Use of a Tracer Gas Technique to Study Mixing in a Low Speed Turbine

AN ASME PUBLIQATION $4. pet copy $2. to ASME Members 81-GT-86 THE AMERICAN SOCIETY OF MECHANICAL ENGINEERS 345 E 47 St., New York, N.Y. 117 The Society shall not be responsible for statements or opinions advanced in papers or in discussion at meetings of the Society or of its Divisions or Sections. or printed in its publications. Discussion is printed only if the paper is published in an ASME Journal or Proceedings.Released for general publication upon presentationf uil credit should be given to ASME. the Technical Division, and the auttior(s). Copyright 1981 by ASME J. D. Denton Whittle Laboratory, Cambridge University Engineering Department, England S. USW Technical Research Centre, Nippon Kokan K.K., Yokohama, Japan Use of a Tracer Gas Technique to Study Mixing in a Low Speed Turbine A method of using a flame ionization detector to study the movement of air containing a small concentration of ethylene is described. Ethylene is chosen because it has almost the same density as air so buoyancy effects are negligible. The technique is applied to flow in a bent duct and in a low speed air turbine. In both cases large scale migrations of the end wall boundary layers onto the suction surfaces are observed. However, in the turbine the span wise movement and mixing of the flow at mid-span is remarkably small. INTRODUCTION Although it seems likely that the flow in a turbomachine is highly turbulent and non-axisymmetric, there have, to the authors' knowledge, been no studies of the rate at which fluid is mixed or transported from one area of the annulus to another. Such flow migrations occur by a combination of turbulent diffusion and stream surface distortion, the latter arising mainly from secondary flows generated near the end walls. The effect of these secondary flows on turbine performance is well known but not well understood, they are also important as regards the migration and dissipation of "hot spots" leaving the combustion chamber and of cooling air ejected from the blades. Similarly the rate at which flow nonuniformities, such as tip leakage jets or regions of separation, are mixed out is of importance in the design of downstream blade rows but little quantitative information on the rate of mixing is available. In calculations of the flow through turbomachines (e.g. by the streamline curvature method) it is usual Contributed by the Gas Turbine Division of THE AMERICAN SOCIETY OF MECHANICAL ENGINEERS for presentation at the Gas Turbine Conference & Products Show, March 9-12, 1981, Houston, Texas. Manuscript received at ASME Headquarters December 11, 198. Copies will be available until December 1, 1981. to assume that fluid particles move along stream surfaces and to completely neglect the effects of stream surface distortion and turbulent diffusion. The effects of this approximation on the resultant solutions has never been fully investigated. However, it is found that throughflow calculations which concentrate losses near the end walls (e.g. Denton (9)) may fail to give reasonable results, or even break down completely, if some allowance for diffusion of loss between stream surfaces is not included in calculations on multistage machines. The amount of mixing introduced to overcome this problem is purely arbitrary as there is no data on which to base the diffusion rates. The early studies of Herzig and Hansen (8), using smoke to visualise the flow, show how complex were the mixing processes in the end wall boundary layers and in the blade wakes. The secondary flows induced by the former have been intensively studied in 2D cascades and an understanding of the flow processes is now emerging (e.g. Langston et al (6) Sjolander (7)). These cascade studies are based on measurements of flow velocity and direction and on surface flow visualisation. It would be useful to supplement them with direct tracing of the paths of fluid particles so that the origin of low energy fluid can be located. Smoke can be used to give some indications of this but it diffuses rapidly in turbulent flows and cannot give a quantitative measurement. It is also difficult to use smoke inside a turbomachine, as opposed to a cascade, because of difficulties in providing sufficient illumination. These problems can be overcome by injecting a tracer gas, whose time average concentration can be measured at any point downstream of the point of injection, to study the mean motion and rate of diffusion of the fluid. The only known use of such a technique in turbomachinery is the work of

