1007

Experimental investigation of the inuence of fouling on compressor cascade characteristics and implications for gas turbine engine performance
D Fouias1 , A Gannan1 , K Ramsden1 , P Pilidis1 , D Mba1 , J Teixeira1 , U Igie1 , and P Lambart2 1 Gas Turbine Engineering Group, Department of Power and Propulsion, School of Engineering, Craneld University, Craneld, Bedfordshire, UK 2 R-MC Power Recovery Ltd, Stamford, Lincolnshire, UK The manuscript was received on 26 February 2010 and was accepted after revision for publication on 14 May 2010. DOI: 10.1243/09576509JPE992

Abstract: This article describes the ndings of a study which examined the inuence of fouling on the behaviour of a cascade and by making use of these results the performance implications for gas turbine engines of exposure to airborne foulants. A suction-type compressor cascade tunnel with a plenum chamber was employed for investigating fouling blade effects. The tests showed that such a testing arrangement allows the extraction of pressure and corrected velocity distribution data downstream of the blades that is comparable with what can be obtained from blow-type cascade tunnels. This study presents experimental results for smooth clean cascade blades and for uniformly fouled blades. For all the cases considered, mid-span-corrected velocity distributions and pressure losses taken one chord downstream of the blades were investigated in order to identify the effects of fouling on the blades. The result of fouling on exit ow angle was investigated as well. In the present study, cascade clean and fouled cases were used to predict real engine performance. Results are obtained in terms of stage polytropic efciency, thermal efciency, useful power, and compressor efciency deterioration. Roughening the cascade blades uniformly with particles of 254 m size, the compressor efciency dropped by 7.7 percentage points. Keywords: compressor cascade, polytropic efciency, thermal efciency, useful power, compressor efciency

INTRODUCTION

The performance of the compressor of an industrial gas turbine can suffer signicantly from fouling due to the ingestion of particles like sand and dust. In very hostile environments when particles mix with oil vapour, the outcome is a substantial loss in power output and cycle efciency due to compressor fouling. In order to recover this performance loss and subject to manufacturers ring temperature limitations, the engine fuel ow could be increased. This, however, would reduce the turbine blades creep life and result
Corresponding

author: Gas Turbine Engineering Group, Depart-

ment of Power and Propulsion, School of Engineering, Craneld University, Craneld, Bedfordshire MK43 0AL, UK. email: d.fouias@craneld.ac.uk
JPE992

in large increases in gas turbine engine operating costs. Therefore, investigating the effects of blade roughness gives valuable knowledge about the mechanisms of fouling and how this can be reduced. Surface roughness effects on compressor blades strongly inuence the performance of gas turbine engines. Gbadebo et al. [1], applied surface roughness on stator blading of a single-stage low-speed axial compressor. With surface ow visualization and exit loss measurements it was shown that the threedimensional (3D) separation at the hub was increased by the presence of roughness. The separation identied by Gbadebo caused a signicant reduction in the stage total pressure rise. In addition, the experimental work illustrated that applying surface roughness between the leading edge and the location of the peak suction caused a signicant reduction in stage performance, whereas applying surface roughness
Proc. IMechE Vol. 224 Part A: J. Power and Energy

1008

D Fouias, A Gannan, K Ramsden, P Pilidis, D Mba, J Teixeira, U Igie, and P Lambart

downstream of the suction peak had a virtually negligible effect. Applying roughness on the blades the stage total pressure rise was reduced, but at higher ow coefcients the effect of roughness was insignicant. Nikuradse [2] investigated the ow in pipes that were roughened with sand grains and divided the ow into three regions. In the rst region, the roughness of the surface is completely covered by the laminar sublayer, and it was noted that surface roughness has no inuence on the ow. In the second region, some roughness elements protruded through the sublayer causing an increase of the pressure drop coefcient. In the third region, the roughness elements protrude completely through the sublayer and the pressure drop coefcient does not increase further. Schafer [3] investigated experimentally the effect of Reynolds number on four multi-stage axial compressors and reported three different regimes of operation characterized by the condition of the blade boundary layer, the laminar separation, the turbulent attached ow with hydraulically smooth blade surface, and the turbulent attached ow with hydraulically rough blade surface. Schafer dened a lower critical Reynolds number below which laminar separation takes place. Above this critical Reynolds number, the efciency increases as the Reynolds number increases. This increase stops when the Reynolds number reaches the upper critical Reynolds number, which is 5 105 , and the blade surface behaviour is hydraulically rough. Also, it is stated that the lower critical Reynolds number for standard production blades is equal to or below the value of 105 . Below this value, signicant blade ow separation occurs and the blade can stall. Bammert and Milsch [4] ran experiments with compressor cascades consisting of roughened NACA 65 series blade sections in different geometrical variations and illustrated that loss coefcients rise by increasing the roughness grade dened as the ratio of sand grain size over the chord length. Also, they illustrated that increasing the surface roughness resulted in a decrease of the ow turning angle. Bammert and Woelk [5] investigated the inuence of blade surface roughness on the aerodynamic behaviour and characteristic of a three-stage axial compressor model by using emery grain uniformly roughened blades. For a relative roughness ks /c = 4.51 103 , the authors reported a reduction in static pressure ratio of 30 per cent and a reduction in overall efciency of 13 per cent. The equivalent sand grain roughness ks represents the size of sand grains which give the same skin friction coefcients in internal passages as the roughness being evaluated [6]. Bons [7], having done a review of surface roughness and effects in gas turbines, states that the rst evidence of roughness inuence takes place when the roughness reaches an admissible level ks,adm = 100c /Rec and this represents a design guideline. The
Proc. IMechE Vol. 224 Part A: J. Power and Energy

