437

Axial flow compressor design†
S J Gallimore Rolls-Royce plc, PO Box 31, Derby DE24 8BJ, UK

Abstract: The purpose of this paper is to set out some of the basic principles and rules associated with the design of axial flow compressors, principally for aero-engines, as well as the practical constraints that are inevitably present. The thrust is primarily on the aerodynamic design but this cannot be divorced from the mechanical aspects and so some of these are touched upon but are not gone into so deeply. The paper has been written from the point of view of the designer and tries to cover most of the points that need to be considered in order to produce a successful compressor. The emphasis has been on the theory behind the design process and on minimizing the reliance on empirical rules. However, because of the complexity of the flow, some empiricism still remains. Keywords: compressor, aerodynamics, design, computational fluid dynamics (CFD), secondary flows, endwall flows, boundary layers, tip clearance flow, rotors, stators, stall, surge, reaction

NOTATION D U V a DH Dp r incompressible flow dynamic head (0.5rV 2) blade speed axial flow velocity enthalpy rise static pressure rise density

1 INTRODUCTION To write a short paper on axial flow compressor design is a demanding task because of the many decisions and interactions that take place in the course of the design process. Complete books have been devoted to the aerodynamics of compressors (e.g. reference [1]), to which the reader is referred for a more detailed description of the phenomena described here. A whole research industry (with the associated literature) as well as considerable engine company effort is expended trying to improve our understanding of these machines. Teams of engineers from various disciplines are employed and it might as well be said at the outset that it is not a precise science and many judgements have to be made based on experience where the science behind the design does not give accurate answers. This lack of accurate prediction influThe MS was received on 20 February 1998 and was accepted after revision for publication on 18 December 1998. † Invited paper for the Special Issue on Turbomachinery Design published in two parts in the Proceedings, Part C, 1999, Vol. 213, Issues C1 and C2.
C02198 © IMechE 1999

ences the design process, starting with the fundamental requirements of a compressor to pass a certain flow at a given pressure ratio and efficiency with adequate operating range for stable operation. To the author’s knowledge there are no methods currently available that guarantee to predict the absolute values of these quantities to a sufficient accuracy that a new design can be said to be risk free. The lack of predictive capability permeates through the whole design process down to the fine details of three-dimensional and unsteady flow behaviour and necessitates the skill and judgement of the engineer at every step. Notwithstanding this, there have been significant improvements in the predictive capability available to the designer over the years, with improved modelling of the flow physics and increased computing power. This has led to improved designs with more certainty of achieving the desired results, but of course the demands for higher efficiency, reduced cost, etc. continue to push the compressor designer beyond the current established practice. The reason for this lack of certainty in the predictions lies in the complicated nature of the flow through axial flow compressors which is described in the next section. The flow is unsteady and three-dimensional and viscous effects play a dominant role, with separations of the boundary layer flows being common. Unfortunately it is these very features that are difficult to model accurately and hence the uncertainty in the design process. Because of this complexity, assumptions have to be made about the flow in order to make it a tractable problem that can be calculated, but the danger of this approach is that some of the important flow physics will be left out of the model and this can lead to unexpected
Proc Instn Mech Engrs Vol 213 Part C

with more complicated modelling being used as the design becomes more refined. This not only means that the wakes of the upstream blade row pass intermittently through the row but also that the back pressure seen by the row also varies because of the upstream potential effect from the downstream row. throughflow design. These headings will be expanded upon below and used as sections to explore the axial flow compressor design process. The design process is iterative and interactive and can broadly be described as design by analysis. Generally. blading design (two-dimensional ) and blading design (three-dimensional ). Figure 1 is a schematic representation of the flow through an axial flow compressor rotor with some of the significant flow features illustrated. In general the flow is unsteady because of the relative motion between successive blade rows in the compressor. it will turn a different amount than the Fig. Because of the change in rotation between successive blade rows any reduction in axial velocity towards the annulus walls produces a change in inlet angle in the frame of reference of the blade row under consideration. It is also true that the effects of blade rows even further upstream can be detected some rows downstream.438 S J GALLIMORE behaviour of the real machine that was not predicted. and is also subject to the effects of shear on the endwalls and within the flow field. be split into roughly four stages: preliminary design. 