I Kerrebrock and Mikolajczak (3) who used helium and a between the chimney and flame chamber and of pre-mixing conductivity probe to study the migration of rotor the flows of sampled air and hydrogen before they enter wakes inside a stator row. The use of a conductivity the flame chamber (Fig.2). probe demands fairly high concentrations of a gas with a different molecular weight (and hence different density) to air. At low flow speeds a gas TO SUCTION PI"" with density different to that of the mainstream is subject to buoyancy effects which may distort its motion. Hence it is preferable to base a tracer gas system on a gas with similar molecular weight to air and also to be able to work with very low JOINING CHIMNEY concentrations of the gas. Such a system could COMPOi N provide a useful supplement to the usual flow LAME CHAMBER velocity and direction measurements obtained in cascades and in turbomachines. IGNITOR TO A."PLIFIF.P GAS DETECTOR SYSTEM The method of measuring the concentration of tracer gas is based on a commercially available flame ionisation detector (FID) which has the advantage of very high sensitivity, being able to detect mass flow rates of hydrocarbon gases as low as 1 1 gm/s (i.e. concentrations of a few p.p.m.). The system comprises the following components connected as shown in Fig. 1. 1. Flame Ionisation Detector (Perkin-Elmer F 11) 2. Amplifier 3. Suction pump 4. Flow meters 5. Flow control valve 6. Air compressor 7. Hydrogen and ethylene cylinders 8. Sampling probe 9. Tracer gas injection probe. AMPLIFIER REF Fig. 1 `,1 CTION PUMP... FID SATI I L.( PROBE 7 I AIR COMPRESSOR HYDROGEN CYLINDER 111 INJECTION P TORE ETHYLENE CYLINDER Complete tracer gas detection system The FID was originally designed as a component of a gas chromatography system and needed some modification to enable it to be used in a situation where a sample of air is sucked through it. The modifications consisted simply of sealing the gap ATI, --31. SAMPT r Fig. 2 Flame ionisation detector ELECTRODE POLARTS INC Vni.TPCE HYDROGEN As shown in Fig. 1 air, containing a small quantity of tracer gas, is sucked into a sampling probe inserted into the flow being studied. The flow rate of the sampled air is monitored by a flowmeter and is kept constant throughout the tests. The sample is mixed with the flow of hydrogen and burnt in the flame chamber. Any hydrocarbon gases present produce ions which are subject to a potential difference between the jet of the flame tube and a collector electrode nearby and so produce a small electric current. This current is directly proportional to the mass flow rate of hydrocarbon gas in the sample. Since the sample flow rate is held constant the amplified current is directly proportional to the concentration of tracer gas in the sample. Ethylene was chosen as the tracer gas since it has almost the same density as air (molecular wt 28). Although it is highly inflammable at high concentrations the flow rates of ethylene used (about 8 cc/min) were so low compared to the flow rate of the air in the turbine as to present no danger. The response time of the system is determined by the time taken for the sample to flow from the sampling probe to the flame chamber which in turn depends on the flow rate of the sampled air and the volume of the connecting system. The latter was kept small by using the minimum length (2.5 m) of 1.5 mm I.D. flexible tube. The maximum flow rate was limited by the rtability of the flame to about 9 cc/min, tolls gave a response time of 3-4 seconds and so enabled a large number of points to be sampled on any one run. The linearity of the output signal was checked by varying the flow rate of the tracer gas whilst holding the sampling probe fixed. The result is shown in Fig. 3 for each of the two amplifier ranges used in the experiments. Any kind of open ended tube can be used as the sampling probe. For the tests reported here conventional total pressure probes were used for convenience. With a single hole spherical head 2

probe it was found that the reading was insensitive to the flow inclination for incidences of up to ± 3 from the hole axis, so long as the sampling flow 1 6 amplifier range Range A Range B x z c/ / LL P4o LL a 2 /..--X.---'4(----'X...' x x -A--x-'-x,...-X x)- - 5 ---- 1 2 3 SCALE PROPORTIONAL TO THE AMOUNT OF GAS Fig. 3 Linearity of detector system rate is held constant. Such a probe was used for measurements in the duct. For measurements in the air turbine a 5 hole spherical head probe of 1.5 mm diameter was used with the central hole used to sample and the two yaw holes approximately nulled so as to direct the flow into the central hole. The only requirement of the system used to inject the tracer gas is that it should cause minimum disturbance to the flow. A 3 mm.d. 