author also states that for a turbulent boundary layer, exceeding ks,adm a modest increase in blade prole losses may occur; however, for a laminar boundary layer, the prole loss could increase by up to a factor of 2. On the other hand, if the laminar boundary layer was already prone to separation, at lower Reynolds numbers roughness-induced transition could minimize separation and reduce prole losses. Tarabrin et al. [8], analysing computational results, reported a 4.5 per cent reduction in mass ow, 4 per cent reduction in pressure ratio, and 2 per cent reduction in compressor efciency, as successive stages of a gas turbine foul for the rst six stages. Tarabrin et al. [9] reported that the gas turbine unit sensitivity to axial compressor fouling decreases by increasing the turbine entry temperature (TET) and keeping the compressor pressure ratio constant. Keeping the TET constant and increasing the compressor pressure ratio, the gas turbine unit sensitivity to fouling increases. They also stated that a 1 per cent decrease in axial compressor efciency of a single shaft gas turbine due to fouling can cause a reduction in useful power (UW) output of 2.82 per cent, keeping the TET and pressure ratio constant. Kurz and Brun [10] developed a model for a twoshaft gas turbine engine with power turbine in order to investigate engine performance degradation. By using 2.1 per cent loss in compressor efciency, 5 per cent reduction in airow, 5 per cent reduction in pressure ratio, and a 0.5 per cent reduction in gas generator turbine efciency, the authors reported an 8.6 per cent reduction in power and an efciency drop of 3.5 per cent. Zaba [11] investigated the effect of fouling on the performance of an industrial gas turbine. He stated that if all the compressor stages are equally and uniformly fouled, the percentage reduction in the compressor volume owrate is approximately equal to the percentage reduction in the compressor efciency. The author also reports that fouling of the rst stages has greater impact on the compressor volume owrate than fouling of the rear stages and the percentage change in the volume owrate is greater than the percentage change in the compressor efciency. For the case he investigated experimentally, he found that for heavily fouled rst stages the change in volumetric owrate is approximately equal to 2.5 times the percentage change in compressor efciency. Meher-Homji and Bromley [12] ran performance simulations for a 39.6 MW industrial gas turbine with a ring temperature of 1377.6 K. Imposing a 6 per cent deterioration in mass ow and 5 per cent deterioration in the compressor efciency, the authors found that the power output drops by 5.5 MW, hence by 14.3 per cent. Howell [13] developed a method to relate cascade tests to the performance of a compressor stage. This is
JPE992

Experimental investigation of the inuence of fouling on compressor cascade characteristics and implications

1009

used here to predict the performance deterioration of a compressor in an engine and the associated changes in the whole engine performance, from changes in cascade performance due to fouling. 2 EXPERIMENTAL FACILITY

centrifugal fan

cascade tunnel exit cone intake

The compressor cascade wind tunnel employed for this investigation is a suction wind tunnel, with a cross-sectional area of 0.043 m2 , and is designed for an air mass owrate of approximately 5 kg/s, which can be varied via a throttling valve installed at the exit. This corresponds to an inlet Mach number of 0.3 and a Reynolds number of 3.8 105 formed with the inlet velocity and the chord length. This value is above the lower critical Reynolds number of about 105 [3], below which laminar separation occurs. The degree of turbulence and the mean velocity of the inlet air ow were measured with a hot wire anemometer as 2.25 per cent and 52.6 m/s, respectively. The test section of the cascade comprises nine 2D blades of NACA 65 thickened prole with circular arc camber line. The blades are set at zero incidence; the blade span is 180 mm, the chord length 60 mm, and the pitch-to-chord ratio s /c 0.8. The cascade blades represent a stator blade mean section of a gas turbine compressor with 50 per cent reaction. The blades have a span-to-chord ratio (aspect ratio) of 3. The selection of the particular number of blades and aspect ratio was made in order to minimize the inuence of the cascade wall boundary layer effects on the test blade section. In a compressor cascade tunnel, the increase in static pressure across the blades causes wall boundary layer thickening and contraction of the ow. Owing to the contraction of the ow the meridional velocity increases. This contrasts with the diffusing action of the compressor cascade itself. As a consequence, the sidewall boundary layer effectively reduces the static pressure increase that theoretically would be achieved. To reduce this effect, Dixon [14] and many others suggest that at least seven blades are needed for compressor cascade testing and that each blade has a minimum aspect ratio of 3. Figure 1 illustrates the cascade tunnel arrangement. A plenum chamber was installed behind the cascade test section. The ideal cascade experiment would discharge the airow from the blading into an innite space (e.g. the atmosphere). In the current project, however, the facility employed a suction pump (fan). Accordingly, the blading was designed to discharge into the largest practical volume feasible to ensure a uniform exit static pressure. In addition, it is equally important that the effect of the plenum chamber shape itself and its exit ducting on the fan does not distort the cascade discharge ow.
JPE992

plenum chamber

cascade test section

settling chamber

Fig. 1

Actual cascade tunnel left side view

ROUGHNESS AND MEASURING PARAMETERS

The three middle cascade blade proles were roughened by covering the surfaces of the blades with very thin double-sided sticky tape with a thickness of 0.09 mm. Application of carborundum allowed the blade surface roughness distribution to be nearly uniform. The carborundum grain sizes were dened with emery grade numbers which represent different sieve sizes via which the grains pass through. The grit numbers involved were 220, 180, 120, and 60 meaning grain sizes of 63, 76, 102, and 254 m, which correspond to K /c values of 0.0010, 0.0013, 0.0017, and 0.0042, respectively. The roughness height K was assumed to be equal to the average size (mean diameter) of the carborundum grains. Traverse pressure measurements one chord upstream of the blades leading edges (see Fig. 2) were taken with an L-shaped pitot static tube. The same pitot static probe was used to evaluate the mass ow by traversing along a cross-sectional plane perpendicular to the ow in the area of the settling chamber. The plane was located at a distance of 50 per cent of the settling chamber width upstream of the rst cascade blade, marked blade 1 in Fig. 2.

