THE AMERICAN SOCIETY OF MECHANICAL ENGINEERS 345 E. 47th St., New York. N.Y

Transcription

1 THE AMERICAN SOCIETY OF MECHANICAL ENGINEERS 345 E. 47th St., New York. N.Y ;:GT;50 The Society shall not be responsible tdi state lments or opinions advanced In papers or eascussionat meetings otthe'societi or ads DiVisiens or Scclices, or printed in its publications. Discussion is printed only It the paper is published in an ASME Journal Authorization to photocopy for internal or personal use is granted to libraries and other users registered with the Copyright Clearance Center (CCC) provided tillarficle ois4dpage is paid to CDC, 222 Rosewood Dr., Danvers; MA Requests for special perrnlision'or bulk reproduction should be addressed to the ASME Technical Publishing Department. CoPYriSdlt 0 tees by ASME AN AGARD WORKING GROUP STUDY OF 3D NAVIER-STOKES CODES APPLIED TO SINGLE TURBOMACHINERY BLADE ROWS John Dunham consultant, UK II I Georges Meauze ONERA, BP72, Chatillon Cedex, France ABSTRACT Computer codes which solve the Reynolds-averaged Navier-Stokes equations are now used by manufacturers to design turbomachines, but there is no consensus among experts about which grids and which turbulence models are good enough to provide a reliable basis for design decisions. The AGARD Propulsion and Energetics Panel set up a Working Group to help to clari& these issues, by analysing predictions (using as wide a range of codes as possible) of two representative but difficult single blade row test cases: NASA Rotor 37 and an annular turbine cascade tested by DLR. This paper summarises the Group's results and conclusions. Recommendations are made about the type and density of grid, which depend on many factors. Although mixing-length turbulence models give good results for quasi-two-dimensional boundary layers, they are essentially unsuitable for turbomachines with their complex endwall flows; it is essential to adopt some kind of turbulent transport model. INTRODUCTION Computational methods for the design and analysis of turbomachine flows have been developed and brought into use progressively over the last fifty years. Codes which solve the Reynolds-averaged Navier- Stokes (RANS) equations in three dimensions are already used by the large engine manufacturers for the advanced design of many engine components. It is, however, recognised by experts in the turbomachinery field that the codes cannot yet be relied upon to predict the efficiency and all the details of the flow except on a comparative basis. The key to improving the predictions is to understand the physical phenomena involved and the features of the codes which model them. The AGARD Propulsion and Energetics Panel has regularly organised.activities on turbomachinery flow prediction methods, and Working Group 26, "CFD validation for propulsion system components", was started in The Group comprised experts in this field from NATO nations, listed in Appendix I. It decided to base its work on two representative turbomachinery configurations. In 1993, the ASME Turbomachinery Committee had issued an open invitation to predict the flow details of an isolated transonic fan rotor, NASA Rotor 37. This proved a challenging case, so the WG decided to select it as one of the two cases; the other was an annular turbine cascade tested by the German Research Establishment DLR. The objectives of the Group were to obtain CFD calculations of the specified flows, and to evaluate the results and the methods used for the purpose of advancing the technology. It was planned to study the effectiveness of the grids and turbulence models used in the various CFD codes and to make recommendation's for the guidance of code developers, and for future research. It was not intended to attempt to rank the specific codes in order of merit or to recommend to manufacturers a "best" code. The Group has reported its work in AGARD-AR-355, expected to be published well before June The present paper has been prepared from that report with the encouragement of AGARD, to enable its conclusions to be discussed by a wider range of specialists. THE TEST CASES The two test cases enabled the Group to assess the ability of CFD to predict several of the aerodynamic phenomena which dominate the performance of turbomachines. The Rotor 37 case included shock waves, corner stall, and tip clearance effects in a fan with supersonic inlet relative Mach number across the whole span. The DLR annular reseede showed the effect of spanwise pressure gradients on the Presented at the International Gas Turbine & Aeroengine Congress & Exhibition Stockholm, Sweden June 2 June 5, 1998