1 is obtained. To simplify this paper the design process can. Because this fluid has a different momentum at inlet to the blade than the freestream flow. however. it is hoped that this description will prove sufficient to allow the approximations used in the various design methods discussed later to be appreciated. somewhat artificially. the flow field that the design process is trying to control will be described. The flow into the blade row is radially non-uniform and can be typified as a freestream region and two annulus wall boundary layers where the flow velocities vary towards the endwall values. It uses a range of tools from the most simple mean line methods to the most sophisticated three-dimensional computational fluid dynamics (CFD). 1 Schematic diagram of flows in a compressor blade row Proc Instn Mech Engrs Vol 213 Part C C02198 © IMechE 1999 . Firstly. 2 DESCRIPTION OF THE COMPRESSOR FLOW FIELD The purpose of this section is to give a description of the main features present in the complex flow field of a multistage axial flow compressor. The design can iterate between the various stages of the design process. mechanical design and stress. If these unsteady effects are ignored. While it is not possible to go into all the phenomena in great depth in this paper. In an attempt to minimize this risk a series of rules have been developed over time that attempt to add back in the missing physics but are not necessarily based on that physics. the relative air inlet angles increase towards the walls. There is a continual interaction with other disciplines such as whole engine modelling and performance. the picture of the flow illustrated in Fig. Each has its own part to play in the overall process. The freestream is turned towards the axial direction and diffused through the blade row. This sets up a tangential pressure gradient which then acts on the endwall boundary layer fluid. each of which has its own objectives.

These are the flows associated with the mechanical design and real geometry of the machine such as shroud leakage flows and bleeds. The precise value of the Reynolds number beyond which no loss reduction occurs depends on details of the design such as the blade surface roughness. They determine the stall behaviour of the blade row and are responsible for at least half the losses. It is also a useful factor in allowing the use of large-scale low- Fig. This is illustrated in Fig. This is a slight misnomer because it may give the impression that they are of secondary importance. reference [2]). This accounts for most engine compressors except for the very smallest at high altitudes. Increasing the Reynolds number of a compressor will generally reduce losses as the boundary layers thin until no further improvement is possible. This was illustrated on an isolated rotor [3] where even the presence of a small axial gap just upstream of a rotor influenced the flow field over a significant part of the span. At blade ends with a tip gap (rotors at the outer casing and cantilevered stators at the hub) the flow is dominated by the flow across the tip clearance. Even if there is no net flow into or out of the annulus the circumferential non-uniform static pressures produced by the blades mean that flow goes both into and out of these gaps in a non-uniform manner. At fixed blade ends the endwall flow cannot sustain the amount of diffusion required in the suction surface endwall corner and it tends to separate. three-dimensional and dominated by viscous effects. and are generally categorized as secondary flows (apart from the transition). Because they are travelling slower than the freestream flow they tend to be centrifuged radially outwards on rotors. 2. In fact it is these flows that tend to dominate the behaviour of multistage compressor blade rows. with the position of the start of the fully turbulent boundary layer moving along the blade chord with time. This makes it extremely difficult to model and predict accurately and so for compressor design to proceed several simplifications need to be made in order to produce practical design tools. Recently. This results in flow under. This separation region is fed by endwall fluid that is overturned by the mechanism just described and results in reduced turning in this region. particularly when the aspect ratios approach unity in the latter stages of a machine. These flows enter and leave the main annulus through axial gaps between rotating and stationary parts of the hub and bleed holes in the outer casing.or overturning and radially non-uniform exit flow angles from the blade row. This flow is driven by the viscous drag of the endwall passing over the blade tip and also by the pressure difference across the blade. 