perspex walls

top

data collection line 1

SS PS 0 mm -40 data collection line 2 blade 1 bottom

120

Fig. 2

Cascade test section

Proc. IMechE Vol. 224 Part A: J. Power and Energy

1010

D Fouias, A Gannan, K Ramsden, P Pilidis, D Mba, J Teixeira, U Igie, and P Lambart

A three-hole cobra probe was used for downstream traverses at a distance of one chord from the blade trailing edge. At this distance, most of the mixing would have taken place and pitchwise ow angle variation will be only small, according to Gostelow [15]. The probe was nulled for the measurements and the probe traversing took place at mid-span. The width of the probe head is 2.4 mm, the height is 0.8 mm, and the total length is 500 mm. The three holes of the probe are located in the same plane parallel to the probe axis, and their diameter is 0.5 mm. The head of the probe is trapezoidal with a characteristic wedge angle = 45 . The pressure readings extracted from this yaw probe were combined in order to provide particular coefcient values that were related to the probe calibration chart so as to work out the ow exit angles of the blades. Velocity within the ow eld was calculated from the measurements of total and static pressure. The blade prole total pressure loss coefcient = P1 P2 P1 p 1 (1)

0.980 0.975 0.970 0.965

p2/Pamb

0.960 0.955 0.950 0.945 0.940 -40 -30 -20 -10 0 10 20 30 40 50 60 70 80 90 100 110 120

pitchwise distance (mm)

Fig. 4

Normalized rear static pressure distribution at 60 mm traverse

was obtained via analysis of the pressure data obtained from the up and downstream traverses. Flow uctuations gave rise to a variation in measured total pressure at any point in the ow of not more than 100 Pa. This represents a variation of plus or minus 0.5 per cent. The total temperature was measured in the test cell and not in the ow. In fact, the ow total temperature was assumed to be the test cell ambient temperature. All data for pressure and temperature were non-dimensionalized. 4 EXPERIMENTAL RESULTS SMOOTH BLADES

The velocity distributions one chord downstream of the cascade were investigated at mid-span for clean (smooth) blades. A statistical average of all the results taken at different ambient conditions is presented in Fig. 3. It can be seen that the passage velocity increases from 81 to 85 m/s towards the upper end of the cascade. This variation occurs because at the cascade exit
88 86 84 82 80 78 76 74 72 70 68 66 64 -40 -30 -20 -10 0 10 20 30 40 50 60 70 80 90 100 110 120

there is a pitchwise gradient in static pressure (see Fig. 4) of about 0.2 per cent. This causes the ow to deviate towards the lower static pressure region and progressively increases the values of exit ow angle 2 at the exit of the cascade blades. This pressure gradient is due to the fact that the settling chamber is smaller than what would be required to allow the cascade to discharge to uniform static pressure. The dimensions of the settling chamber were dictated by size limitations of the test location. When employing suction-driven cascade experiments, exit conditions in terms of static pressure are always greatly inuenced by the downstream discharge conditions. Given that the pressure gradient identied above is of modest proportions this was judged not to constitute an insuperable obstacle to the obtaining of useful data with the present rig conguration. The measured wakes were typically around 10 mm wide with the velocity defect falling from 83 to 73.5 m/s in the wake. Figure 5 illustrates a statistical average of pressure loss coefcient values calculated at different ambient conditions. The wake loss coefcient of the middle blade reaches a maximum value of 0.16 and the left and right blade loss coefcients both reach a value of around 0.18. The loss coefcient corresponding to

0.30

Total pressure loss coefficient w

0.25

Outflow velocity (m/s)

0.20

0.15

0.10

0.05

0.00 -40 -30 -20 -10

10

20

30

40

50

60

70

80

90 100 110 120

pitchwise distance (mm)

pitchwise distance (mm)

Fig. 3

Statistical average velocity distribution at 60 mm traverse

Fig. 5

Statistical average loss distribution at 60 mm traverse

JPE992

Proc. IMechE Vol. 224 Part A: J. Power and Energy

Experimental investigation of the inuence of fouling on compressor cascade characteristics and implications

1011

44 43 42 41 40 39 38 37 36 35 34 33 32 31 30 29 28 27 -40 -30 -20 -10

Exit flow angle (degrees)

Blade 6

Blade 5

Blade 4

Roughened blades
0 10 20 30 40 50 60 70 80 90 100 110 120

Clean blades

pitchwise distance (mm)

Fig. 7

Illustration of roughened blades

Fig. 6

Statistical average exit ow angle distribution at 60 mm traverse

the three passages investigated is almost constant at a value of 0.04. It is important to note that all the cascade experiments have been undertaken at a nominal inlet Mach number of 0.3. In fact, the range of Mach number varies from Mach 0.297 to Mach 0.31. Examples of typical variation of losses with Mach number over a range of incidences reported [16, 17] show that for Mach number less than 0.5 losses are independent of Mach number. For the test cases considered here, experimental examination of losses at Mach numbers less than 0.3 has not been undertaken. These values are somewhat lower than Mach 0.5, which is characteristically found in the mid-span of the rst stage of industrial compressors. However, the fan capacity did not allow the experiments to reach that number. A statistical representation of the distribution of blade exit ow angle, 2 , is illustrated in Fig. 6. It can be observed that the middle blade wake exit ow angle is approximately 36 whereas in the passage the exit ow falls to 34 . In addition, there is a small (1 ) decrease in the maximum value of the wake exit ow angle, from one side of the cascade to the other.

the blade surface because a number of small roughness particles are fully submerged in the laminar sublayer and do not protrude or are not felt by the turbulent ow. Nevertheless, they must be taken into account in terms of the arithmetic mean value. Schafer [3] used an effective roughness height k which describes the peaks rather than the average and is dened as the difference between the arithmetic averages of the ten highest peaks and the ten deepest grooves existing per millimetre length. The measuring length used by Schafer was 5 mm. Schafer [3], after measuring the surface roughness of several blades, found a correlation between the effective roughness height k and the arithmetic average height Ra (see equation (2)) k = 8.9 Ra (2)