'L' shaped tube was used in the turbine and 3 mm holes in the duct walls were used in the bent duct. MEASUREMENTS IN A BENT DUCT A bent duct with 12 turning and rectangular cross section was used in developing the system. The results obtained are of interest because they can be compared with detailed measurements of the secondary flow in the duct reported by Bruun (2). A view of the duct is shown in Fig. 4. The inlet is connected directly to the exit contraction of an open circuit wind tunnel giving inlet wall boundary layer thicknesses of about 17 mm. The mainstream velocity is about 19 m/s. The ethylene was injected through 3 mm holes drilled through the duct wall at positions shown as A, B, C in Fig. 4. The sampling probe was an 'L' shaped 3 mm diameter spherical head probe which was traversed across the exit of the duct. Contours of concentration measured at duct exit are shown in Fig. 5. When the tracer was injected at the centre of the duct cross section (point A in Fig. 4) the contours (Fig. 5A) are almost circular and show little mixing. The slight elongation to the right hand side of the contours may be an effect of the probe stem whilst the outward displacement of the peak relative to the point of injection is due to the secondary velocities in this region. (see Bruun (2)). 4 Fig. 4 Layout of bent duct Tracer gas injected into the wall boundary layers is subject to strong secondary flows. Fig. 5B shows that fluid in the lower wall boundary layer moves inwards along the lower wall and then up the inner wall to about i of the half height. It must be concluded that very little of the fluid entering in the inner half of the lower wall boundary layer remains on the lower wail. Bruun's (2) measurements show that strong secondary flow vortices develop which, in the lower half of the duct, sweep the wall boundary layer round in an anticlockwise direction. This is confirmed by Fig. 5C which shows the result of injecting tracer gas on the inner wall of the bend. Fluid entering in the inner wall boundary layer first moves up the wall towards the centreline and then separates from the wall and moves outwards to become entrained into the passage vortex. The long concentrated 'tail' of the contour in Fig. 5C reveals this vortex motion. Fig. 5 Concentration contours at exit of bent duct. Points of injection A,B,C as marked on Fig. 4. 3

Similar meausrements have been obtained with other points of injection of the tracer gas and the results are in broad agreement with Bruun's measurements of the secondary flow. Rather surprisingly, when tracer was injected into the outer wall boundary layer no strong migration was detected. diffusion of the tracer gas is remarkably small at mid-height although there is a slight upwards elongation of the contours caused by the turbulence behind the probe stem. On the hub and casing there is more diffusion, as expected, both because of the higher turbulence levels in the boundary layers F E D CB 118 SCALE M 1 CENTRIFUGAL FAN THROTTLE \ GUIDE VANES RESEARCH TURBINE STAGE FLOW Fig. 6 Layout of low speed air turbine MEASUREMENTS IN AN AIR TURBINE The low speed air turbine used is described in detail by Hunter (1). It compric.es an inlet contraction, single stage turbine, large centrifugal fan and throttle as shown in Fig. 6. The turbine is of free vortex design with no swirl at entry or exit so that in the absence of viscous effects and secondary flows the axial velocity would be everywhere uniform. The flow in the turbine has been measured in great detail by Hunter (1) whose results provide a basis for interpreting the tracer gas measurements. For these tests the turbine was operated at its design point (i.e.with zero exit swirl) and with trip wire at exit from the contraction to ensure turbulent annulus boundary layers. The thickness of these boundary layers was measured at 21.3 mm at the hub and 22.1 mm at the casing just in front of the nozzle blades. There were no trip wires on the turbine blades, however, and as shown by Hunter (1) this leads to largely laminar boundary layers on the nozzle blades. The inlet axial velocity was 17.5 m/s giving incompressible flow with Reynolds numbers of 4.4 x 1 5 and 3.3 x 1 5 for the stator and rotor respectively. Mixing Without Blades In order to provide a basis of comparison for the mixing within the blade rows, measurements were first made of the rate of mixing of the tracer gas in the annulus upstream of the blades. The positions of injection and sampling are shown as points A and B in Fig. 6, the distance between the points being 48 mm. Fig. 7 shows results when the tracer was injected at mid-height and on the hub and casing. The and because of the longer transit time of the lower velocity fluid. The distance travelled by the gas in this case is similar to that used for subsequent measurements in the turbine so the spread of the contours in Fig. 7 provides a minimum which can be compared with that found in the flow through the blade rows. 15 75 15 CASING Fig. 7 Mixing in annulus upstream of turbine blades 4