2 Schematic diagram of unsteady transition C02198 © IMechE 1999 speed compressors for detailed aerodynamic research.AXIAL FLOW COMPRESSOR DESIGN 439 freestream flow in the pressure gradient. Because of the effect of the wakes the boundary layer flows in compressors tend to behave more like turbulent ones rather than laminar ones and this means that blades with Reynolds numbers above about 2×105 will all have essentially the same type of boundary layer characteristics and not be subject to laminar boundary layer separations.g. The process of transition of the blade surface boundary layers is complex and has been the subject of extensive study recently (e. another category of flows has received attention and their importance in determining compressor behaviour has become more widely appreciated. On the blade surfaces the blade boundary layers do not behave in a two-dimensional way. if not fully understood. An example of the exit flow field measured on a low-speed compressor serves to illustrate this and is shown in Fig. This work has revealed that the process is unsteady and heavily influenced by the incoming wakes from the upstream blade row. The tip clearance flow interacts with the flow already on the endwall to produce a complex flow pattern that is often typified by a vortex type structure as the tip clearance flow rolls up and passes downstream. The flow patterns just described are well known. The data are taken from reference [4] and show contours of measured total pressure defect taken downstream of the third stator and circumferentially averaged flow angles and loss coefficients for the Proc Instn Mech Engrs Vol 213 Part C . These flow disturbances interact with the main flow field and can change the nature and magnitude of the secondary flows as well as producing losses. The result of the types of flow just described is that the flow field in a compressor blade row is extremely complicated and can be categorized as being unsteady. In stator blade rows the radial pressure gradient imposed by radial equilibrium tends to force the blade boundary layers radially inwards. 3. The process of transition varies with time as the wakes pass over the blade.

has not been touched upon. This breakdown occurs near rotor tips in discrete patches which rotate in the same direction as the rotor but at between 30 and 70 per cent of the rotor speed. The effect of the tip clearance is again seen clearly in the underturning of the flow measured downstream of stator 3. This situation will be fairly unusual even in a well-designed high-speed multistage compressor operating over its full speed range. perhaps only occurProc Instn Mech Engrs Vol 213 Part C ring at one speed. 4 Schematic diagram of a part span rotating stall C02198 © IMechE 1999 . There is a significant increase in the losses in this region. For ‘modal oscillations’ to occur the compressor has to be axially well matched. Near the outer casing there is both under. stall or surge. associated with the tip clearance flow.and overturning of the flow as well as an increase in the losses. The stators are cantilevered and the tip flow at the hub of the stator is clearly shown in the contours. It is now generally accepted that stall occurs in compressors when the flow breaks down under increasing adverse pressure gradients. commonly caused by high incidence at a rotor tip [5]. and the other identified with a longer lengthscale known as a ‘modal oscillation’.440 S J GALLIMORE Fig. 3 Example of the complexity of flow in a compressor blade row. This is called part span rotating stall and can have more than one stall cell initially but as the stall develops they tend to coalesce into one cell. leading to a bulk instability of Fig. The ‘modal oscillations’ are related to an instability of the whole compression system while the ‘spike’ type of instability can be related to a more local disturbance in the flow. Two types of breakdown have been identified. Towards the outer casing there is evidence of separation in the suction surface endwall corner. one associated with a short lengthscale disturbance known as a ‘spike’. illustrated in Fig. [4]) same blade. Eventually the instability in one blade row can disrupt the flow in the whole compressor. 4. So far in this description one of the most important phenomena in compressors. and consequently the most significant mechanism for practical purposes is the ‘spike’ type. (After Howard et al. with none of the stages being significantly more stalled than the rest.