2 laser anemometer survey stations Stet Sails 2 St, 3 Sto.Stal pressu re ratio adiabatic efficiency % Hub usage Ma Aidal Direction. an Ian Fig 1 NASA Rotor 37 measuring stations T959b span -90% span span \SO% Wan 3016 span * measured TRANscode (Baktesm-Lornax) TRANSCock (Spate) 6 CANARI (k-c) g OVERFLOW (Spatart) mass flow! choked mass flow Fig 3 Overall performance of Rotor 37 complex secondary flow field. The test cases did not, on the other hand, include important transition effects, heat transfer, or high incidence conditions, nor of course the three-dimensional separations found within a centrifugal impeller. More importantly, the Group (by choosing isolated blade rows) did not attempt to cover the difficult area of row-to-row interference due to unsteady flow. The test programmes were carried out in well-developed test installations.)17 Fig 2 DLR cascade measuring stations operated by experienced research groups. In both programmes a wide range of measurements was made, using proven instrumentation and data acquisition methods, and reduced by accepted and traceable procedures. The Rotor 37 test case is well known. It was part of a research programme involving four related axial-flow compressor stages intended to cover a range of design parameters typical of aircraft turbine engine high-pressure (core) compressor inlet stages. Design and overall stage performance results were reported by Reid and Moore (1978) and by Moore and Reid (1980). A particular operating point was chosen for traverses to be compared with predictions; this point was at design speed, 98% choked flow, and an overall pressure ratio of 2.1. At this condition, the relative inlet Mach number is supersonic across the whole of the span. Measurements were made using conventional probes and using a laser anemometer, at locations shown in Fig.l. Full details are available on a floppy disc from NASA Lewis Research Center. The DLR annular cascade is a scaled version of a subsonic low aspect ratio turbine nozzle guide vane row. Measurements were made at a particular operating point (outlet Mach number 0.74) with a 5- hole probe downstream of the stator and with a three-dimensional Laser-Two-Focus anemometer inside the passage and downstream. Additionally the upstream boundary layer was traversed using a "fishmouth" pitot probe. The measurement planes are shown in Fig.2. Full details of the case are available on a floppy disk from DLR. THE COMPUTATIONS The members of the Working Group invited research workers to undertake computations of the test cases using any 3D RANS codes they wished to use. Computations using sixteen different codes were supplied to the WO, in some cases with a range of grids and turbulence models. They are listed in Table 1. Not all the codes in common use in industry were included, but most of the important types were represented. In all, a wide range of grids and turbulence models were employed. The predictions were not "blind"; nevertheless, the overall performance of both test cases predicted by most codes fell short of the accuracy engine designers need as a basis for decision-making. In both cases, it proved difficult to clarify all the details of the flow from the measurements alone, but with the help of the CFD results a reasonably complete understanding of all the flow fields was reached. The secondary flow in a turbine cascade has been well known for many years, but the complexities of the Rotor 37 flow have 2 2

3 OA 0.2 y 41 I ; V. I. :es ; tr 41: eq. 4 measured TRANSCode (Baldwin-Lomax) TFtANSCode (Sealan) CANARI (k-e) OVERFLOW (Spatart) pressure ratio temperature ratio adiabatic efficiency % Fig 4 Rotor 37: spanwise distribution of pitchwise-mean performance (mass-averaged) 1 A Os Os Esimrnort a TRACES (72K Grid) (2SEK Odd) e TRACE_S (500K GAM T TASOlow (72K 01Ic0 20 O to 80 relative pit0/1 % Fig $ Mach number variation across pitch at mid-span (Hildebrandt) only recently been illuminated by experiments and CFD studies conducted by NASA Lewis. Rotor37 Fig.3 shows some typical overall performance predictions, in comparison with the measurements. It will be seen that the shape of the curves, and the flow range, are well predicted, but the levels of pressure ratio and efficiency are widely spread. Looking at all the predictions (not just those plotted), the spread of pressure ratio predictions was 0.18 and the spread of efficiency predictions was 6%. Generally, the pressure ratio was too high and the efficiency too low. 100 Fig.4 shows the pitchwisc-mean pressure ratio, temperature ratio and efficiency levels at 98% choked flow, for the same codes. Denton (1996) has given a good global analysis of the flow in this compressor. Chima (1996) and Suder and Celestina (1996) analysed the tip leakage flow, and Hah and Loellbach (1997) and Shabbir et at (1997) analysed the hub corner flow. The mid-span region is dominated by a strong shock attached at the blade leading edge. This shock interacts strongly with the suction side boundary layer. The boundary layer after the shock may separate either up to the trailing edge according to some authors, or reattach before the trailing edge. The computed relative Mach number is always lower downstream of the shock than its measured value, and the shock is located too far upstream (provided a fine mesh is used), implying that the mass flow used in the simulation is too low. Chima's simulations