Mixing in the Nozzle Blades To study the flow through the turbine nozzle blades tracer gas was injected at an axial location shown as point C in Fig. 6 and sampled at point D. The gas was injected at various circumferential positions relative to the nozzle leading edge. When the tracer is injected at mid-height into a streamline passing very close to the nozzle leading edge stagnation point the concentration contours downstream are shown in Fig. 8A. It is surprising that the mixing is only slightly greater than that in the uniform upstream flow despite the fact that the gas was subject to the turbulence in the blade wake. Also surprising is the fact that the pitchwise diffusion is no greater than that in the spanwise direction. Some radial inwards flow is known to occur in the nozzle boundary layers and wake (Hunter(1)) but there is little sign of this in the concentration contours. The reason that no inward C A S I NG ----- APP. POSITION OF WAKE fluid is transported towards the suction surface but here the peak in concentration is less prominent and the contours spread more widely in the pitchwise direction. This is surprising since Hunter's results show a more intense and concentrated vortex at the hub than at the casing. The spread of the contour to the right of Fig. 8c may be explained by the considerable overturning of the flow near the end wall but the spread to the left cannot be easily explained. It may be relevant that, because of the higher turning on the hub, the flow path from the trailing edge to the sampling point is 1.36 times that at the casing. The preceding results show that fluid entering the nozzles on the endwalls is transported towards the suction surface of the blades. This fluid must be replaced by fluid entering just off the endwalls which is transported towards the walls and forms new endwall boundary layers. It is instructive to try to find the point of origin of the fluid in the new endwall boundary layers. To this end tracer gas was injected at points 7.5, 15 and 25 mm from the endwalls into streamlines passing close to the pressure surface (boundary layer thickness about 22 mm). With injection at 7.5 mm from the walls the results were simil a r to those of Fig. 8. With injection at 25 mm from the walls no strong cross flow could be detected. However, interesting results, shown in Fig. 9, were obtained with injection at 15 mm from the walls. The contours shot,- a peak close 25 2 1 T" A CASING Fig. 8 Mixing through nozzle blades movement is detected is probably because the boundary layers on the nozzle are extremely thin. Cascade tests show the displacement thickness at the trailing edge to be about.7 mm. Hence the amount of tracer entrained into the boundary layers and wake may be too small to be detected. Similar results were obtained with injection midway between adjacent blades at mid-height. Figure 8B shows the result when gas was injected on the casing into a streamline passing close to the nozzle pressure surface. The point of injection is marked N in the figure. It can be seen that the fluid on this streamline has been moved tangentially through a considerable distance and most of it ends up on the suction surface displaced slightly inwards from the casing. The position of the peak contour coincides with that of the vortex observed by Hunter. Tracer gas injected on the casing at different pitchwise positions gave similar results hence it can be concluded that all the fluid entering the nozzle row close to the casing wall ends up in the core of the passage vortex. The situation is only slightly different at the hub as shown in Fig. 8C. Again the boundary layer Fig. 9 Mixing through nozzle blades with injection 15 mm from annulus walls to the point of injection but displaced from it towards the wall, and secondary peaks on the wall in the wake of the adjacent blade. The two peaks are connected by contours showing appreciable concentration on the end wall over the whole pitch. It is concluded that fluid entering the nozzle row at about half the inlet boundary layer thickness from the walls moves first radially towards the walls and then onto the walls to form the new wall boundary layers. There is very strong crossflow in the new boundary layer and some of the fluid entering it is transported across the passage to the suction surface of the adjacent blade where it forms the second peak in the contours. This fluid is not entrained in the passage vortex as was fluid entering closer to the walls. These findings are in agreement with the results of Langston et al (6). 5