EXPERIMENTAL RESULTS ROUGHENED BLADES

Roughness was applied uniformly on the three middle cascade blades. The particles used were carborundum of 63, 76, 102, and 254 m average size stuck with very thin double-sided sticky tape on the blades (see Fig. 7). Roughness is usually described by the arithmetic average value Ra, dened as the arithmetic mean of the absolute departures of the roughness prole from the mean line. This is the line tted through the prole where the areas of the prole above and below this line are equal [3]. However, this arithmetic mean value Ra is not sufcient in dening the hydrodynamic characteristic of
JPE992

For this project the parameter Rz was used as well, which is the average height difference between the ve highest peaks and the ve lowest valleys within a sampling length of 0.8 mm [18]. A surface texture-measuring device (Taylor Hobson Sutronic 25) was used involving a stylus following the surface of the blades in the streamwise and spanwise directions and taking roughness measurements at stations of 25 per cent, 50 per cent, and 75 per cent of the chord and 25 per cent, 50 per cent, and 75 per cent of the span. All the span and streamwise measurements were averaged, and the results are presented in Tables 1 and 2 for upper and lower blade surfaces, respectively. The number of sampling lengths assessed
Table 1 Upper surface roughness
Upper surface streamwise roughness (m) Ra 0.6 7.2 10.11 14.62 21.47 Rz 4.67 40.11 51.33 73.67 107.44 k 5.34 64.41 90.00 130.14 191.05

Particle size (m) 0 63 76 102 254

Upper surface spanwise roughness (m) Ra 0.42 6.73 10.02 14.44 22.71 Rz 2.56 37.22 51.78 73.44 114.00 k 3.76 59.93 89.20 128.55 202.13

Proc. IMechE Vol. 224 Part A: J. Power and Energy

1012

D Fouias, A Gannan, K Ramsden, P Pilidis, D Mba, J Teixeira, U Igie, and P Lambart

Table 2

Lower surface roughness

Lower surface streamwise roughness (m) Ra 0.47 7.38 10.11 14.33 22.44 Rz 3.22 41.11 51.11 72.33 109.22 k 4.15 65.66 89.99 127.57 199.76

Particle size (m) 0 63 76 102 254

Lower surface spanwise roughness (m) Ra 0.37 7.4 10.4 14.53 24.04 Rz 2.11 41.33 53.22 74.22 122.78 k 3.36 65.86 92.56 129.35 214.00

from the measuring device was 5, and a mean value for every roughness parameter measured was calculated within an assessment length l dened from the sum of the sampling lengths. This provides a better statistical estimate of the parameters measured value. Bons [7] collected a wide variety of proposed correlations which have been employed to convert measurable surface roughness parameters (Ra, Rz , or Rq ) to equivalent sand grain roughness ks (including equation (2)). Bons noted that many of the correlations vary by up to a factor of 5. This shows that no single correlation appears to capture all of the relevant physics for both engineered and service-related roughness. The effect of fouling on the pitchwise velocity distribution at one chord downstream of the cascade blades was investigated for clean and fouled blades. The corrected velocity, dened as the ratio of exit ow velocity to the square root of the ambient temperature (V2 / Tamb ), in the middle of the passage remains almost constant at a value of 4.9 (Fig. 8). However, as the fouling level increases this ratio undergoes significant changes. Increasing the particle size to 63, 76, 102, and 254 m the velocity ratio in the wake drops from 4.39 (clean blade) to 4.21, 4.14, 4.10, and 3.91, respectively. These values correspond to percentage falls of 10.4 per cent, 14 per cent, 15.5 per cent, 16.3 per cent, and 20.2 per cent from the passage value of 4.9. This indicates that the wake-corrected velocity behind the roughened blades drops continuously though not linearly. In addition, the wake widens as the fouling level increases. However, the increase is not gradual. For
0 microns
5.20

clean blades the wake is approximately 10 mm thick. However, for fouling levels of 63, 76, and 102 m the wake thickness remains almost constant at 24 mm. Applying particles of 254 m on the blades, the wake thickness increases considerably to about 39 mm. As seen in Fig. 8 all the wakes move towards the right with respect to the 0 m wake as the fouling level is increased. This happens due to the higher boundary layer growth of the suction surface of the blades compared to the pressure surface boundary layer. The effect of the pitchwise static pressure increment towards the upper end of the cascade (see Fig. 4) behind the blades is assumed not to be responsible for this wake shift towards the upper end of the cascade. This is because the ow Mach number there is very low. For such cases, probable incidence changes do not cause signicant changes in the blade drag coefcient, which is related to the thickness of the boundary layer affecting the blade wake. The pressure losses corresponding to the all three middle passages behind the blades were investigated. Figure 9 illustrates the pressure loss distribution one chord downstream of the blades towards the cascade exit streamtube. Taking into account the middle passage, as the fouling level applied on the blade surface increases from 0 m (smooth blades) to 63, 76, 102, and 254 m the total pressure loss coefcient increases from 0.16 to 0.25, 0.27, 0.285, and 0.35, respectively. This indicates that the wake total pressure loss coefcient behind the roughened blades rises continuously but not linearly. The passage loss varies smoothly around the value of 0.04 and it seems that the fouling does not affect this area signicantly. It was noted that as the fouling level increases from 0 to 76 m, the middle passage exit ow angle (one chord downstream) corresponding to the nulling point in the passage between blades 4 and 5 increases from 34 to 37 . Increasing the fouling level further to 102 and 254 m, the passage exit ow angle at the same point gets values of 38 and 39.5 (see Fig. 10). Hence, the passage exit ow angle seems to increase in a lower proportion as the roughness increases more
0 microns
0.40

63 microns

76 microns

102 microns

254 microns

63 microns

76 microns

102 microns

254 microns

Total pressure loss coefficient w

0.30

V2/ Tamb (m/s / K0.5)

5.00 4.80 4.60 4.40 4.20 4.00 3.80 3.60 -40 -30 -20 -10 0 10 20 30 40 50 60 70 80 90 100 110 120