The procedure is to iterate through the mean line prediction program. These must be achieved at shaft speeds appropriate for the compressor and turbine.. The reaction is the ratio of the static pressure rise achieved in the rotor to the overall stage static pressure rise. in a multistage compressor it is possible for one or perhaps more blade rows to be stalled without the compressor surging. Blade turning is either specified or is taken from correlations that relate blade deviation to the blade two-dimensional profile. etc. then no amount of later changes to blade profiles or numbers will be able to rectify the situation. technology level. However. High V /U a will give relatively higher DH/U2 and hence the stage pressure ratio for a given blade loading. which is in effect just solving the mean velocity triangles through the compressor. number of stages and overall length. but by just specifying the turning of the air imposed by the blades. The anticipated level of tip clearance in the compressor is an important parameter because of its large influence on both the efficiency and surge margin [6 ]. but at this point some parameters are used to guide the decisions without recourse to a full knowledge of the flow field. such as choosing too few stages or too short a compressor for example. the general velocity level in the compressor will be higher which may present problems in terms of high Mach numbers and so possibly losses in the front stages and high exit Mach numbers from the compressor which could cause problems with the diffuser and combustor design and so increase the losses there. compared with a military or industrial application. For 50 per cent reaction the relative velocity triangles through both rotor and stator are identical and they have equal diffusion. exit Mach numbers above 0. The velocity triangles provide knowledge of the stage loading. as quantified by the incompressible static pressure rise coefficient (Dp/D) which is discussed below. The details of the correlations used will vary because they will be heavily dependent on the past experience of the organization carrying out the design but fundamentally they all do the same thing. The basic inputs to the design will be requirements for a certain flow capacity. For the compressor designer it is important to be able to predict the onset of stall because generally it is not satisfactory to operate a compressor in stall for any significant part of engine operation. Some decisions can be made on the flow angles which will determine the reaction of the compressor. and axial velocity. Previous experience will play a major role in assessing what is an acceptable design in the circumstances. C02198 © IMechE 1999 inviscid and one-dimensional.4 can give high losses in certain circumstances. blade numbers and flow angles have been arrived at that are deemed satisfactory. blade numbers. pressure ratio. The mechanical restrictions of compressor length and radius to fit in with the whole engine arrangement need to be met.AXIAL FLOW COMPRESSOR DESIGN 441 the flow through the whole compressor which is called surge. However. until an annulus.. it may be judged from the preceding description of the phenomenon that predicting stall onset is a difficult task and remains one of the biggest challenges to research into compressor aerodynamics. such as the annulus lines. giving a smaller compressor. the endwall regions. timescales and cost. At first sight this might seem to be the obvious Proc Instn Mech Engrs Vol 213 Part C . 3 PRELIMINARY DESIGN This is the phase of a design where the basic outline of the compressor is established. Viscous effects are included by using losses taken from correlations to represent the annulus loss and blade losses. These assume that the flow through the compressor is steady. This typically occurs at low speed with the front of the compressor stalled but the rest of the machine being healthy enough to support it. Because of this. efficiency and surge margin for a range of engine operating points. etc. Depending on the nature of the project the designer will have more or less freedom to change these parameters. that may exist. DH/U2.25 are to be avoided and rotor 1 tip Mach numbers above 1. Annulus wall boundary layer blockage is also calculated from correlations or perhaps by simple methods such as that of Stratford [7]. A civil engineering project will require a different balance of risk.. However. Typically. Changing the fundamental architecture of the compressor later on in the project will have serious consequences. The regions of the flow that tend to dominate the behaviour of a compressor. are not modelled well in this process and so empirical adjustments have to be made to any theoretical limits to blade loading. Some decisions can be made without postulating blade profile inlet and exit angles. For a new engine there will be many iterations before the compressor arrangement has been fixed to fit in with whole engine requirements of flow. At this point in the process the prime considerations are blade and endwall loadings. The design decisions will have to be taken against the project background of acceptable risk. number of stages. These are fundamentally linked to the three-dimensional viscous nature of the real flow field. With an initial estimate of efficiency this will give the pressure ratio. once the a flow angles have been decided. However. The basic tools to deal with this part of the design are mean line performance prediction programs. efficiency and surge margin at the required shaft speed. etc. Examples of typical correlations may be found in the work of Miller and Wasdell [8] and Wright and Miller [9]. it is this part of the process that is the most crucial of all because if fundamental mistakes are made here. V /U. relatively simple and fast tools tend to be used to screen out the various possibilities before the most promising options are subjected to a more detailed analysis. It will also reduce the annulus area for a given mass flow.