Mixing Within the Rotor Blades At a fixed point behind the rotor all flow properties vary periodically at blade passing frequency, about 45 Hz in this turbine. The detector system cannot follow such a high frequency and the measured concentration can be expected to be a time average value at the sampling point. For a fixed blade row, in the absence of secondary flow, the circumferential displacement of all streamlines passing through the row will be the same between points at the same axial positions upstream and downstream of the row. For a rotating blade row the absolute flow is unsteady and we must consider path lines rather than streamlines. The path lines of fluid particles originating at the same fixed point upstream of a rotating blade row differ for different relative positions of the point and the blades at the time of particle release. If we consider the flow relative to the blade row the 'source' of particles moves circumferentially and the particles it releases move on different relative streamlines through the row. Those moving along relative streamlines near the blade pressure surface take longer to pass through the row than those near the suction surface. Although the circumferential displacement of all the relative streamlines is the same in the relative frame, on transfering back to the absolute frame the differences in transit time, At, will cause different absolute circumferential displacements, U At (where U is the blade speed), of the path lines. Fluid particles passing near to the pressure surface will therefore be displaced circumferentially in the direction of rotation relative to those passing near the suction surface. The time difference At can only be calculated accurately from numerical blade to blade solutions but Smith (4) gives an approximate method of estimating it from the blade geometry and circulation. Fluid entering the rotor boundary layers should have a greatly increased transit time and hence a large circumferential displacement. These effects make interpretation of the tracer gas measurements behind the rotor much more difficult than those behind the nozzles. When studying flow through the rotor tracer gas was injected at axial position E and sampled at point F in Fig. 6. The point of injection could be moved circumferentially so as to lie either within or between the nozzle wakes. With injection at mid-height midway between nozzle wakes the results shown in Fig. 1A were obtained, the point of injection is off the graph to the left. As expected the contour is drawn out in the direction of rotation but the radial diffusion is no greater than that through the nozzle blades. The circumferential extent of the contour coincides very closely with that obtained from Smith's (4) method of estimating the difference in transit time, which in this case was about 4.5 x 1 3 seconds. The contours show no sign of the long 'tail' expected from fluid entrained in the rotor boundary layers although very low concentrations of gas were detected well to the right of the contours shown. This again implies that the surface boundary layers are very thin and so only a very small quantity of tracer gas is entrained into them. Results from tests with injection into the nozzle wake at mid-height are similar except for a radial shift of the peak towards the hub, probably caused by radial inflow in the wake upstream of the rotor. DIRECTION OF ROTATION Fig. 1 Mixing through the rotor blades Figure 1C shows the result of injecting tracer into the hub boundary layer. There is increased mixing in both radial and circumferential directions compared to the nozzle row. The latter is expected from the difference in transit times and the former is consistent with the much stronger passage vortex which Hunter found at the rotor root. This in turn is a result of the higher turning. The peak of the contour again coincides closely with the centre of the vortex. The origin of the smaller peak on the hub itself is not completely clear, its location to the left of the main peak suggests that it may contain fluid which bas been greatly over-turned in the new hub boundary layer. Injection of gas on the casing produces the contour pattern of Fig. 1B. There is a well defined peak well off the end wall and the circumferential spread is less than at midspan due to the reduced transit time difference at the tip. The 'tail' of the contour extending towards the casing on the right hand side may be due to tip leakage fluid which is underturned by the rotor and leaves it with swirl in the direction of rotation or maybe simply the result of increased transit time in the casing boundary layer. Mixing Through The Whole Stage The final stage of the experiments was to study the mixing through the whole stage. Tracer gas was injected at point C (Fig. 6) and sampled at point F. Results are shown in Fig. 11. At mid - span the diffusion is similar to that through the rotor alone but with an outwards displacement of the left of the contours. This displacement could be due to decondary flow causing fluid near the suction surface to move outwards even at mid span, but if so it should also have been detected in the flow through the rotor alone. Fig. 11C shows contours obtained when the tracer was injected on the hub. These are significantly different from those of Fig. 1C the peak in concentration now being on the hub. This is explainable by the fact that, as shown by Fig. 8C, 6

