You are on page 1of 11

Proceedings of GT2006

ASME Turbo Expo 2006: Power for Land, Sea and Air
Proceedings
May 8-11, of GT2006
2006, Barcelona, Spain
ASME Turbo Expo 2006: Power for Land, Sea and Air
May 8-11, 2006, Barcelona, Spain

GT2006-90004
GT2006-90004

THE IMPACT OF SURFACE ROUGHNESS ON AXIAL COMPRESSOR PERFORMANCE DETERIORATION

Elisabet Syverud and Lars E. Bakken


NTNU, Norwegian University of Science and Technology,
Department of Energy and Process Engineering,
N-7491 Trondheim, Norway

ABSTRACT N Shaft speed, [rpm]


Axial compressor deterioration due to removable p Pressure, [kPa]
deposits is a major concern in the operation of gas turbines. r Blade radius, [m]
It is important to fully understand the flow mechanisms in R Gas constant, [J/kgK]
order to successfully monitor and clean the engine. Re Reynolds number
A test program on the GE J85-13 jet engine quantified T Temperature, [K]
the increased surface roughness and the distribution of salt U Blade velocity, [m/s]
deposits in an axial compressor. The test data showed good V Relative air velocity for the stator, [m/s]
agreement with published data for stage performance W Relative air velocity for the rotor, [m/s]
deterioration.
This paper compares the GE J85-13 test data on surface β Blade angle, [deg]
roughness to previously published work on surface ϕ Flow coefficient, Ca/Utip
roughness in compressors. The effect of surface roughness δ Displacement thickness, [m]
on the stage characteristics is modeled using theory for ε Defined in Eq. (10)
frictional losses, blockage and deviation. The results are γ Ratio of specific heats
compared to test data. η Isentropic efficiency
The most significant effect of increased roughness is λ Darcy friction factor
found to be the variation in the flow coefficient.
θ Momentum thickness, [m]
ρ Air density, [kg/m3]
Keywords: Axial compressor, roughness, stage
σ Solidity
characteristics, GE J85-13
ΨT Stage work coefficient
NOMENCLATURE c p ΔTt
ΨT =
A Flow area, [m2] 2
U tip
a Hydraulic friction loss factor
cp Specific heat at const. pressure, [J/kgK] ΨP Stage pressure coefficient
γ
C Absolute air velocity, [m/s] ⎡ ⎤
c p t ,1 ⎢
T ⎛ p ⎞ γ −1

Ψ = 2 ⎢⎜ ⎟ − 1⎥
P t ,2
c Chord length, [m]
U tip ⎜⎝ pt ,1 ⎟⎠
Deq Equivalent diffusion ratio ⎢⎣ ⎥⎦
F
GE
Head – flow exponent
General Electric
ω Total pressure loss coefficient
Ξ Mean value of H and θ along surface
H Form factor (δ/θ = 1.1 in this work)
i Incidence
Subscripts
j Boundary layer parameter for PS
a Axial
kCLA Average roughness, [μm]
eq Equivalent
ks Surface roughness, [μm] max Maximum surface velocity
K Defined in Eq. (7) PS Pressure side
m& Mass flow rate, [kg/s] R Rotor
M Mach number
1 Copyright © 2006 by ASME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 02/19/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use


rel Relative reliable data on the surface roughness found in gas turbines
s Static condition operating under realistic conditions.
S Stator In a recent test program on a GE J85-13 jet engine, all
SS Suction side three conditions are fulfilled. The jet engine was
t Total condition deteriorated by ingesting salt water, and the effect of
1 Inlet fouling was measured both by test cell performance
2 Discharge measurements, and by analysis of the internal surface
deposits. Axial compressor geometry and individual stage
INTRODUCTION characteristics are available from open literature. This
Recoverable gas turbine deterioration is the subject of provides a good basis for analyzing the effects of gas
great interest, due to the adverse impact even small turbine fouling. The test program and the results are
amounts of fouling can have on gas turbine performance, described in an earlier publication [9]. The test data
and the substantial number of commercial washing systems showed good agreement with published data.
available for performance recovery. Several papers have The purpose of this paper is to analyze the
been presented on the general causes and effects of gas effectiveness of two-dimensional models in the prediction
turbine fouling [1,2] and numerous theses have been of performance degradation due to salt ingestion in the GE
written on the subject (see refs. 3-5 for recent examples). J85-13. Analytical expressions for frictional losses found in
In addition, the current state-of-the-art compressor washing literature are applied to the known surface roughness of the
systems are presented in two recent publications [6,7]. GE J85-13. The results are compared to measurements.
Fouling in the axial compressor comprises airborne The discussion starts with a discussion on the surface
matters or aerosols which are deposited on the front roughness of fouled axial compressors, because this subject
surfaces of the inlet shroud and the internal gas path of the is not fully covered in the literature.
axial compressor, restricting the flow and causing
increased boundary layer thickness both at the blade SURFACE ROUGHNESS IN FOULED AXIAL
profiles and along the end-walls of the annulus and hub. It COMPRESSORS
has long been known that intake depression is a preferred There are only a few references describing surface
method for monitoring front stage fouling in axial roughness associated with axial compressor fouling in real
compressors[8,9]. However, this parameter is not applications. The deposits will vary with the operation and
commonly available in real applications, and compressor the airborne fouling at the site.
pressure ratio, compressor isentropic efficiency and power Blade surface roughness measured in a helicopter
reduction are commonly used substitutes for monitoring engine after 1500 hours of flight is reported [14]. The
compressor deterioration. relative roughness levels were in the range of 10.9E-3 in
The fundamental effect of surface roughness on the the first stage to 0.7E-3 in the last stage.
boundary layer development is given by [10]. However, Roughness measurements on a Rolls-Royce turbofan
this work covers the inherent roughness in the machine due engine after a “long period of airline operation” is reported
to the manufacturing process, and not the additional [15]: Along the centerline, the average roughness (kCLA)
roughness due to fouling. was 1.5-2 μm, which corresponds to an equivalent sand
The ASME test code on compressors [11] does not roughness of 160μm.
cover the effects of surface roughness in axial compressors. In the current work, Struer’s replicas were used to
The modeling of gas turbine fouling has been remove the deposits along the centerline of the stator vanes
approached by simulating the effect of a certain loss in in the first three stages of the GE J85-13 engine after the
compressor performance, often quantified as a certain engine was degraded with salt mist. The method is very
percentage loss in flow rate and loss in compressor simple, and can be applied to other gas turbines as well, but
pressure capability. Although such models are effective in requires disassembly of the gas turbine casing. The
predicting the relative changes in overall gas turbine deposits (salt crystals in this case) were analyzed in an
performance degradation, the models cannot improve our electron microscope to find representative pictures of grain
understanding of the local effects of fouling in axial size and distribution. Comparing the results to the work by
compressor stages, and therefore are not successful in Schlichting [16] gave the equivalent sand roughness and
modeling partial performance recovery due to maintenance relative roughness for the first three stages. The results are
activity like compressor washes. given in Table 1. Further description is given in [9]. One of
The modeling of local effects of fouling is often the striking results is the difference between the pressure
approached using two-dimensional stage stacking and the suction sides. As shown in Table 1, the greatest
techniques and relying on general stage maps [12] and/or roughness was found on the pressure side of the vanes.
the correlations of test data from two-dimensional cascade However, the surface roughness is not uniform along the
loss measurements [13]. The success of such models vanes. Only the surface roughness at the centerline was
depends on the following three factors: Firstly, reliable data analyzed in the microscope.
on the increased losses associated with surface roughness;
secondly, detailed knowledge of the axial compressor
geometry, including blade profile geometry; and thirdly,