changes later on in the design process may have implications for the mechanical design of the engine. The detailed choice of design levels for these types of parameter depends on the application. The aspect ratio also plays an important role in determining the surge margin of the compressor. Typical design values for a compressor would be about 0. These are primarily dependent on the matching of the compressor and the surge margin requirements.442 S J GALLIMORE choice because it evens out the loading across the blade rows. but this will vary depending on the type of operation that the particular compressor is required to perform. However.45. causing losses and increased deviation. which indicates that the choice of reaction for its own sake is not a crucial design parameter in determining efficiency and surge margin and is more likely to be the result of the other design considerations discussed above. if the rotor 1 Mach numbers are high then some inlet flow swirl angle in the direction of rotation would reduce them. The diffusion factor is a measure of the diffusion on the suction side of a blade. However. stagger and cambers as input to correlations to predict losses and deviations. Mach numbers. Preliminary rules for the axial gaps between the blades will need to be established to allow sufficient room for surge deflections and the decay of the blade potential fields to reduce the vibration forcing on the blades. However. However.45 are typical. A value of unity would indicate that all the dynamic head had been turned into static pressure and the flow had stagnated. reference [11]). but it is generally thought that low aspect ratio blades control the endwall boundary layer flow better and so delay the onset of stall. near the endwalls the loadings will always be higher than the mean levels because of the annulus wall boundary layers. Mach number levels are also chosen by the designer. There is no agreed scientific explanation for this. can be found in reference [1]. C02198 © IMechE 1999 . However. There are successful compressors with 50 per cent reaction.g. If the inlet flow angle is axial and there is no requirement for variable stators on the compressor for part speed surge margin then perhaps axial flow into the first rotor is the best option. design values of around 0. These use the blade numbers. this is usually taken into account in the acceptable mean levels of loading used as design criteria which will be somewhat below the equivalent values that are appropriate for a purely two-dimensional cascade. Common examples of diffusion parameters are the Leiblein diffusion factor and the incompressible static pressure rise coefficient. The trend has been for compressor aspect ratios to become lower over the years. If this becomes too large the blade boundary layer will separate. A more complete discussion on the choice of reaction. Lower aspect ratios tend to give more surge margin at a given two-dimensional blade loading. This happens because of the possibility of recovering some of the velocity defect in the blade wake as it passes through the downstream blade row and so reducing the wake mixing losses. The choice of reaction will also influence the bearing loads which need to be monitored. Dp/D is fundamental to the velocity triangles and is simply a measure of the static pressure rise through a blade row non-dimensionalized by the inlet relative dynamic head. high reaction (zero stator exit flow angle) and axially varying reaction. Care must be taken at this point to acknowledge inlet radial distributions of total pressure. etc. This does not happen because the annulus and blade boundary layers have less stagnation pressure than the freestream and will separate before the static pressure rise is achieved. Again. The fundamental parameters that are used in compressor design are measures of flow diffusion and velocity. There is no sure way of choosing the aspect ratio. the reaction is not necessarily a completely free choice. The balance between profile and secondary losses might influence the choice of aspect ratio. previous experience and the amount of surge margin required. Some decisions will also be made on blade incidence settings at this point. At exit from the compressor there is generally a requirement for axial flow into combustion chambers. The correlations for loss and deviation will generally be based on the loading coefficients described above and the blade tip clearance. There is some evidence to suggest that there is an optimum (smaller) axial gap between blade rows that gives the best efficiency (e. The inlet flow angle to the compressor will be determined by what is upstream. loadings. Dp/D (or the De Haller number if preferred). The overall length of the compressor may also influence the choice of aspect ratio. While this is not a fundamental constraint. the demonstrated efficiency gains are small and it is generally the mechanical constraints that determine the minimum axial gap. The predicted losses and efficiencies give the designer another guide to the choice of blade numbers. High Mach numbers will give high dynamic heads and reduce the loading coefficients described above. Once preliminary blades have been set up at the design point the mean line programs can be used to predict a set of overall characteristics for the compressor. but one measure that can be used is that derived by Koch [10] which is based on a diffuser analogy of the compressor blade passage. as will the constraints on overall compressor length. which reaches similar conclusions to those here. The limits used in a particular design will be based on previous experience. If there is a significant radial variation then the loading of some parts of the blade Proc Instn Mech Engrs Vol 213 Part C will be significantly above the mean levels and so a reduced mean level of design loading would be appropriate in these cases. This creates separations which in turn produce loss and increased deviation. Of course. but the swirl will have to be removed by the point at which the flow leaves the compressor and this increases the stator loadings. excessive Mach numbers give shock losses that can result in low efficiencies if there is insufficient work or turning in the blade row.