have shown that an increase of 0.24% for the mass flow (from 98% to 98.24%) is sufficient to fit the experimental blade-toblade Mach number distribution. This value (0.24%) is also of the same order as the difference between the avenged choked mass flow deduced from the simulations (20.86 kg/s) and the experimental value (20.93 kg/s). Perhaps this explains the discrepancy in shock location. Alternatively, it may be that the shock/boundary interaction modelling is inadequate, or that inaccuracies in predicting the flow in the endwall regions have led to errors in the spanwise distribution of the flow, and hence an overestimate of the axial velocity around mid- span. Another feature of the mid-span predictions is that at station 3 the wake is always too deep (Fig.5), while at station 4 the wake deficit is greatly reduced. The experimental deficit is 1/3 lower than in the simulation near the trailing edge. This could be due to the way a laser vclocimeter or a steady CEO code interprets a periodically unsteady wake. The excessive mixing of the wake into the mainstream between stations 3 and 4 is believed to result from using too coarse a grid in the blade-to-blade direction in the wake region. Most of the predictions give a good efficiency distribution at mid-span. 3

4 Predicted rotor exit profiles (with cavity, fine mesh) 1.0 to a to 0.8 c Fig 6 Limiting streamlines, showing corner stall (Kang and Hirsch) total pressure ratio Near the hub wall, the measurements of absolute stagnation pressure at station 4 show a region of low pressure at 20% of the blade span, which is present over a wide range of mass flow at design speed. It is clear that a corner stall occurs near the suction side corner. The resulting flow is illustrated in Fig.6, using one of the codes which predicted its presence. This corner stall reduces the axial momentum in the region of the stagnation pressure deficit near the hub, and this axial momentum is redistributed all over the blade span, thereby reducing the work done by the blades and the overall pressure ratio. NASA found that the flow in the hub region was seriously affected by the presence of a small axial gap (0.75 mm) in the hub annulus line just upstream of the rotor leading edge (because the hub wall ahead of that point did not rotate). Shabbir et al (1997) demonstrated by both measurements and computations that, although the gap led only to a blind cavity, air was pumped in and out of it as the rotor rotated, and this strongly influenced the observed corner stall. The presence of a shock wave even at the hub results in sufficiently large pitchwise static pressure changes to cause the pumping. Shabbir et al (1997) showed that CFD predictions in which the inflow-outflow was simulated predicted the corner stall, while the same code without the inflow-outflow did not This discovery came at a late stage of the Working Group's activities, and only one other member had time to try to model the gap-flow in his code. Without the gap, only a few codes predicted the corner stall, which is extremely sensitive to small modifications of the boundary conditions, the radial mesh distribution, and the turbulence model. The Baldwin-Lomax model predicts corner stall if the empirical constant C.*, is set to its originally proposed value of 0.25, but other cases make it clear that C,k should really be about unity (when a corner stall is not predicted). Fig.7 shows how one code, independent of the NASA study, was also able to predict the corner stall by simulating the flow in and out of the gap; but the corner stall only appeared when a grid was used which was fine in the radial direction. The tip wall region is dominated by a strong tip leakage flow. Chima (1996) and Suder and Celestina (1996) have presented some detailed experimental and CFI) results for the flow in this region. The main conclusions of these studies are that the leakage flow behaves almost as an inviscid jet owing to a strong supersonic expansion; the limiting surface between the tip leakage jet and the main crossflow is a Fig 7 Effect of pressure ratio on predicted flow with cavity modelled (Hall) region of strong shear attached to the blade corner on the suction side; and at the exit from the tip gap the particle traces exhibit clearly a negative axial velocity. The interaction of the clearance vortex with the passage shock leads to a flow separation on the casing wall. The mixing of the leakage flow with the primary flow extends over 10% of the blade span and also probably far downstream. Near the casing, most of the simulations produce a lower efficiency than measured. The high value of the losses seems to appear as a consequence of the interaction between the passage shock and the leakage flow vortex. The flow reaction is probably too abrupt across the shock, which is a common feature of all equilibrium turbulence models, such as the mixing length or linear k-e models. Low values of the turbulent viscosity are predicted near the tip wall, but the resulting separation