2 Copyright © 2006 by ASME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 02/19/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use


Stator Equivalent sand Relative therefore to be accounted for separately. These losses are
vane roughness, ks roughness, assumed to be unchanged in fouled compressors.
[μm] ks/c
1 PS 25 11E-4 Friction
1 SS 10 4E-4 Two approaches are taken for the estimation of
2 PS 20 11E-4 frictional losses: i) the relatively new approaches on
2 SS 15 8E-4 Reynolds number dependence verified for centrifugal
3 PS 15 10E-4 compressors and ii) the classical approach of blade profile
3 SS 2 0.6 E-4 losses in axial compressors. The fundamental mechanisms
of internal hydraulic losses are founded in the Moody chart
Table 1 Salt crystal grain size and distribution at mean for pipe friction and should be equivalent regardless of
line [9] compressor type. In the following sections, both
The losses associated with surface roughness have been approaches are developed.
measured in three different test programs:
i) Milsch [17] tested the effect of roughness on the two- Equivalent Reynolds correction model
dimensional cascade losses and applied emery paper of It is generally recognized that a variation in Reynolds
grade 220, 80, 40 and 24, uniformly and completely on the number influences the turbomachinery performance. For
cascade profiles, giving a relative roughness of 0.56E-3 to turbocompressors most of the correction formulas
5.56E-3 for the cascade tests. The test results make an proposed in the past are generally considered to be
excellent database for the effect of roughness on the inadequate. Some correction procedures were too
suction and pressure side boundary layers of NACA complicated to use, while others did not reflect the
profiles. variation in the Reynolds dependent losses and led to an
ii) Bammert and Woelk [18] tested the effect of excessively high correction of the efficiency especially at
roughness on a three stage axial compressor with trailing high Reynolds numbers. A detailed review of the Reynolds
edge modified NACA 65-Series as the basic blade profile. dependent losses, both numerically and experimentally
The roughness was applied uniformly and completely on [20,21] reveals the impact on efficiency, head and flow
the blades using emery paper of grade 320, 220 and 120. coefficients. These centrifugal compressor correction
This gave a relative roughness of 1.5E-3 to 4.5E-3 in the procedures are widely accepted and verified by numerous
tests. The tests give the effect of surface roughness on the factory performance tests [21]. Revised correction
overall performance of a compressor and show the procedures are included in the ASME test code [11];
narrowing in the range of the constant speed lines and a however, the correction to flow coefficient is not included
strong influence on the flow coefficient with increasing in the latest revision of the standard.
roughness. However, the data are not very practical for In the fundamental approach, the ratio between the
fouled compressors because these rarely will see uniform internal hydraulic losses and the changes to the friction
distribution of the surface roughness on all stages. factor is constant for a given compressor [21]. This
iii) Elrod, King and Poniatowski [19] tested the effect approach is taken in the following, where the change in
of surface roughness and incidence angle on the two- surface roughness is modeled equivalent to a change in the
dimensional cascade with NACA 65-Series profiles and at Reynolds number for un-fouled surfaces, using an
critical Reynolds number (4.5E5). Three different relative equivalent Reynolds number in the equations for the
roughness levels of 2.1E-3 to 9.4E-3 were recessed into the hydraulic losses. This approach also accounts for sub-
surface on the first 25% chord on the suction side of the critical boundary layer flows and are therefore not
profile. Increased roughness was found to remove the restricted to the completely rough flow regime where
leading edge separation bubble, thicken blade boundary Reynolds number changes can be neglected.
layers, increase wake thickness and overall total pressure Because the Reynolds number is known, the associated
losses. However, freestream turbulence was found to have Darcy friction factor can be calculated using the Haaland
a proportionally greater effect on the cascade performance explicit equation [22]:
than increased surface roughness. −2
⎧⎪ ⎡ 6.9 ⎛ k s / c ⎞1.11 ⎤ ⎫⎪
λ = ⎨−1.8log ⎢ +⎜ (1)
⎟ ⎥⎬
LOSS MECHANISMS DUE TO GAS PATH ⎪⎩ ⎣⎢ Re ⎝ 3.7 ⎠ ⎦⎥ ⎭⎪
FOULING
Fouling in the gas path will potentially increase the skin Equation (1) matches the Moody diagram within 2%.
friction, decrease the deviation angle and affect blockage.
Fouling also causes concern about reductions in the overall Frictional loss model
stability. These fundamental loss mechanisms are reviewed The hydraulic losses due to friction are modeled as a
using literature studies and experimental data from a GE ratio between the hydraulic losses and the change in
J85-13. friction factor using the following approach:
The correction method applies only to the internal i) The Darcy friction factor for clean conditions, λclean,
hydraulic losses. The leakage and mechanical losses have is calculated from Eq. (1) at given Reynolds number and