Fig. 11 Mixing through the whole stage at entry to the rotor, the gas will have moved slightly away from the hub, the rotor secondary flow must then bring much of this fluid back to the hub to replace the rotor inlet endwall boundary layer. The low concentrations of tracer in the hub boundary layer at entry to the rotor will be entrained into the_passage vortex and may account for the low peak which was recognisable at the point marked N in Fig. 11C. This point coincides with the peak in Fig. 1C. At the casing, Fig. 11B shows the peak of the contour at the same point as in Fig. 1B but with the major axis of the contours tilted upwards to the right. This tilt is consistent with strong inwards flow on the rotor suction surface and outwards flow on the pressure surface coupled with underturning of the tip leakage flow. It is therefore probably a complex result of the combined secondary and tip leakage flow at the rotor tip. DISCUSSION AND CONCLUSIONS The gas detection system described can be used to measure the concentration of any hydrocarbon gas in a sample of air. Ethylene is the most suitable tracer gas for use when buoyancy effects need to be minimised, as when tracing the flow through a turbomachine. The system responds to the mass flow rate of tracer gas in the sampled air and so it is important to keep the sampling rate constant. The rate of hydrogen supply to the flame should also be maintained fairly constant as this affects the +he Aetector. When these conditions are satisfied the system responds linearly to the concentration of hydrocarbon tracer gas in the sampled air and so can be used for quantitative studies of mixing processes in any air flow. The results presented show that the system can be used to obtain useful information about the mixing processes in turbomachines. The time required to obtain each contour plot was about 1; hours using manual traversing and recording of the results. The results obtained are broadly consistent with the measurements of flow in the turbine obtained by Hunter (1). The amount of mixing at mid-span seems surprisingly small but it must be remembered that this is a lightly loaded, single stage machine with low turbulence at inlet and very low losses (total-total efficiency about 93%). It is to be expected that more highly loaded turbines, particularly those with lower aspect ratio and higher inlet turbulence, will have much more intense mixing. The radial migration of fluid is also remarkably small at mid-span which suggests that only a very small proportion of the flow is entrained into the blade boundary layers and wake where spanwise flow is known to occur. It would be a useful extension to experiment with direction injection of tracer gas into the blade boundary layers and from the trailing edge into the wake. The mixing near the endwalls is much more intense than at mid-span and the measurements confirm that much of the fluid in the inner half of the annulus boundary layers is swept completely off the end walls and ends up in the core of the passage vortex. At exit from a blade row it is found that the fluid on the endwalls originated in the outer part of the inlet boundary layer. At exit from a rotor row interpretation of the concentration contours is made difficult by the pitchwise spread induced by the different'transit times of fluid particles on the pressure and suction surfaces. Smith's approximation is found to predict this spread with remarkable accuracy at mid-span. ACKNOWLEDGMENT The authors would like to express their gratitude to Mr. E.C. Deverson and his staff at the Whittle Laboratory for help in developing the detection system and with operating the turbine. REFERENCES 1 Hunter, I.H., "Endwall Boundary Layer Flows and Losses in Axial Turbomachines". Ph.D. Thesis, Cambridge University, September 1979. 2 Bruun, H.H. "An Experimental Investigation of Secondary Flow Losses in Bends with Rectangular Cross Sections" Cambridge University Engineering Department, CUED/A-Turbo 95, 1979. 3 Kerrebrock, J.L., and Mikolajczak, A.A. "Intra-Stator Transport of Rotor Wakes and its effect on Compressor Performance" ASME Paper 7-GT-39. 4 Smith, L.H. "Secondary Flow in Axial Flow Turbomachinery" Trans. ASME, 77, 1955. 5 Perkin Elmer Ltd.,"Introduction to Gas Chromatography". Manual for Flt Gas Chromatography System. 6 Langston, L.S., Nice, M.L., and Hooper, R.M. "Three Dimensional Flow Within a Turbine Cascade Passage" ASME Paper 76-GT-5. 7 Sjolander, S.A. "The Endwall Boundary Layer in an Annular Cascade of Turbine Nozzle Guide Vanes" Tech. Rept. No. ME/A 75-4, Carleton University, December 1975. 8 Herzig, H.Z., and Hansen, A.G. "Visualisation Studies of Secondary Flow with Application to Turbomachines". Trans. ASME, 77, 1975 9 Denton, J.D. "Throughflow Calculations for Transonic Axial Flow Turbines" ASME J.Eng. Pow., 1, April, 1978. 7