of the wall boundary layer leads to predicted losses which are too high fluffier away from the wall. The mixing length model generates more losses in the tip region than the k-e model. The interaction between the leakage flow and the primary flow is certainly far more complex than a simple boundary layer situation for which the mixing length model was developed. All the complex physical phenomena described have to be treated along the tip wall, for which a mixing length turbulence model only "sees" a single length and a single velocity scale, while the tip wall flow is in fact dominated by several different lengths and velocities. DLR cascade Fig.8 shows the pitchwise-mean distributions of total pressure and flow angle downstream of the cascade predicted by some typical codes, in comparison with the measurements. When all the predictions supplied to the Group were compared, the overall pressure loss predicted by some of the codes was as much as 40% in error. The measurements of total pressure show that the secondary loss region near the hub is confined close to the wall by the spanwise static pressure gradient, and that the location of the peak loss is generally well predicted, though its magnitude varies very widely between the 4

5 0.8 fraction of span measured TRANSCode (Baldwin-Lomax) TRANSCode (Simian) CANARI k-s ) TRACE-S (extended k- e) pressure ratio limy angle deg Fig 8 DLR cascade: spanwise distribution of pitchwise-mean performance codes. The secondary loss region at the casing, on the other hand, is convected well away from the wall, the measured loss peak being at 55% span. Most codes were unable to predict the location and magnitude of this peak. Fig.9 shows some of the contour plots of total pressure, compared with the measurements, showing the marked distortion of the wake, which only some of the codes reproduced. It is clear that the codes are generally unable to predict correctly the highly three-dimensional secondary flow and the resulting migration of low energy fluid on to the suction surface and then diagonally inwards until its discharge from the trailing edge. CANARI's k-c predictions improved on predictions using a mixing length model. TRANSCode obtained much better predictions with the Spalart turbulent transport model than with the Baldwin- Lomax mixing length model. TRACE-S used two different grid topologies, and also (for a C-H grid) tried several turbulent transport models. The calculations using an 0-H-H-H grid predicted a poor loss distribution, due to difficulties at the junction of the grid regions just after the trailing edge, showing the need to have a fine aligned grid in the wake region; but the effect of changing the turbulence model from k-c to k-to was small. The contour plots of turbulent viscosity revealed remarkably large differences between the turbulence models. The turbulent viscosity computed by TRACE-S with different two-equation models differ by up to an order of magnitude (even in the mainstream region) and the same observation applies to the CANARI results using the Michel and k-c models. It seems possible that some of the differences are due not so much to the turbulence model itself as to the way in which it is coded. Choice of algorithm The algorithms used for the Group were mostly of the time-marching type, but pressure correction methods were also represented. While the details of algorithm must surely control the stability and convergence of the code as well as its running time, the Group has no evidence to suggest that it has any effect on the accuracy of the converged result, if one can be obtained. Grid construction The desirable features of a computational grid are well known: it should be fine, have approximately square cells, and be aligned with the stream. Since these properties are impossible to achieve simultaneously in a turbomachine context, the choice of grid always represents a compromise between the various desirable properties of the grid and the complications introduced into the algorithms when complex mixed grid schemes are chosen. It is also known that the choice cannot be dissociated from the algorithm and the turbulence model. Nevertheless, some general conclusions are possible in the case of a structured grid. No unstructured grid solutions were offered to the Group. In the present test cases, no single grid type stood out as being superior to the others. In general, the 0- and C-grids proved better in the leading edge region, although the present test cases, being at nearly zero incidence, were relatively insensitive to the leading edge region. At the trailing edge, the C-, I-, and H-grids that aligned with the wake provided the best wake definition. The 0-grid solutions provided high resolution of the flow near the trailing edge but diffused the wake too rapidly as the grid opened up further downstream. The most successful grids used by contributors were an H-I-H grid (because it aligned with shock waves) and an overset 0-H grid. The grid lines must be clustered progressively near solid surfaces, and ought to be clustered in any regions of strong shear. All the solutions submitted used grids well clustered near surfaces. The total number of cells varied widely, with several contributors conducting grid refinement studies to establish how many cells were needed to make the solution grid-independent to engineering accuracy. The minimum number of cells must depend on the flow being computed, and the algorithm and turbulence model, so it is difficult to generalise. However, it became clear that in the cases used by the WG at least 50 grid lines hub-to-tip, around 50 blade-to-blade (if wall functions are 5