3 Copyright © 2006 by ASME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 02/19/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use


assuming for the clean surface a relative roughness of Blade profile loss model
5ks/c, 9 ks/c, and 3 ks/c for stages 1, 2, and 3, respectively. The blade profile losses of axial compressors are
ii) The Darcy friction factor for fouled conditions, frictional losses and the wake mixing losses caused by the
λfouled, is calculated from Eq. (1) at the same Reynolds boundary layer flow across the blades and vanes. Because
number as for clean conditions but at the roughness level the calculation of the turbulent viscous flow in the
found in Table 1 for the suction surface. boundary layers is very complex, the determination of
iii) The equivalent Reynolds number is calculated from blade profile losses depends on empirical correlations
Eq. (1) using a friction factor equivalent to the fouled developed half a century ago for two-dimensional
condition and a roughness equivalent to the clean surface. incompressible flow in cascades. For new designs, these
iv) The critical friction factor for clean conditions, methods have virtually been abandoned, overtaken by
λclean,critical is calculated from Eq. (1) at a Reynolds number computerized fluid dynamic tools[24]. However, the
above critical condition and a roughness equivalent to the empirical correlations are still recognized for fundamental
clean surface. understanding and for preliminary design.
v) Finally, the relative change in hydraulic losses is To obtain an estimate of the losses associated with the
calculated from Eq. (2) [21]: fouling in the gas path of the GE J85-13, the old
λclean correlations are revisited. A similar approach has been
(1 − a ) + a found in literature but without validation to test data [25];
1 − ηclean λclean,critical
= (2) however, the current approach separates the frictional
1 − η fouled λ fouled losses associated with the suction and the pressure side of
(1 − a ) + a the blades and vanes, and includes several sources for the
λclean,critical effect on increased roughness on the blade profile losses.
where a is the fraction of hydraulic internal losses to total The GE J85-13 was designed with slender, transonic
losses. This fraction is taken as 0.4 at the design point blades for stage 1 and 2, and subsonic profiles for the
according to Howell’s breakdown of losses in an axial remaining stages, see Appendix A. In the current work, the
stage [23]. Fraction a is 0.7 for centrifugal compressors GE J85-13 compressor data are compared to data for the
[21]. NACA 65-Series.

Impact on head and flow at design point Cascade loss correlations


At the design point, the impact of hydraulic losses on In the early days, extensive cascade testing was carried
the isentropic head and flow coefficients are found from out to develop profile loss data of different profile families
the following relationships which are validated from at low Mach numbers. Because total pressure loss is
empirical data on centrifugal compressors [21]. Although fundamental in determining compressor blade
the polytropic head and efficiency are used in the original performance, a non-dimensional total pressure loss factor is
work, the isentropic head and efficiency are considered in used as the basic parameter. The loss is expressed by the
this paper. Applying the equations to a single stage, the end total pressure loss coefficient, ω [13,26,27]:
results should not be affected. To gain confidence in the
data, the relations should be carefully reviewed with data Δpt Δpt / pt ,1
on axial compressors; however, with the lack of such data, ω= = −γ
(5)
the equations for centrifugal compressors are applied here pt ,1 − ps ,1 γ −1 γ −1
1 − (1 + 2
M rel ,1 )
without further adaptation. 2

Ψ clean
P
η The correlation of blade profile loss data was simplified
= 0.5 + 0.5 clean (3)
by the observation that the empirical data on total pressure
Ψ fouled
P
η fouled
loss associated with the boundary layer development on a
2D cascade is effectively correlated by the momentum-
F thickness ratio and to a lesser extent the form factor of the
ϕclean ⎪⎧ Ψ clean
P
⎪⎫
=⎨ P ⎬ (4) wake[26]:
ϕ fouled ⎩⎪ Ψ fouled ⎭⎪
−3
θ σ cos β1 2 ⎡ 2 ⎤⎡ θ σ H2 ⎤
ω = 2( ) ( ) ⎢ ⎥ ⎢1 − ( ) ⎥
(6)
The head-flow exponent, F, in Eq. (4) is given as 0.5 c cos β 2 cos β 2 ⎣ 3 − 1/ H 2 ⎦ ⎣ c cos β 2 ⎦
for centrifugal compressors[21]. Our experience and
experimental data found in literature [18], show a much Lieblein showed that it was possible to correlate the
stronger influence on the flow coefficient at fouled data on total pressure loss on a suction surface diffusion
conditions. These effects are further described in the factor based on the velocity distribution for NACA 65-
section on blockage loss mechanisms. Series cascades operating at their minimum loss incidence.
The relationships for head and flow given in Eq. (3-4) The method was further developed by establishing an
are compared to the result of the blade profile loss model equivalent diffusion ratio which correlates the momentum
given in the next section. thickness data both at minimum loss as well as at greater