0.20

PS

SS

0.10

0.00 -40 -30 -20 -10

10

20

30

40

50

60

70

80

90

100 110 120

pitchwise distance (mm)

pitchwise distance (mm)

Fig. 9 Fig. 8 Roughness wakes at 60 mm traverse

Roughness cases loss distribution at 60 mm traverse

JPE992

Proc. IMechE Vol. 224 Part A: J. Power and Energy

Experimental investigation of the inuence of fouling on compressor cascade characteristics and implications

1013

41
0.30

smooth blades - statistical average data

blades with double sided sticky tape

Exit flow angle (degrees)

40
Total pressure loss coefficient

39 38 37 36 35 34 33 0 20 40 60 80 100 120 140 160 180 200 220 240 260

0.25

0.20

0.15

0.10

0.05

0.00 -40 -30 -20 -10 0 10 20 30 40 50 60 70 80 90 100 110 120

particle size (microns)

pitchwise distance (mm)

Fig. 10

Blade 45 passage exit ow angle versus particle size

Fig. 11

Effect of adding thickness on the total pressure loss coefcient

than 102 m. Up to the 102 m the increase in exit ow angle is 4 , and from 102 to 254 m the increase is reduced to 1.5 only. This can be attributed to the increase of the passage blockage which progressively reduces the deviation and therefore the exit ow angle. Though the trend is unmistakable, the exact angular variation is subject to the measurement accuracy estimated at 1 , as represented through the error bars in Fig. 10.

effect on total pressure loss coefcient, especially for the two side cascade blades. 7 THEORETICAL BACKGROUND Howell [13] developed a uid dynamic theory whose basis is knowledge of the 2D ow past cascade blades and of the appropriate correction factors needed to provide the mean stage conditions in an actual compressor. In Howells work, the blade prole drag and lift coefcients are obtained as follows CDp = cos3 m p0 s 2 c (1/2) V1 cos2 1 s CL = 2 cos m (tan 1 tan 2 ) c (3) (4)

BLADE THICKNESS EFFECT

The effect of blade thickness due to roughness has not been taken into account in the current study. Gbadebo et al. [1], in order to separate the effect of thickness from the effect of roughness, performed tests by covering the leading edge/peak suction region with thin cardboard strips of similar thickness to that of the emery paper used for applying roughness. Comparing contours of stage pressure rise coefcient for thickened blades with these of smooth and roughened blades, they found that the thickness has negligible contribution to wake thickening. According to Saravanamuttoo et al. [16], test results for subsonic compressor blade sections shows that, at low Mach numbers, the losses for zero incidence are very low compared to those at higher incidence and higher Mach number. The current cascade runs with an inlet Mach number of 0.3 and at a nominal incidence of 0 . The corresponding losses are expected to be low and depend only secondarily on blade thickness. However, at high Mach number and incidence far away from nominal, as expected, the losses increase dramatically. In this case, blade thickening due to roughness signicantly increases the pressure losses. The effect of adding thickness to the middle cascade blades by attaching the double-sided sticky tape was investigated experimentally. Figure 11 shows that addition of the double-sided sticky tape has almost no
JPE992

where tan m = 1 (tan 1 + tan 2 ) 2 (5)

In order to proceed towards a representation of a real compressor stage, it was considered that the stage reaction R is 50 per cent. Then knowing the blade pitch, the blade height, and the blade lift coefcient values, the overall drag coefcient can be calculated CD = CDp + CDa + CDs (6)

The drag coefcient, related to wall annulus friction losses, is CDa , and the drag coefcient CDs , which describes the secondary losses, particularly those that give rise to trailing edge vortices, can be calculated as follows, employing Howells method CDa = 0.02 s /H CDs =
2 0.018CL

(7) (8)

Annulus wall boundary layers give rise to spanwise distributions of axial velocity, which tend to be more
Proc. IMechE Vol. 224 Part A: J. Power and Energy

1014

D Fouias, A Gannan, K Ramsden, P Pilidis, D Mba, J Teixeira, U Igie, and P Lambart

peaky as the air passes through the compressor stages. Because of this, the axial velocity at the blade midspan is higher than the design value and less work will be generated. Theoretically an increase in the work will take place at the blade ends, but due to stalling at these areas no increase in work occurs and the result is a reduction of the work done for the whole blade below the design value. The work done factor , which is the ratio of actual rotor blade whirl velocity to the ideal change of whirl velocity [19], applies a correction for all these losses, and the mean value for multi-stage compressor used by Howell [13] was 0.86. Howell and Bonham [20], in order to plot the stage characteristics curve using cascade testing data, employed the stage temperature rise coefcient cp Ts /0.5U 2 (to use Howells notation), stage efciency s (or polytropic efciency p ), and stage pressure rise coefcient ps /0.5 U 2 . These quantities had to be calculated for different incidences as follows [21] cp T s Va = 2 (tan 1 tan 2 ) 2 (1/2)U U 2 CD s = 1 sin 2m CL c p Ts ps = s 2 (1/2) U (1/2)U 2 (9) (10) (11)

blade row in the stage of the compressor would need separate cascade test data. Only in the case of a 50 per cent reaction stage can the results of a single cascade test be applied to both blade rows. For the case of smooth blades, the values of stage ow coefcient Va /U and stage loading coefcient H /U 2 were 0.524 and 0.247, respectively. Another parameter used as an input to the Turbomatch simulation was the percentage deterioration in the cascade passage non-dimensional mass ow 0.5 (W1 T1 /P1 )p obtained from the experimental results due to fouling. This was assumed to be equal to the percentage reduction in non-dimensional mass ow of the industrial gas turbine engine examined. 8 EXPERIMENTAL PERFORMANCE SIMULATION RESULTS

