, a one percent increase in component efficiency results in to 58.9% for the case examined). Technological challenges aris-
an increase in cycle thermal efficiency of about 2.5%. ing from such advanced cycles include increased compressor exit
Given the stage performance results listed in Table 3, and ac- temperature and the maintainance of compressor efficiency as the
counting for the assumptions of the model, it seems reasonable exit corrected flow decreases with size and increased pressure ra-
to assume that an advanced compressor might have a theoreti- tio.
cal maximum polytropic efficiency as high as 95%. Assuming
a baseline polytropic efficiency of 92% in both the compressor
and turbine, this increase in compressor efficiency alone would ACKNOWLEDGMENT
result in an increase of 1.6% in cycle efficiency, and increasing This research was sponsored by the Fundamental Aeronau-
both compressor and turbine efficiency to 95% would allow an tics Program at the NASA Glenn Research Center under Coop-
increase of thermal efficiency of up to 7.0% for fixed turbine in- erative Agreement Number NNX08AW63A. The authors would
let temperature, with potential for larger increases as the turbine also like to acknowledge J. J. Adamczyk, N. A. Cumpsty, M.
temperature is increased. Increased component efficiency is only Drela, and W. K. Lord, for their contributions throughout the
one enabling technology required for these advanced cycles; the duration of this study, and A. H. Epstein, N. A. Fitzgerald, A.