6 SS PS measured TRACE-S with k-co turbulence model (Lisiewicz) TRACE-S with k-e turbulence model (Lisiewicz) HAH3D with k-s turbulence model (Hah) TRANSCode with Baldwin-Lomax turbulence model (Stapleton) TFtANSCode with Spalart turbulence model (Stapleton) Fig 9 DLR cascade: contours of total pressure ratio 6

7 used), and around 300,000 cells in all are needed if the pitchwisemean performance is to be resolved. If wall functions are not used, a finer grid is needed near the walls. To capture the three-dimensional detail of the secondary flow vortices in the DLR cascade, more than 100 grid lines hub-to-tip may be needed. For overall performance and for blade surface pressure distributions, on the other hand, a grid of around 200,000 cells may be adequate. However, the WG believes that to obtain an accurate solution it is necesssary to have a good turbulence model as well as a large number of grid points.' The "pinched tip" model of the tip clearance region chosen by some contributors is unsatisfactory, in that the actual clearance is not used; some empirical "effective" clearance is chosen instead. From the fully-gridded solutions submitted, the Group was unable to recommend a minimum number of cells within the clearance region. Choice of turbulence model Turbulence models can broadly be divided into algebraic or mixing length types (including the popular Baldwin-Lomax model) and turbulent convection types (including k-e models and one-equation models). All the Rotor 37 solutions contributed to the WG using the Baldwin-Lomax model produced very similar results, and they were generally inferior to the predictions obtained using turbulent convection models in those regions where the flow is separated or highly three-dimensional. It is logical that any turbulence model requiring a "distance to the nearest wall" must encounter serious difficulties in such a region. For flows in which the viscous phenomena are primarily of the nature of a boundary layer, any of the well-known turbulence models are adequate, since they were set up for boundary layers. But the grid needs to be adequately fine near the walls. It is generally considered that for algebraic models a r value less than 5 is advisable unless wall functions are used; for codes in which the calculation extends filly to the wall using a low Reynolds number turbulence model a r value less than I is needed (and as low as 0.1 for heat transfer calculations). In codes using wall functions, y 4 = 50 seems adequate. Massive differences were noted between the values of eddy viscosity predicted by different turbulence models. These differences led to differences in loss prediction which took some of the loss predictions outside an acceptable range of accuracy. It appears that much of the difference may result not from the modelling concept but from the way it is implemented within the particular code. The WG was unable to identify any one turbulence model which always gave good loss predictions. It is well known that this is an area of continuing vigorous research, and it needs to be. Transition predictions were not important for the test cases chosen by the WG, but it is well known that many current CFD codes cannot predict transition or re-attachment satisfactorily. Predictions of the flow downstream of a leading edge separation bubble can be critical to aerodynamic loss prediction. Transition prediction is also the key to good heat transfer prediction. CONCLUSIONS The two test cases chosen proved to be two challenging examples which encompassed some (but not all) of the key aerodynamic features of turbomachinery flows. 