The cascade tunnel results were analysed according to Howells theory [13] in order to relate the cascade with an actual stage. The blade prole loss coefcient was calculated and then the overall drag coefcient was estimated. Calculating the lift coefcient and the mean cascade ow angle m , the polytropic efciency p related to an actual stage was calculated. The actual stage polytropic efciency calculated was then compared with the polytropic efciency of an industrial gas turbine (165 MW power output). Running cascade experiments for different levels of fouling, different values of actual stage polytropic efciency were calculated and for each case the percentage deterioration in polytropic efciency was calculated compared to smooth blade cases. The fall in polytropic efciency obtained from the experiments was used as an input to the gas turbine zero-dimensional performance simulation tool, Turbomatch [22, 23], developed at Craneld University, for the examination of the performance of a large industrial engine. In this study, it is assumed that the percentage deterioration in polytropic efciency of this engine equals the percentage deterioration of the polytropic efciency derived from the experimental cascade expressed as stage results. An assumption is made that the mid-radius reaction is 50 per cent. It is worth noting that isolated cascade tests in other than 50 per cent reaction stages can only be applied to one of the two blade rows of the stage at appropriate stagger and camber angles. The other
Proc. IMechE Vol. 224 Part A: J. Power and Energy

The polytropic efciency was calculated for smooth and fouled blades of the current project compressor cascade tunnel, Fig. 12 showing that for fouling levels of 63, 76, 102, and 254 m, the percentage deterioration in polytropic efciency is 2.2 per cent, 3.9 per cent, 4.9 per cent, and 7.7 per cent, respectively. The conversion of the loss data from the cascade tests into a typical change in polytropic efciency of the real compressor stage takes into account the change in outlet angle as given in Fig. 10. Table 3 shows the percentage reduction of the cascade passage non-dimensional mass ow with respect to the fouling particle size applied on the blades.
8 7 6 5

p %

4 3 2 1 0 0 20 40 60 80 100 120 140 160 180 200 220 240 260

Fouling particle diameter (microns)

Fig. 12

Experimental percentage deterioration in polytropic efciency due to fouling

Cascade passage percentage reduction in non-dimensional mass ow
0.5 /P ) % ( W 1 T1 1 p

Table 3

Particle size (m) 0 63 76 102 254

0.000 0.4 0.5 0.7 1.7

JPE992

Experimental investigation of the inuence of fouling on compressor cascade characteristics and implications

1015

0 microns 102 microns 0.370 0.360 0.350 0.340

63 microns 254 microns

76 microns

th

0.330 0.320 0.310 0.300 0.290 0.280 0.270 1100 1200 1300 1400 1500

TET (K)

Fig. 13 Thermal efciency versus TET

TET=1100 K TET=1400 K 15.5 TET=1200 K TET=1500 K TET=1300 K

increases by 25.5 per cent. Therefore as the fouling particle size increases, the percentage thermal efciency gain in the same level of TET increase gets higher. As a result of the diverging constant pressure lines in the temperatureentropy diagram (Fig. 15), the UW progressively increases with increasing TET. As a result of compressor fouling, the compressor efciency reduces and the compressor work increases. As TET increases, the effect of increasing compressor work with increasing fouling decreases. Finally, the outcome is that the engine performance measured by the thermal efciency is less sensitive to component inefciency as TET increases. This fact is illustrated by the reducing range of thermal efciency change with increasing TET shown in Fig. 13. In conclusion, the engine performance deterioration due to fouling is highest at low TET. From the Turbomatch results obtained it was shown that as the fouling level increases the UW of the gas turbine decreases, but increasing the TET this drawback could be handled (see Fig. 16). Taking into account the

15.0

PR

14.5

14.0

13.5 0 20 40 60 80 100 120 140 160 180 200 220 240 260

Fouling particle diameter (microns)

Fig. 14

Pressure ratio reduction due to fouling

For this study, it was assumed that the compressor is fouled uniformly all the way through. Turbomatch performance simulation runs were carried out while varying the TET between 1100 and 1500 K, combined with the effect of fouling. The thermal efciency of the engine was examined rst. The thermal efciency th increases as the TET increases (Fig. 13), and this can be attributed to the parallel increase of the compressor pressure ratio (Fig. 14). Increasing the particle size from 0 to 254 m for constant TET levels the pressure ratio was found to decrease by 1.3 per cent. However, keeping the fouling level constant and increasing the TET from 1100 to 1500 K the pressure ratio increased by 8.3 per cent. As the fouling increases, the pressure ratio drops and the thermal efciency decreases (i.e. referring to values at constant TET in Figs 14 and 13, respectively). Considering the case of smooth blades (0 m), the TET increases from 1100 to 1500 K and the thermal efciency increases by 15.5 per cent. Taking into account the case of the highest fouling of 254 m for the range of the same TET increase, the thermal efciency
JPE992

Fig. 15 Temperature versus entropy [24]

0 microns 102 microns 200 190 180 170 160 150 140 130 120 110 100 90 80 70 60 1100 63 microns 254 microns 76 microns

UW (MWatts)

1200

1300

1400

1500

TET (K)

Fig. 16

UW versus TET

Proc. IMechE Vol. 224 Part A: J. Power and Energy

1016

D Fouias, A Gannan, K Ramsden, P Pilidis, D Mba, J Teixeira, U Igie, and P Lambart