4 Copyright © 2006 by ASME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 02/19/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use


incidence angles for NACA 65(A10) and C4 profile data where jPS represents the form factor and the momentum
(both of 10% thickness-to-chord ratio) for unstalled flow thickness along the pressure side.
regimes [26]. The equivalent diffusion ratio is readily It was expected that the total skin friction coefficient
calculated from inlet and outlet conditions and avoids the would be lower than the conventional flat-plate value
need for detailed surface velocity data which are required based on inlet velocity Reynolds number, due to the
with the diffusion factor. In addition, the equivalent increase in negative velocity gradients with increased
diffusion ratio is essentially independent of solidity, air diffusion. However, the limited experimental data on
inlet angle and blade camber. The equivalent diffusion ratio NACA 65 and C4 profiles at different diffusion showed
was adapted to a general expression valid in a rotating good agreement with Eq. (9) with ε equal to 0.004 [26].
frame of reference and implementing changes in axial For the approximate case of zero diffusion along the
velocity [26]: pressure side, this represents a friction factor, λ, of 0.016
cos β 2 Wa ,1 ⎡ cos 2 β1 ⎤ and a roughness level of 4E-4 from the Moody chart. For
Deq = ⎢1.12 + 0.0117(i ) + 0.61
1.43
K⎥ the first stage of the GE J85-13 compressor, this roughness
cos β 2 Wa ,2 ⎣ σ ⎦ (7)
level corresponds to a surface roughness of 9μm which can
r2 Wa ,2 U ⎛ r2 ⎞ be manufactured relatively easily [10] and seems realistic
K = tanβ1 − tanβ 2 − 1 ⎜1 − 22 ⎟
r1 Wa ,1 Wa ,1 ⎝ r1 ⎠ for the early cascades.
In this work, the change in momentum thickness due to
Equation (7) is valid for NACA 65-Series profiles. For roughness is obtained from Eq. (9) and Eq. (10), assuming
stator rows, the blade speed is zero, and W is replaced by negligible changes in the surface diffusion and the factor Ξ
V. In this work, incidence is only accounted for in the rotor :
rows. θ fouled λ fouled
= (11)
Cascade data for momentum thickness can be θclean λclean
correlated on the equivalent diffusion factor [26]:
The Darcy friction factors used in Eq. (11) are derived
from from Eq. (1).
θ
= 0.004 ⎡⎣1 + 3.1( Deq − 1) 2 + 0.4( Deq − 1) 2 ⎤⎦ (8) A second source of information is found from a large
c series of cascade testing which was conducted to obtain
detailed information on the efficiency drop due to
The above Eq.(6-8) are used to calculate the profile roughness [17,28]. The tests were done on the NACA 65-
losses for three first stages of the GE J85-13 in clean and Series profiles, at 10% thickness to chord ratio and at
fouled conditions. For fouled conditions, the momentum different design lift coefficients and solidities. The surface
thickness is modified for roughness according to roughness was applied using emery grains glued onto the
expressions given in the following section. surface giving surface roughness levels up to 5.56E-3. All
tests were at low Mach numbers and at the minimum loss
Influence of roughness on momentum thickness incidence. The data includes momentum thickness
The influence of roughness on the momentum thickness distribution along the blade chord for both the suction and
is studied using three different sources found in literature. pressure sides of the blades.
The above correlation for total pressure loss was based The last source of information on the influence of
on the assumption that the skin friction coefficient will be surface roughness is found in the work by Koch &
small compared to the changes in the suction surface Smith[29]. This work is founded in the correlations of
diffusion term, and that the changes in the total wake Eq.(6-10), but provides corrections to the momentum
momentum thickness will result primarily from the thickness and form factors for variations in Mach number,
diffusion contribution of the suction surface boundary streamtube convergence (axial-velocity-density-ratio),
layer. Accordingly, the momentum thickness correlation is Reynolds number, and surface roughness. The data are
derived from the following equation [26]: validated only at minimum loss incidence. The surface
roughness correction to momentum thickness at a Reynolds
θ ε numbers of 1E6 is expressed as the polynomial equation
= (9)
c ⎛ Vmax,SS ⎞ given in Eq. (12), valid up to a relative roughness of ks/c =
1 + Ξ ln ⎜ ⎟ 12E-4 [29]:
⎝ V2 ⎠
θfouled
2
⎡k ⎤ k
where the factor Ξ represents the mean value of the form =91887 ⎢ s ⎥ + 945.87 s + 0.8908 (12)
factor and the momentum thickness along the surface, and θclean ⎣c⎦ c
ε represents a lump factor for skin friction and pressure
surface diffusion ratio, rewritten with the Darcy friction Figure 1 compares the data on momentum thickness
factor: change due to roughness in the above discussed literature.
⎛ Vmax, PS ⎞ The figure shows that the measured increase in momentum
ε= 1
8 ( λPS + λSS ) + jPS ln ⎜ ⎟ (10) thickness on the pressure side follows the data of the Darcy
⎝ V2 ⎠ friction factor for rough pipe flow, while the suction side
data follows the Koch & Smith relations for the low
camber airfoil. The steep change in momentum thickness at
5 Copyright © 2006 by ASME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 02/19/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use


the suction side of the airfoil of highest camber is not i) The surface roughness for the fouled condition is
predicted by any of the methods. given in Table 1. For the clean condition, a
relative roughness of 5ks/c, 9 ks/c, and 3 ks/c is
Koch&Smith [29] assumed for stages 1, 2, and 3, respectively.
NACA 65-(6)10 SS [17] ii) The change in momentum thickness due to
NACA 65-(6)10 PS [17] surface roughness is calculated from Eq.(12) for
NACA 65-(12)10 SS [17] the suction side and from Eq.(11) for the pressure
NACA 65-(12)10 PS [17] side. The results for the suction and pressure side
Darcy friction factor [22] are added to obtain the overall change in
6 momentum thickness.
iii) The total pressure loss coefficient is calculated for
clean and fouled conditions from Eq. (6).
iv) The total pressure loss is calculated for clean and
Relative change in momentum thickness .