This will also give the greatest surge or stall margin. handling bleeds or casing treatments are required to enable the compressor to operate satisfactorily over the speed range. At this point in the design process this flow phenomenon is not included in the models used to predict compressor performance and consequently the onset of stall is related. In particular. However. It is possible to establish stability criteria based on the performance of the individual stages such as those discussed in references [9] and [12]. the practical approach is to try to have the smallest clearance that can be mechanically designed. C02198 © IMechE 1999 However. they can be used with some confidence to predict relative changes between machines. front stages are forced towards stall while the rear stages are driven towards negative incidence. During the preliminary design phase the decision will have to be made whether to have cantilevered or shrouded stator blade rows. because of mechanical and vibration considerations. To the author’s knowledge there is no clear evidence to suggest that one solution is better than the other in all cases. Shrouds have the advantage of removing the tip flow and resultant losses associated with cantilevered designs. These rely on the accurate prediction of stage characteristics and even then do not guarantee success.AXIAL FLOW COMPRESSOR DESIGN 443 Correlations are also used to calculate annulus and secondary flow losses and blockage.g. a small value. They are too complicated for computation and prediction and so again recourse is made to correlations and empirical rules. This needs to be assessed at part speed as well as at design to establish whether variable stators. It must be remembered that in a multistage compressor it is possible for an individual stage to be stalled while the machine as a whole remains stable. whereas it only needs one such crack to release part of a cantilevered stator. some leakage flow through the shroud will still be present and this will interact with the main annulus flow. However. The initiation of stall is a complex. Generally the preliminary design programs available cannot be expected to give the correct absolute values of efficiency and surge margin. On the other hand the cantilevered solution allows thinner stator blade sections near the hub. There is some evidence that an optimum tip clearance exists that gives maximum efficiency [13. The simplest of these again does not include the major. The methods used to determine the surge margin at this stage of the design process are heavily empirical. for the shroud design to run at a very small clearance because the seal fins can be allowed to rub more readily than the free end of a stator. reference [10]). The level of the optimum clearance is about 1 per cent of blade chord. It is generally acknowledged that stall initiates in the endwall regions of compressors. The blockage is equivalent to the displacement thickness of a boundary layer but in reality has a wider role than this because it is part of the semi-empirical input to the calculations that may be adjusted through experience to give better agreement between the calculations and experimental data. This means introducing the radial or spanwise dimension into the design using a throughflow procedure. It is possible. and this can be advantageous in high Mach number applications. the correlations used for blade and endwall losses are generally based on traditional blade profile types such as DCA or NACA without any three-dimensional design features such as endbends. however. changing the secondary flows around the hub and increasing the loss. This can lead to improved performance retention over the lifetime of the compressor as the clearances increase. The blade incidence settings at design may need adjusting at this point to compensate for the off-design operation. a shrouded stator will need to crack through completely at two radial locations before any blade is lost. through correlations and experience. The fact that there are both shrouded and cantilevered compressor designs in service perhaps indicates that the choice is not a clear one. three-dimensional phenomenon which is still not fully understood. it is at this stage that the radial matching of the compressor is established and more attention can be given to the endwall regions. Consequently. 4 THROUGHFLOW DESIGN Once the preliminary design phase has screened out some of the proposals it is necessary to examine the remaining options in more detail. particularly at rotor tips where the tip clearance plays a dominant role. Other aspects of the flow field that can be used to correlate with stall are blade loadings and incidence as well as limiting diffusion (e. Blockage levels can be calculated using integral Proc Instn Mech Engrs Vol 213 Part C . In the simplest throughflow methods the effect of the endwall boundary layers is included by specifying a blockage and extra loss in the calculations. which for most practical applications is difficult to achieve mechanically. Perhaps the most accurate method of assessing the surge or stall line is to calibrate the methods against the known performance of similar machines using a range of criteria such as those mentioned above. significant flow features for similar reasons to the preliminary design methods. The tolerance of the blades to damage from debris passing through the compressor in service also needs to be considered. as well as other flow properties. unsteady. For the compressor designer the choice is often made to stay with the style of design that has worked adequately in similar compressors in the past. At this point the estimation of surge margin becomes important. This is not fully understood but is likely to be associated with the balance of losses associated with the tip clearance flow mixing out into the blade passage and the losses associated with the secondary flow in the corner of the blade suction surface and the endwall. 14]. At part speed. to the bulk loading and flow properties available.