case of 0 m (smooth blades), incorporating a fouling level of 254 m on the blades at the same TET of 1100 K the UW drops by 19.2 per cent and only by 9.8 per cent for the level of 1500 K. In order for the engine to recover the original UW at 0 m after suffering from fouling level of 254 m at the TET of 1100 K, it has to increase its TET by almost 50 K. For the case of 102 m passing to 254 m, the TET should increase less in order to recover the original useful clean engine output. Also, for constant values of fouling level, the UW was increased linearly with respect to the TET increase. The decrease in UW as the fouling level (particle size) on the blades increases is caused due to a reduction in the engine mass ow capacity. Using predicted data obtained with the Turbomatch code, Fig. 17 shows that for the industrial engine whose design point is at a TET of 1378 K, the rate of reduction of compressor efciency with roughness particle size increase is nearly independent of TET. Referring to the TET value of 1100 K, as the particle size increases from 0 to 254 m the compressor efciency falls as stated earlier. Also, increasing the TET from 1100 to 1500 K the compressor efciency c keeps falling by almost 1.2 per cent for the smooth (0 m) and all the fouling cases examined. This drop in efciency is due to the fact that the engine non-dimensional mass owrate decreases as the TET increases. This happens because for a compressor constant speed running line, as the TET increases the pressure ratio increases (moving from point A to point B), as Fig. 18 illustrates. However, the non-dimensional mass owrate of the engine falls following the trend of the efciency line passing from the design point shown in Fig. 19. Keeping the TET constant and increasing the fouling level gradually, the performance drawbacks in terms of percentage compressor efciency, thermal efciency, and UW deterioration are illustrated in Table 4. From the experimental cascade results correlated with the real uniformly roughened fouled engine via Howells theory, it was found that by increasing the roughness

N / Ndp =0.56 N / Ndp=0.79 point A, TET=1100 K 40 35 30 25

N / Ndp=0.6 N / Ndp=0.9 point B, TET=1500 K

N / Ndp=0.67 N / Ndp=1

B A

PR

20 15 10 5 0 0.060 0.080 0.100

0.120
0.5

0.140

0.160

0.180

0.200

WeTe /Pe

Fig. 18

Industrial engine pressure ratio versus nondimensional mass owrate

N / Ndp =0.56 N / Ndp=0.79 point A, TET=1100 K 1.100 1.050 1.000 0.950

N / Ndp=0.6 N / Ndp=0.9 design point

N / Ndp=0.67 N / Ndp=1 point B, TET=1500 K

0.900 0.850 0.800 0.750 0.700 0.060 0.080 0.100 0.120 0.140
0.5

0.160

0.180

0.200

WeTe /Pe

Fig. 19

Industrial engine compressor efciency versus non-dimensional mass owrate

Percentage efciency and UW reductions due to fouling of industrial gas turbine
0 m 63 m 2.2 84.2 4.6 191.5 2.5 0.307 2.9 0.360 1.4 76 m 3.9 80.4 8.9 187.4 4.6 0.297 5.9 0.355 2.7 102 m 4.9 78.3 11.3 185.0 5.8 0.292 7.6 0.352 3.4 254 m 7.7 71.3 19.2 177.3 9.8 0.274 13.2 0.344 5.6

Table 4

c (%) UW (1100 K) (MW) UW (1100 K) (%) UW (1500 K) (MW) UW (1500 K) (%) th (1100 K) th (1100 K) (%) th (1500 K) th (1500 K) (%)

0 88.2 0.0 196.5 0.0 0.316 0.0 0.365 0.0

Fig. 17

Compressor efciency versus TET

up to a level of 254 m the drawbacks in compressor efciency can be as high as 7.7 per cent. Taking into account these deterioration percentages in terms of compressor efciency, increasing the fouling towards the level of 254 m at 1100 K TET, the percentage deterioration in the UW ( UW (1100 K) per cent) produced by the engine was found to be 19.2 per cent. This percentage deterioration was eliminated almost by half, as the TET was increased to 1500 K. For the same TET increase and fouling particle size of 254 m, the percentage deterioration in thermal
JPE992

Proc. IMechE Vol. 224 Part A: J. Power and Energy

Experimental investigation of the inuence of fouling on compressor cascade characteristics and implications

1017

efciency was eliminated by 57.6 per cent. For lower levels of fouling, and increasing the TET in the same range, similar trends in terms of UW and thermal efciency were found. Therefore, it can be stated that one performance deterioration inhibitor when fouling presents, is the parameter TET which must be increased properly. 9 CONCLUSIONS

The analysis of the cascade experiments showed that as a result of increasing the fouling level, the width and the depth of the wakes increased signicantly. More signicantly it was observed in all fouling cases that the wakes shifted towards the blade suction side due to the increased boundary layer growth on the suction surfaces of the blades. Increasing the roughness to 254 m caused the total pressure loss coefcient (measured one chord downstream of the blades) to double (i.e. from 0.16 to 0.35). Also, by increasing the level of fouling, the wake-corrected velocity defect and the total pressure loss coefcient rise measured downstream of the blades were observed to vary gradually though not linearly. From the experimental results, it can be observed that in the middle of the blade passage, the increase in the exit ow angle when the fouling level was raised from 0 to 102 m was reduced by almost half when fouling was further increased from 102 to 254 m. This is due to the resulted added exit ow blockage related to the widening of the wakes as the roughness increases. This article presents a method of establishing a relationship between the compressor cascade tested and a real compressor stage. Howells theory of correlating cascade data with compressor stages was employed. The performance simulation tool Turbomatch was used to examine a real engine subject to the fouling comparable to the four different fouling cases examined experimentally. The results showed that by increasing the fouling level, the deterioration in both polytropic and overall efciencies of the compressor increases continuously. By increasing the fouling level to 254 m, the percentage deterioration in compressor efciency reached a value of 7.7 per cent when compared to the smooth blades. The compressor efciency falls when the fouling level on the blades increases due to the pressure ratio degradation caused. However, for the compressor examined, by increasing the TET from 1100 to 1500 K the efciency falls by 1.2 per cent, no matter how severely fouled the compressor may be. Again, in order for the engine to recover the loss in UW due to the increasing compressor work caused by fouling, the TET must increase. Increasing the TET by 36.4 per cent (from 1100 to 1500 K) and keeping the fouling level constant, the engine percentage performance
JPE992