5 fouled condition from Eq. (5)


v) The stage efficiency is calculated from the
following equation:
−γ
4 ⎡ γ − 1 2 γ −1 ⎤
1 − η profile = (ωR ) ⎢1 − (1 + M R ,1 ) ⎥
⎢⎣ 2 ⎥⎦ (13)
−γ
3 ⎡ γ − 1 2 γ −1 ⎤
+ (ωS ) ⎢1 − (1 + M S ,1 ) ⎥
⎣⎢ 2 ⎦⎥
2

Fraction of hydraulic losses to total losses


1 Eq.(13) is an expression for stage losses considering
only the blade profile losses. Applying this equation to an
axial compressor will determine the theoretical blade
profile losses in a stage. This is compared to the
0 compressor total stage losses [30] to obtain a split between
0.000 0.001 0.002 0.003 0.004 0.005 the hydraulic losses and the total losses at the design point:
Relative roughness, ks/c
1 − η profile
Figure 1 Momentum thickness change due to = χ1 (14)
roughness at the trailing edge 1 − ηtotal
The overall momentum thickness of the blade should be
the sum of the momentum thickness for the suction side Table 2 shows the results of Eq.(14) for the GE J85-13
and the pressure side. As can be seen in Figure 1, the in clean conditions and at 98% speed.
pressure side will have a significant influence on the
change in overall momentum thickness. Applying the ϕ 1 − η fouled 1 − ηtotal χ1
relationship given in Eq.(12) will only adjust for changes Stage
Eq. (13) [30] Eq.(14)
in the suction side roughness level, and will therefore be
1 0.545 0.042 0.110 0.38
inadequate in the overall estimation of blade profile loss.
2 0.542 0.041 0.108 0.39
In this work, the overall change in momentum thickness
is modeled using Eq. (12) for the suction side roughness 3 0.543 0.046 0.118 0.39
and Eq. (11), for the pressure side. The overall momentum Table 2 Ratio of hydraulic loss to total loss for GE
thickness is the sum of the calculated momentum thickness J85-13 at 98% speed setting
for both surfaces.
The results in Table 2 compares to the 70% ratio found
Frictional loss model
for the centrifugal compressors (factor a in Eq. (2)) which
The change in blade profile losses due to increase in again shows that other loss factors, like tip leakage, shock
friction is calculated based on the data on surface losses, and annulus losses are more significant in axial
roughness found in Table 1. For the rotor blades, the compressors than centrifugal compressors. Compared to
surface roughness was not measured. From observations, Howell’s breakdown of losses [23], the ratio of 38% seems
the roughness of the rotor blade deposits are defined as reasonable at the design point. The data in Table 2 serves
50% of the roughness found on the stator vanes. The as an example of axial compressors. However, only
frictional loss model calculates the blade profile loss for manufacturers with a large data base of axial compressor
both rotor and stator and by taking the roughness of the data can verify the results.
suction surface and pressure surface into consideration:
6 Copyright © 2006 by ASME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 02/19/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use


Impact on head and flow at design point
Milsch [17] NACA 65-(12)10
The stage pressure coefficient, Ψp, can be expressed in
terms of the total pressure loss coefficient to obtain an Mal'tsev and Shakhov [14]
expression of the influence of the blade surface losses on 0
Ψp. From Eq. (5) and the definition of Ψp the following
expression is derived for stage pressure ratio for the fouled
-1
case:

.
γ -2
pt , S ,2, fouled ⎡ P 2
U tip ⎤ γ −1
= ⎢ Ψ clean + 1⎥

Change in deviation, deg


pt , R ,1 ⎣⎢ c p Tt , R ,1 ⎥⎦ -3
⎡ γ −1 2 ⎤ (15) −γ

− (ωR , fouled − ωR ,clean ) ⎢1 − (1 + M R ,1 ) ⎥ γ −1


-4
⎢⎣ 2 ⎥⎦
−γ
⎡ γ − 1 2 γ −1 ⎤
− (ωS , fouled − ωS ,clean ) ⎢1 − (1 + M S ,1 ) ⎥ -5
⎣⎢ 2 ⎦⎥
The stage total pressure coefficient for fouled -6
conditions at the design point is given by the following
expression:
γ
-7
⎡ ⎤
⎢⎛ pt , S ,3, fouled ⎞ γ −1 0.000 0.002 0.004 0.006 0.008 0.010
− 1⎥⎥
c p Tt , R ,1
Ψ Pfouled = 2 ⎢⎜⎜ p ⎟⎟ (16)
U tip Relative roughness, ks/c
⎢⎣⎝ t , R ,1 ⎠ ⎥⎦
Figure 2 Change in deviation due to roughness
The method does not include any correction to the flow
coefficient. In addition, the method assumes critical Blockage
Reynolds conditions, and cannot compensate for Reynolds Surface deposits will cause internal blockage of the
numbers in the sub-critical range. throat areas due to the increased profile thickness,
increased 3D separations in the end-wall regions, and
Deviation increased wake thickness. The increased thickness of the
The flow turning evolves from the pressure distribution wake is accounted for by the blade profile losses.
around the profile. Increased surface roughness will affect The increased profile thickness due to deposits will
the pressure distribution and the deviation angle (being the cause a decrease of the throat area in the gas path. The
deviation from the blade trailing edge angle to the outlet measurements of the GE J85-13 showed a maximum salt
flow angle from the blade row). crystal size of 160μm along the profile mean line of the
Two data sources are found on the effect of surface second stage. Although scarcely distributed, this salt crystal
roughness on deviation: i) Milsch measured the change in increases the local profile thickness by 13%.
flow turning angle measured at different surface roughness Increased roughness is known to cause significant 3D
levels, profile thickness distributions, and solidities at a separation at the design point of a single stage compressor
Reynolds number of 4.3E5. The roughness elements were with inlet guide vanes and controlled diffusion airfoils
applied uniformly and completely on the cascade profiles (CDA) [15]. For smooth surfaces, 3D separation occurs
[17]. ii) Mal’tsev and Shakhov measured the difference in only at lower flow coefficients.
flow lag angle at different roughness levels, incidence Visual observation of the rotor deposits in the GE J85-
angles and solidities at Reynolds number of 2.4E5. The 13, showed heavy deposits along the hub and annulus,
roughness elements were applied at the back of the blade especially at the pressure side of the blade, with the
profile [14]. The data are compared in Figure 2 for deposits covering up to 30% of the total surface area. The
solidities of 1.33 and 1.21, respectively. The difference deposits are interpreted as a visualization of 3D separations
between the datasets can be explained by the location of in the gas path, although a direct comparison cannot be
surface roughness in the measurements. made to the above referred work [15] where the roughness
The Milsch data, measured for NACA 65-(12) 10 blade was applied only to the profile suction side. The pressure
profiles, represents the best data set for the GE J85-13 at side deposits found in the GE J85-13 should be further
full load and is used for modeling the effect of surface analyzed to fully understand the flow mechanisms and to
roughness on deviation. This is contrary to previously investigate its influence on overall compressor stability and
published work [9], where the deviation was assumed surge range.
insignificant based on the data from Malt’sev and Shakhov The end-wall losses are often represented by a simple
[14]. blockage factor. A more sophisticated model [29] allows
for the estimation of end-wall losses based on a model
7 Copyright © 2006 by ASME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 02/19/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use