8 Copyright © 2012 by ASME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 03/20/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 Huang, G. F. Kiwada, J. S. Sabnis, and J. B. Staubach for helpful namic Design and Analysis Using the Euler Equations.
comments and suggestions. GTL Report 187, Massachusetts Institute of Technology.
[16] Drela, M., 2009. “Power Balance in Aerodynamics Flows”.
AIAA Journal, 47(7), July, pp. 1761–1771.
[17] Drela, M., and Giles, M. B., 1987. “Viscous-Inviscid Anal-
REFERENCES ysis of Transonic and Low Reynolds Number Airfoils”.
[1] Koch, C. C., and Smith, L. H., 1967. “Loss Sources and AIAA Journal, 25(10), October, pp. 1347–1355.
Magnitudes in Axial-Flow Compressors”. Journal of Engi- [18] Huang, A., 2011. “Loss Mechanisms in Turbine Tip Clear-
neering for Power, 98(3), July, pp. 411–424. ance Flow”. MS Thesis, Massachusetts Institute of Tech-
[2] Koch, C. C., 1981. “Stalling Pressure Rise Capability of nology, Cambridge, MA, May.
Axial Flow Compressor Stages”. Journal of Engineering [19] Hall, D. K., 2011. “Performance Limits of Axial Turbo-
for Power, 103, pp. 645–656. machine Stages”. MS Thesis, Massachusetts Institute of
[3] Leroy H. Smith, J., 2002. “Axial Compressor Aerodesign Technology, Cambridge, MA, February.
Evolution at General Electric”. Journal of Turbomachinery, [20] Benzakein, M. J., 2001. “Propulsion Strategy for the 21st
124, July, pp. 321–330. Century: A Vision Into the Future”.
[4] Wright, P. I., and Miller, D. C., 1991. “An Improved [21] Wennerstrom, A. J., 1989. “Low Aspect Ratio Axial Flow
Compressor Performance Prediction Model”. In European Compressors: Why and What It Means”. Journal of Turbo-
Conference of Turbomachinery: Latest Developments in a machinery, 111, October, pp. 357–365.
Changing Scene, IMechE. [22] Wisler, D. C. Advanced compressor and fan systems. Lec-
[5] Denton, J. D., 1993. “Loss Mechanisms in Turboma- ture notes.
chines”. Journal of Turbomachinery, 115, October, [23] Horlock, J. H., 2003. Advanced Gas Turbine Cycles. Perg-
pp. 621–656. amon.
[6] Storer, J. A., and Cumpsty, N. A., 1994. “An Approximate [24] Cumpsty, N. A., 2003. Jet Propulsion, 2nd ed. Cambridge
Analysis and Prediction Method for Tip Clearance Loss in University Press.
Axial Compressors”. Journal of Turbomachinery, 116, Oc-
tober, pp. 648–656.
[7] Yaras, M. I., and Sjolander, S. A., 1992. “Prediction of Appendix A: Stage Efficiency Calculation
Tip-Leakage Losses in Axial Turbines”. Journal of Turbo- The loss in stage efficiency is calculated as the sum of the
machinery, 114, January, pp. 204–210. efficiency debits due to the four loss sources considered.
[8] Young, J. B., and Wilcock, R. C., 2002. “Modeling the Air-
Cooled Gas Turbine: Part 2–Coolant Flows and Losses”.
η = 1 − (∆ηsurf + ∆ηwake + ∆ηwall + ∆ηtip ) (10)
Journal of Turbomachinery, 124, April, pp. 214–221.
[9] Cumpsty, N. A., 1989. Compressor Aerodynamics. Long-
man Group UK Ltd. This amounts to seven terms that need to be calculated: the ∆η’s
[10] Lieblein, S., and Stockman, N. O., 1972. “Compressibility in both the rotor and the stator for blade boundary layer, end
Correction for Internal Flow Solutions”. Journal of Air- wall boundary layer, and wake mixing dissipation, and the rotor
craft, 9(4), April, pp. 312–313. clearance flow mixing loss.
[11] Halstead, D. H., Wisler, D. C., Okiishi, T. H., Walker, H. P.,
Hodson, H. P., and Shin, H.-W., 1997. “Boundary Layer
Surface velocity distribution.
Development in Axial Compressors and Turbines: Part 1 of
Each of the loss sources depends on the assumed surface
4 – Composite Picture”. Journal of Turbomachinery, 119,
velocity distribution. The velocity triangles of the stage profile
January, pp. 114–127.
(see Figure 2a) can be determined in terms of φ , ψ, and either a
[12] Hodson, H. P., and Howell, R. J., 1992. “Bladerow Interac-
specified incoming absolute swirl angle, α1 or stage reaction, Λ.
tions, Transition, and High-Lift Aerofoils in Low-Pressure
Turbines”. Journal of Engineering for Gas Turbines and 
Power, 114, October, pp. 603–611.  α1,spec 
[13] Smith, L. H., 1966. “Wake Dispersion in Turbomachines”. α1 = 2(1 − Λspec ) − ψ (11)
 tan−1
Journal of Basic Engineering, 88, pp. 688–690. 2φ
[14] Dickens, T., and Day, I., 2011. “The Design of Highly  
−1 ψ
Loaded Axial Compressors”. Journal of Turbomachinery, α2 = tan + tan α1 (12)
φ
133, July.
[15] Drela, M., 1986. Two-Dimensional Transonic Aerody- α3 = α1 (13)

9 Copyright © 2012 by ASME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 03/20/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use

  
1−1 Blade boundary layer dissipation.
β1 = tan − tan α1 (14)
φ The losses associated with blades and vanes is approximated
 
−1 1 using the two-dimensional stage profile. An integral boundary
β2 = tan − tan α2 (15)
φ layer method is used to solve numerically for the trailing edge
boundary layer quantities in terms of the blade velocity distribu-
Both the rotor blades and stator vanes are assumed to have tion and Reynolds number.5
circular arc camber lines. The camber length (which, under the
assumption of thin blades, is a good approximation to streamwise    ∗ ∗ 
u(`) δTE θTE θTE
surface length), is then given by F , Re −→ , , (21)
uin Cs Cs Cs
  1
Cs 2 (αin − αout )

= , (16) Applying Equation (4), the associated loss is determined in terms
c sin 21 (αin − αout )
of the calculated boundary layer kinetic energy thickness and the
profile geometry.
where αin is the blade-relative incoming flow angle (β1 for the
rotor and α2 for the stator) and αout is the blade-relative exit flow
angle (β2 for the rotor and α3 for the stator).   3 ∗
1 Cs 1 θTE φ2
The blade spacing is set by specification of either the solid- ∆ηsurf = σ . (22)
2 c cos αout Cs ψ
ity σ of a blade row or a diffusion factor D (from which σ is
calculated explicitly).
 End wall boundary layer dissipation.