deterioration in terms of UW and thermal efciency was reduced by almost 50 per cent. As the TET increases, the thermal efciency of the engine increases as well, as a result of the increase in compressor pressure ratio. Also, for the same range of TET increase, the higher the level of fouling on the blades, the higher the gain in thermal efciency, compared to clean blades. Keeping the TET constant and increasing the fouling level, the pressure ratio drops and the thermal efciency decreases. The UW increases linearly with the TET, and for constant TET, by increasing the fouling level, the power reduces due to the reduction in the mass ow capacity of the engine. At lower TETs, the decrease in UW due to fouling is higher than that corresponding to higher TETs as a result of the extra margin in UW incorporated, as the diverging lines of the temperatureentropy diagram show. ACKNOWLEDGEMENTS The authors thank Paul Lambart, Russell Gordon, Andy Lewis, and Jonathon ODonnell from the R-MC Power Recovery Limited for their technical advice and support. Authors 2010 REFERENCES
1 Gbadebo, S., Hynes, T., and Cumpsty, N. Inuence of surface roughness on three-dimensional separation in axial compressors. ASME J. Turbomach., 2004, 126, 455 463. 2 Nikuradse, J. Laws of ow in rough pipes. NACA TM 1292, National Advisory Committee for Aeronautics, 1933. 3 Schafer, A. Experimental and analytical investigation of the effects of Reynolds number and blade surface roughness on multistage axial ow compressors. ASME J. Eng. Power, 1980, 102, 513. 4 Bammert, K. and Milsch, R. Boundary layers on rough compressor blades. ASME paper no. 72-GT-48, 1972. 5 Bammert, K. and Woelk, G. U. The inuence of the blading surface roughness on the aerodynamic behavior and characteristic of an axial compressor. ASME J. Eng. Power, 1980, 102, 283287. 6 Zhang, Q., Goodro, M., Ligrani, P., Trindade, R., and Sreekanth, S. Inuence of surface roughness on the aerodynamic losses of a turbine vane. ASME J. Fluids Eng., 2006, 128, 568578. 7 Bons, J. A review of surface roughness effects in gas turbines. ASME J. Turbomach., 2010, 132, 021004-1 021004-16. 8 Tarabrin, A. P., Bodrov, A. I., Schurovsky, V. A., and Stalder, J. P. An analysis of axial compressor fouling and a cleaning method of their blading. ASME paper no. 96-GT-363, 1996.
Proc. IMechE Vol. 224 Part A: J. Power and Energy

1018

D Fouias, A Gannan, K Ramsden, P Pilidis, D Mba, J Teixeira, U Igie, and P Lambart

9 Tarabrin, A. P., Bodrov, A. I., Schurovsky, V. A., and Stalder, J. P. Inuence of axial compressor fouling on gas turbine unit performance based on different schemes and with different initial parameters. ASME paper no. 98-GT-416, 1998. 10 Kurz, R. and Brun, K. Degradation in gas turbine systems. ASME J. Eng. Gas Turbines Power, 2001, 123, 7077. 11 Zaba, T. Losses in gas turbines due to deposits on the blading. Brown Boveri Rev., 1980, 1280, 715722. 12 Meher-Homji, C. B. and Bromley, A. F. Gas turbine axial compressor fouling and washing. In Proceedings of the 33rd Turbomachinery Symposium, Houston, Texas, 2004, pp. 163191. 13 Howell, A. R. Fluid dynamics of axial compressors. Proc. Instn Mech. Engrs, 1945, 153, 441452. 14 Dixon, S. L. Fluid mechanics and thermodynamics of turbomachinery, 1998 (Butterworth-Heinemann, Oxford, UK). 15 Gostelow, J. Cascade aerodynamics, 1st edition, 1984 (Pergamon Press, Oxford, UK). 16 Saravanamuttoo, H. I. H., Rogers, G. F. C., and Cohen, H. Gas turbine theory, 5th edition, 2001 (Prentice Hall, Essex, UK). 17 NASA SP-36. Aerodynamic design of axial-ow compressors. Scientic and Technical Information Division, National Aeronautics and Space Administration, Washington, District of Columbia, 1965. 18 Taylor, H. A guide to surface texture parameters (manual), 2004 (Taylor Hobson Limited, Leicester). 19 Ramsden, K. W. Axial compressor design and performance. Course notes, 2006 (Craneld University, UK). 20 Howell, A. R. and Bonham, R. P . Overall and stage characteristics of axial-ow compressors. Proc. Instn Mech. Engrs, 1950, 163, 235248. 21 Howell, A. R. Design of axial compressors. Proc. Instn Mech. Engrs, 1945, 153, 452462. 22 Pachidis, V., Pilidis, P ., Talhouarn, F., Kalfas, A., and Templalexis, I. A fully integrated approach to component zooming using computational uid dynamics. ASME J. Eng. Gas Turbines Power, 2006, 128, 579584. 23 Celis, C., Long, R., Sethi, V., and Mangion, D. On trajectory optimization for reducing the impact of commercial aircraft operations on the environment. ISABE-2009 1118, 2009. 24 Pilidis, P. Gas turbine theory and performance. Course notes, 2002 (Craneld University, UK).

k ks ks,adm K l N p P Pamb R Ra Re c Rq Rz s T Tamb U V W Subscripts 0 1 2 3 4 a c dp e m p s th

roughness parameter equivalent sand grain roughness admissible sand roughness roughness height (particle size) assessment length engine rotational speed static pressure total pressure ambient pressure stage reaction arithmetic average roughness Reynolds number based on true chord and inlet conditions root mean square roughness ISO ten-point height roughness parameter blade pitch total temperature ambient temperature blade speed velocity mass owrate air ow angle blade metal angle yaw probe wedge angle change in value efciency work done factor density

APPENDIX Notation c cp CD CL H blade chord specic heat at constant pressure drag coefcient lift coefcient blade height, enthalpy

stator outlet rotor inlet, cascade inlet rotor outlet, cascade outlet stator inlet stator outlet axial, annulus compressor design point engine compressor inlet mean prole, peak, passage, and polytropic secondary, stage thermal

Abbreviations CW PR PS SS TW compressor power pressure ratio pressure surface suction surface turbine power

Proc. IMechE Vol. 224 Part A: J. Power and Energy

JPE992