relating the displacement thickness and the tangential force point for the stage characteristics. The stage design point is
thickness. Both of these effects come as a result of the end- given by [31].
wall boundary layer, and represent a measure of the mass Only data for the three first stages are included in this
flow reduction as a result of its presence. However, this work because these stages have the heaviest deposits and
loss model does not take into account changes due to the largest contribution to overall compressor performance.
roughness.
Stage pressure coefficient
Impact on head and flow at design point Figure 3 shows predicted change in stage pressure
The blockage is accounted for by reducing the suction coefficient for stages 1, 2, and 3 due to increased
volume flow rate, which causes a horizontal shift in the roughness as measured in the GE J85-13. Test data for the
flow coefficient of the stage characteristics. Head is GE J85-13 at full load is included for comparison.
assumed to be independent of blockage. None of the frictional loss models are effective in
The model is based on data for suction volume flow predicting the large variation in measured total pressure
[18]: coefficient from clean to deteriorated condition. The
ϕ fouled ⎡ ks ⎤
2
k degradation in the stage performance due to roughness is
= 10000 ⎢ ⎥ − 90 s + 1 (17) underpredicted by both models. However, the most
ϕclean ⎣c⎦ c significant effect is the deterioration in flow coefficient
which causes a large shift in the stage performance. The
Blockage is calculated using the average relative shift in performance from clean to deteriorated condition
roughness found on the stator vanes (Table 1). explains the importance of evaluating the individual stage
characteristics when calculating the overall effect of engine
Leading edge separation deterioration. The stage characteristics of the GE J85-13
Heavy leading edge deposits were found at the leading are very sensitive to changes in the flow coefficient, and
edges of the GE J85-13. Such deposits will cause the the compressor may operate at positive slopes of the
suction surface flow to separate at lower incidences, and is characteristics during normal operation. This causes
expected to reduce the pressure capacity of the compressor inherent stability problems in the compressor section,
and cause early rotating stall [18]. which is expected to be unusual in modern axial
The effect of leading edge separation is not modeled in compressor design. However, without this knowledge of
the current work for two reasons: i) it was not possible to the individual stage characteristics, it would be virtually
obtain test data along a single speed line, and ii) only the impossible to analyze the root cause of the engine
test data measured at 98-100% speed are used in the deterioration sensitivity.
verification. The salt deposits are located in the stagnation In Figure 3, it should be noted that the stage 1
point of the profile. Because the salt was injected at 98% maximum pressure coefficient point is not validated[30];
shaft speed, the incidence angle remains unchanged and hence no experimental data exists that verifies the
there will be no additional flow separation due to the performance curve towards surge. The nominal operating
leading edge deposits for the data points used in this work. points are well situated at the negative slope of the stage
No stall or other instability problems were encountered characteristics, and there are no data available in the
during testing. literature that validates the location of the maximum point
of this characteristic. Hence, the apparent overshoot of the
characteristic in the deteriorated condition merely indicates
IMPACT ON COMPRESSOR STAGE
that the maximum point of the characteristics is located at a
CHARACTERISTICS
higher pressure coefficient than given in literature[30].
The design point ratios ηclean/ηfouled, φclean/ φfouled, and
ΨPclean/ΨPfouled are assumed to be the same at all points on Stage efficiency
the stage characteristics. This is in agreement with the Figure 4 shows the predicted change in stage
original work on centrifugal compressors [21]. efficiency for stages 1, 2 and 3 due to a roughness increase
The loss mechanisms are applied to the first three as measured in the GE J85-13. Test data for the GE J85-13
stages of the GE J85-13, to simulate the effect of the at full load is included for comparison.
measured roughness due to salt deposits. Again, the significant shift in stage performance is due
Test data from GE J85-13 are used to validate the to a large shift in the flow coefficient, coupled with a large
models. The data are recorded at full load and constant inherent sensitivity in the stage design. Despite these facts,
exhaust gas temperature, with varying exhaust nozzle area. which are specific to the GE J85-13 compressor design, the
This provides data at nearly constant shaft speed and at models are underpredicting the changes in the stage
various air flow rates. One representative dataset is efficiency due to increased roughness.
included for each of the clean and deteriorated condition.
The same dataset as previously published [9] is used in this
work.
All data are calculated at reduced condition where the
reference condition is full load performance (engine
military setting) in the GE J85-13 test data, and the design
8 Copyright © 2006 by ASME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 02/19/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use


GE J85-13, clean GE J85-13, clean

GE J85-13, deteriorated GE J85-13, deteriorated

Design (clean engine) Design (clean engine)

Equivalent Re correction model Equivalent Re correction model

Blade profile loss model Blade profile loss model

Blade profile loss model with blockage Blade profile loss model with blockage

1.4
1.05
Stage 1 reduced total pressure coefficient

Stage 1 reduced isentropic efficiency


1.3

1.00
Ψ /( Ψ at design point)

η/(η at design point)


1.2

1.1
0.95
p

1.0
p

0.90
0.9

0.8
0.85
0.6 0.7 0.8 0.9 1.0 1.1
Reduced flow coefficient 0.6 0.7 0.8 0.9 1.0 1.1
Reduced flow coefficient
1.10
Stage 2 reduced total pressure coefficient

1.05 1.05

Stage 2 reduced isentropic efficiency


Ψ /( Ψ at design point)