18 contributors submitted in all 49 solutions to one or both of them, with a wide range of grid geometries and number of grid points, and several different turbulence models. (Unfortunately, no unstructured grid solutions were offered.) The overall performance of both the rotor and the cascade were predicted, and the spread of predictions and their accuracy levels was much wider than a designer would consider acceptable. With considerable help from the CFD solutions, it proved possible to identify the physical phenomena involved in great detail, and to assess the extent to which the codes could account for them. NASA investigations showed that the hub corner stall observed in Rotor 37 was strongly affected by a small gap in the annulus line, which made modelling of the corner stall particularly difficult. No one code, no single grid scheme, and no one turbulence model proved able to predict all the flow patterns quantitatively; but the Working Group was able to identify some of reasons for this and to suggest the right direction for future research. It is recommended that mixing length turbulence models should be discarded in favour of turbulent transport models. Turbomachinery specialists need to focus on finding which model is best. If losses are to be predicted, around 300,000 well-chosen grid points are needed per blade row, refined in regions of high aerodynamic shear. If no wall function scheme is used, a finer grid is essential near the walls, and hence many more grid points. If fuller details of the three-dimensional flow pattern are needed, a finer radial grid and hence 500,000 or more points in all are needed, depending on the turbulence model. The temperature rise and aerodynamic loss predictions were generally too high in the region of the Rotor 37 measurements affected by the tip clearance flow. Some intensive research should be focused on tip clearance effects. ACKNOWLEDGEMENTS The AGARD Propulsion and Energetics Panel and the present authors express their grateful thanks to the NASA Lewis Research Center and to the DLR for providing the test cases, and so willingly responding to many questions about them; to the contributors of the solutions (Table 1), to the members of the Working Group, and especially to the authors of the chapters in the AGARD report: Prof M.Bassi, W.I.Calvert, Dr R.A.Delaney, Prof F.Leboeuf, and Dr L.A.Povinelli, whose analysis of the results formed the basis of this paper, and Prof Ch.Hirsch, Dr V.Couaillicr and Prof G.K.Serovy. REFERENCES Chima, R.V., 1996, "Calculation of Tip Clearance Effects in a Transonic Compressor Rotor", ASME paper 96-GT-114, accepted for publication in ASME Transactions Denton, J.D., 1996, "Lessons from Rotor 37", 3rd ISAIF Meeting. Beijing Dunham, J. (ed.), 1998, "Propulsion and Energetics Panel Working Group 26 on CFD Validation for Propulsion System Components", AGARD-AR-355 Hah, C. and Loellbach, J., 1997, "Development of Hub Corner Stall and its Influence on the Performance of Axial Compressor Blade Rows", ASME Paper 97-GT-42 Moore, R.D. and Reid, L., 1980, "Performance of Single-Stage Axial Flow Transonic Compressor with Rotor and Stator Aspect Ratios of 1.19 and 1.26, respectively, and with Design Pressure Ratio of 2.05", NASA TP

8 Reid,L. and Moore, R.D., 1978, "Design and Overall Performance of four Highly Loaded, High-Speed Inlet Stages for an Advanced High- Pressure-Ratio Core Compressor", NASA TP 1337 Shabbir, A., Celestina, Mi., Adamczylc, J.J., and Suazisar, A.J., 1997, "The Effect of Hub Leakage Flow on two High Speed Axial Flow Compressor Rotors", ASME Paper Suder, K.L. and Celestina, M.L., 1996, "Experimental and Computational Investigation of the Tip Clearance Flow in a Transonic Axial Compressor Rotor", ASME Jul of Turbomachinery, vol.118, no.2, p.218 Table 1 Contributors of CEO solutions to the test cases contributor affiliation code grid type grid size Tu model Rotor 37 Amon. Florence U. TRAF30 C 8011< Eel Calvert DERA TRANSCarie H 363K B-L H 3631< Spalart China NASA SWIFT HCO 4471< EI-L. Couaillier ONERA CANARI HON 821K Michel OH 577K Michel OH 577K B-L. OH 7251< k-e OH 5771< Seinen Denton Cambridge U. TIP3D H 205K miring length Rah NASA HAK3D I 4471< km Hall Allison ADPAC HOH (no hub gag) 338K B-L HON (hub INIIM 338K 84- Hikasteant Munich T.U. TASCflow HIM 72K k-e TRACE-S HIFI 721< km HIFI 25DK k-c HIH 500K k-c Hindi Brussels U EURANUS/TURBO NH 479K 8-4. Hutchinson ASC TASCrlow H 1031< km H 242K km McNulty Allison ADPAC HON 195K B-I. NON 338K 8-1. HON 403K Shatter NASA VSTAGE H 276K B-4. H 276K SKE H 276K CKE Weber Allison OVERFLOW HON 5213K Simian HO 8861( Spalart DLR cascade Bassi-Sayini Ancona U. IL C COMO 125K kma Bergamo U. C medium 4011< kw Chime NASA RVC3CD C 478K 8-1. Couvilier ONERA CANARI HON 659K Michel HON 6591< km DadonemZe Palma Bail Poly C 125K B-I. Deraon Cambridge U. 1IP30 H 200K mixing length Hah RASA HAN3D I k-e Lisiewicz DLR TRACE-S CH 330K k-e CH 330K extended k-e OHHH 380K k-c CH 330K k-o CH 330K SPeirel Marone Florence U. FLOSSED H 12061< k-m McNulty Allison ADPAC C the 159K 94. C Medium 312K 8-1. C coarse 1226K BM Stapleton OEM TRANSCode H 324K B-1. H 774K B-k. H 850K B-L H 3241< SpaLart Appendix 1 Members of the Working Group Members of the AGARD Propulsion and Energetics Panel: Dr Georges Meauzi (chairman) Prof Mike Bardon Prof Charles Hirsch Keith Garwood Prof Francis Leboeuf Prof lose J.Salva Monfort Prof Walter F.OBrien Dr Eng P.Psaroudalcis Prof Giovanni Torella Prof Dr-Ing Heinrich Weyer non-panel members: Prof Francesco Bassi W.John Calvert Dr Vincent Couaillier Dr Bob Delaney Dr John Dunham (editor) Prof W.G.Habashi Prof Francesco Martelli Dr Lou Povinelli Prof George K.Serovy Prof Ahmet S.Ocer Fr Ca Be UK Fr Sp US It It Ge It UK Fr US UK Ca It US US Tu 8