 σspec The boundary layer loss is estimated assuming a constant
cos αin ψ dissipation coefficient with the end wall velocity distribution,
σ=   (17) uwall (x0 , y0 ).

 2 D−1+ cos α in φ
cos αout
Z 1Z 1 0 , y0 ) 3
φ 2 cos ξ 2 u wall (x
∆ηwall = CD dx0 dy0 , (23)
ψ AR 1 + (ri /ro )±1 0 0 ux
The linear profile velocity distribution can be determined
from the velocity triangles, blade shape, and blade spacing. In the present analysis, the integral in Equation (23) is estimated
using 5 × 5 point Gaussian quadrature.
  
u(`) ` cos αin ψ ` cos αin
= 1− 1± + (18)
uin Cs 2σ (Cs /c) φ Cs cos αout
Wake mixing dissipation.
The blade-relative end wall velocity distribution is assumed to The wake mixing loss is determined using a control volume
vary linearly across the blade pitch from suction to pressure side. analysis of the downstream mixing region shown in Figure 4.
Conservation of mass and momentum are used to determine the

change of flow angle and the static pressure drop due to mixing
urel = u p (x0 ) + y0 us (x0 ) − u p (x0 ) (19) in terms of boundary layer quantities. Once the velocities and
pressures downstream are known, the loss can be determined in
where us and u p are the profile suction and pressure side terms of the trailing edge boundary layer quantities by applying
velocities defined in Equation (18), and x0 = x/cx is the Equation (3).
non-dimensional axial direction, and y0 = rθ /s(r) is the non-
dimensional pitch-wise direction on the wall being considered.
For the rotor hub and both stator end walls, uwall = urel . The ve- φ2 
∆ηwake = [M − (1 − D) + (1 − D)2 ](1 − D)
locity on the rotor casing is estimated by vector addition of the ψ
 
blade tip velocity and the blade-relative end wall velocity urel , 1 (1 − D) − K (1 − D)3
assuming the direction of urel is equal to the blade stagger angle + − , (24)
2 cos2 αout cos2 αm
ξ.

s  2
1
uwall = urel cos ξ 1+ − tan ξ (20) 5 We do not present the details of the numerical computations or the boundary
φ (urel /ux ) cos ξ layer method here. For details of a similar implementation, see reference [17].

10 Copyright © 2012 by ASME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 03/20/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 where D, M, and K are related to the boundary layer displace- by
ment, momentum, and kinetic energy thicknesses,
q
   d ṁ = ρCd τ u2s − u2p dz, (30)
δ∗ Cs σ
D = TE , (25)
Cs c cos αout
   where Cd is a discharge coefficient to account for the detailed
θ Cs σ
M = TE , (26) behavior of the flow within the gap (a value of 0.8 is seen to give
Cs c cos αout
 ∗   results that match experimental data [6]).
θ Cs σ Given an assumed tip velocity distribution, Equation (29)
K = TE , (27)
Cs c cos αout can be integrated along the blade chord and divided by the stage
work to determine the loss in efficiency. Rather than use the
and the mixed-out flow angle is given by assumed profile distribution, a rectangular distribution was used.

  tan α2 − tan α1
1−D−M us − u p ≈ ux (31)
αm = tan−1 tan αout . (28) σ
(1 − D)2
2
us + u p ≈ ux (32)
cos ξ
Tip clearance flow dissipation.
With the assumption of isentropic flow through the rotor tip
clearance, Equation (6) can be rewritten as This gives the same blade loading at the tip while minimizing
the clearance loss. The resulting loss in efficiency can then be
expressed as
dΦ = us (us − u p )d ṁ. (29)
   √
τ Cs 2φ ψ(ψ cos ξ + 2φ σ )
∆ηtip = Cd (33)
The leakage mass flow, driven by the pressure difference, is given ro − ri c (2φ σ cos ξ )3/2

11 Copyright © 2012 by ASME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 03/20/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use