1.00

0.95 1.00
η/(η at design point)
0.90
p

0.85 0.95
p

0.80

0.75 0.90

0.70
0.6 0.7 0.8 0.9 1.0 1.1
0.85
Reduced flow coefficient 0.6 0.7 0.8 0.9 1.0 1.1
Reduced flow coefficient

1.10 1.05
Stage 3 reduced total pressure coefficient

Stage 3 reduced isentropic efficiency

1.05
Ψ /( Ψ at design point)

1.00 1.00
η/(η at design point)

0.95

0.90 0.95
p

0.85
p

0.80 0.90

0.75

0.70 0.85
0.6 0.7 0.8 0.9 1.0 1.1 0.6 0.7 0.8 0.9 1.0 1.1
Reduced flow coefficient Reduced flow coefficient

Figure 3 Stage 1, 2, and 3 total pressure coefficient


Figure 4 Stage 1, 2, and 3 Isentropic efficiency

9 Copyright © 2006 by ASME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 02/19/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use


IMPACT ON OVERALL AXIAL COMPRESSOR deviation in axial compressors. Models found in literature
PERFORMANCE are validated based on measurements on a GE J85-13.
Increased roughness is known to cause significant None of the loss models are effective in predicting the
changes in the overall performance of a three stage large variation in measured total pressure coefficient from
compressor[18]. The compressor was designed with NACA clean to deteriorated condition.
65-Series profiles and with free vortex distribution. Emery The blockage model introduced in this work
grains of equal size were uniformly glued onto the entire underpredicts the blockage measured in the GE J85-13.
surface of the blades. Three complete cases were measured, The most significant effect of salt deterioration is the
with equivalent surface roughness ks/c, of 1.5E-3, 2.5E-3, reduction in the flow coefficient. Some of this sensitivity
and 4.5E-3. The results on total enthalpy rise are given in may be explained by the long, slender blade design of the
Fig. 5. GE J85-13, which are inherently sensitive to 3-D
Figure 5 compares the overall performance of the first separations. However, variation in flow coefficient is the
three stages of the GE J85-13 to the data for the three stage dominant effect in the GE J85-13 tests, and should not be
compressor[18]. The GE J85-13 data are measured at 97- ignored when modeling the effects of deterioration.
98% speed. The test data shows no significant influence of The pressure side deposits found in the GE J85-13
the total enthalpy rise, while the suction volume flow is should be further analyzed to fully understand the flow
strongly affected. The salt crystal size found along the mechanisms and to investigate their influence on overall
mean line in the deteriorated GE J85-13 resembles the 320 compressor stability and surge range.
emery paper. Further studies are required to verify the findings of
this paper. Although, the limited testing done on the GE
Smooth J85-13 indicates the main loss mechanisms due to surface
Emery grade 320
roughness, the geometry and stage characteristics of the
GE J85-13 may not be entirely representative for modern
Emery grade 220 engines. Further verification should be performed by
Emery grade 120 companies with first-hand knowledge of current aero-
design of axial compressors.
GE J85-13 Clean
The authors will suggest a fundamental review of the
GE J85-13 Deteriorated loss models and thereby a revision of the ASME and ISO
1.1 standard codes on axial compressor testing to account for
the effects of surface roughness and Reynolds number
variation in axial compressors. Such work must include the
1
.

effect on flow coefficient and should be extended to cover


Reduced total enthalpy rise

similar effects in the centrifugal compressors.


0.9
REFERENCES
[1] Kurz, R., and Brun, K., 2001, “Degradation in gas turbine
0.8 systems,” ASME J. of Engineering for Gas Turbines and
Power, 123, pp.70-77.
[2] Meher-Homji, C B, Chaker, M A, Motiwala, H M, 2001,
0.7 "Gas turbine performance deterioration," 30th
turbomachinery symposium, Texas A&M: 2001, pp. 139-176.
[3] Baker, J.D., 2002, “Analysis of the sensitivity of multi-stage
0.6 axial compressors to fouling at various stages,” Master’s
thesis, Naval postgraduate school, Monteray, CA, USA.
[4] Genrup, M., 2003, “Theory for turbomachinery degradation
0.5
and monitoring tools,” Thesis for Degree of Licentiate in
0.8 0.85 0.9 0.95 1 1.05 1.1 Engineering, Lund Institute of Technology, Lund, Sweden.
[5] Brekke, O., 2004, “Gas turbine performance deterioration,”
Reduced suction volume flow
Master’s thesis, NTNU, Trondheim, Norway.
[6] Stalder, J.-P., 2001, "Gas turbine compressor washing state of
Figure 5 Measured total enthalpy rise the art: Field experiences," ASME Journal of Engineering for
With respect to stability issues, the narrowing of stage Gas Turbines and Power, Vol.123, April, pp.363-370.
[7] Mund, F., Pilidis, P., 2004, “A review of online washing
characteristics due to early stall, as observed in the
systems,” Proceedings of ASME Turbo Expo 2004, June 14-
reference data [18] given in Fig. 5 is not experienced in the 18, Vienna, Austria, GT2004-53224.
GE J85-13 tests. Thus, the large shift in suction volume [8] Scott, J.N., 1979, “Axial compressor deterioration by
flow found in the GE J85-13 test data is realistic. measuring air intake depression,” National Research Council,
Third symposium on gas turbine operations and
CONCLUSIONS maintenance, Ch 13.
This paper describes the fundamental aspects of [9] Syverud, E., Brekke, O. and Bakken, L.E., 2005, “Axial
increased surface roughness on friction, blockage and compressor deterioration,” ASME Proceedings of GT2005,

10 Copyright © 2006 by ASME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 02/19/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use


June 6-9, Reno-Tahoe, NV, ASME GT2005-68701. Accepted [28] Bammert, K. and Milsch, R., 1972, “Boundary layers on
for publication in the ASME Journal of Turbomachinery. rough compressor blades,” ASME Gas Turbine and Fluids
[10] Schäffler, A., 1980, “Experimental and analytical Engineering Conference & Product Show, March 26-30,
investigation of the effects of Reynolds number and blade 1972, San Francisco, CA, USA; 72-GT-48.
surface roughness on multistage axial flow compressors,” [29] Koch, C.C. and Smith, L.H., 1979, “Loss sources and
ASME J. of Engineering for Power, 102, pp. 5-13. magnitudes in axial-flow compressors,” ASME J. of
[11] ASME PTC10-1997, 1997, “ASME performance test code on Engineering for Power, pp. 411-424.
compressors and exhausters.” [30] Milner, E.J., and Wenzel, L.M., 1975, “Performance of a J85-
[12] Kurzke, J, 1996, "How to get component maps for aircraft 13 compressor with clean and distorted inlet flow,” NASA
gas turbine performance calculations," ASME International TM X-3304.
Gas Turbine and Aeroengine Congress & Exhibition, [31] Willoh, R.G., and Seldner, K., 1969, “Multistage compressor
Birmingham, UK, June 10-13. simulation applied to the prediction of axial flow
[13] Lieblein, S, 1959, “Loss and Stall analysis of compressor instabilities,” NASA TM X-1880.
cascades,” ASME Journal of Basic Engineering, September [32] Smith, L.H., 2002, “Axial compressor aerodesign evolution
1959, pp. 387-400. at General Electric,” ASME J. of Turbomachinery, 124, pp.
[14] Mal’tsev, Yu.N., and Shakov, V.G., 1989, “Influence of 321-330.
roughness of deposits in compressor cascade on flow lag [33] Hager, R.D., 1977, “Analysis of internal flow of J85-13
angle (English translation),” Izvestiya VUZ, Aviatsionnaya multistage compressor,” NASA TM X-3513.
Tekhnika, 32, pp.80-82. [34] Tesch W.A., and Steenken, W.G., 1976, “Blade row dynamic
[15] Gbadebo, S.A., Hynes, T.P., Cumpsty, N.A., 2004, “Influence digital compressor program: Volume 1 J85 clean inlet flow
of surface roughness on three-dimensional separation in axial and parallel compressor models,” NASA CR-134978.
compressors,” Proceedings of ASME Turbo Expo 2004,
Vienna, Austria, GT2004-53619. APPENDIX A: GE J85-13 AXIAL COMPRESSOR
[16] Schlichting, H., 1979, Boundary-layer theory, McGraw-Hill DESIGN
Inc., New York, NY, 7th edition.
[17] Milsch, R., 1971, “Systematische Untersuchung über den The J85 evolved in the 1950s as the first real attempt by
Einfluss der Rauhigkeit von Verdichterschaufeln auf den General Electric to employ transonic compressor
Gitterwirkungsgrad,” Doktor Ingenieur Dissertation, technology. The J85 holds a free-vortex design with high
Technischen Universität Hannover, Germany. aspect ratio blades and transonic design conditions in the
[18] Bammert, K. and Woelk, G.U., 1979, “The influence of the front stages. GE almost always used a modified NACA
blading surface roughness on the aerodynamic behavior and 65-Series thickness distribution with a circular arc mean
characteristic of an axial compressor,” Proceedings of ASME line for the subsonic blades, and relied on NACA transonic
Gas Turbine Conference 1979, San Diego, Ca, 79-GT-102.
compressor double circular arc design for the transonic
[19] Elrod, W.C., King, P.I., Poniatowski, E.M., 1990, “Effects of
surface roughness, freestream turbulence, and incidence blades. This is assumed for the J85-13 as well. It should be
angle on the performance of a 2-D compressor cascade,” noted that GE realized much later that the NACA 65 A10
ASME International gas Turbine and Aeroengine Congress & airfoils were a poor design choice for the rotor hubs [32].
Exhibition, Brussels, Belgium, June 11-14, Paper number 90- The desired cylindrical casing and the high stage
GT-208. pressure ratios led to a large hub slope. The rising hub and
[20] Wiesner, F.J., 1978, “A new appraisal of Reynolds number the transonic Mach numbers at the tip were not
effects on centrifugal compressor performance,” ASME Gas satisfactorily handled at the time. The design leads to high
turbine conference, London, UK, April 9-13, 78-GT-149. total pressures at the tip, a starving hub and unsatisfactory
[21] ICAAMC group, 1987, “Influence of the Reynolds number
stall line, and required substantial development trimming
on the performance of centrifugal compressors,” ASME Gas
Turbine Conference and Exhibition, Anaheim, CA, USA, and the addition of stages until two satisfactory models of
May 31-June 4, 87-GT-10. the J85 were achieved and produced [32]. The problems
[22] White, F.M., 1991 Viscous fluid flows, McGraw-Hill, Inc., existed in use as well, and lead to redesign of the bleed
Second edition. valves in the 1970s after the twin engine F5 aircraft had
[23] Cumpsty, N, 1989, Compressor Aerodynamics, Longman encountered severe stall problems in the leeward J85
Scientific & Technical. engine. The problems encountered with the J85 led to
[24] Halstead, D.E.; Wisler, D.C., Okiishi, T.H.; Walker, G.J.; substantial information on stage performance being
Hodson, H.P., Shin, H.-W., 1997, “Boundary Layer released in unclassified documentation. The engine is
Development in Axial Compressors and Turbines: Part 1 –
therefore extensively documented, more than what is
Composite Picture,” Transaction of the ASME, Journal of
Turbomachinery, 119, January, pp. 114-127.
common. [30-31,33-34].
[25] Marson, E., 1992, “Effect of manufacturing deviations on
performance of axial flow compressor blading,” ASME Gas
Turbine and Aeroengine Congress and Exposition, June 1-4,
1992, Cologne, Germany; 92-GT-326.
[26] Lieblein, S., Rodebush, W.H., 1956, “Theoretical loss
relations for low-speed two-dimensional-cascade flow”,
NACA TN 3662.
[27] NASA SP-36, 1965 “Aerodynamic design of axial flow
compressors,” NASA Lewis Research Center, Cleveland,
OH, USA.

11 Copyright © 2006 by ASME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 02/19/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

You might also like