with some stages not running at their design conditions at the overall compressor design operating point. typical of those used in the past. which shows a comparison of the axial velocity profiles and relative flow angles measured downstream of a rotor in a low-speed compressor with the results from several calculations of the flow field. etc. such as the diffusion factor. If the level of blockage specified is not correct then the compressor will be mismatched from front to back. The throughflow will give more details of the endwall conditions. 5. It also radially redistributes the high endwall temperatures caused by the endwall losses. Indeed. Using a Fig. The line labelled as design in both plots is that produced by a conventional throughflow calculation.g. references [19] and [20]). is that including some sort of spanwise mixing in the throughflow model is important if accurate predictions of radial total pressure and temperature distributions are to be achieved. with one mechanism being a turbulent diffusion type of process caused by the unsteady flow and wakes coming from upstream blade rows and another being the radial components of deterministic secondary flows in the blade rows. Both mechanisms have a role. An example of the improvements that can be gained by taking into account these real effects is shown in Fig. which tend to offload the blade rows there. etc. are similar to those used in the preliminary design phase. axial velocity profiles. The data from the throughflow calculations is split into sections along a streamline for each blade.444 S J GALLIMORE boundary layer methods. 18]. The throughflow effects are fed into the blade-to-blade design via the height of the stream tube associated with each streamline. The causes of spanwise mixing have been the subject of considerable research (e. A relatively common design feature is to change these radial distributions to promote increased throughflow velocities in the hub region. tip clearance flow and spanwise mixing gives the much more realistic radial profiles of the viscous throughflow calculation. What is clear. except that radial distributions of these quantities are now available. this calculation gives results of similar accuracy to those from the Denton three-dimensional multistage CFD method [21]. In the past this was achieved by using correlations to define the blades from a series of standard profiles. but it is now more usual to use bladeto-blade calculation programs to define the blade shapes. This will lead to predictions of endwall loadings. (From Gallimore [16 ]) Proc Instn Mech Engrs Vol 213 Part C C02198 © IMechE 1999 . however. 16 ] and more realistic spanwise variations in flow properties can be obtained by using a throughflow including the effects of spanwise mixing [17. These individual sections are then designed before being put together to define the whole blade. 5 Measured and predicted rotor exit axial velocity and flow angle. one of the simplest and most accurate methods being that of Stratford [7]. the endwall loadings can be controlled to give acceptable levels. The addition of the real effects of endwall shear.. Again. at this point the parameters used to judge the loadings. Spanwise mixing is particularly important in multistage compressors because it prevents the continuous growth of the endwall boundary layers through the machine by mixing the endwall losses towards mid-span. The need for the empirical blockage can be removed by modelling the endwall boundary layers in a more direct manner [15. with the relative magnitudes depending on the particular compressor and the operating point. This can lead to reduced efficiency and surge margin. due to the effects of radial equilibrium and streamline curvature effects. 5 BLADING DESIGN ( TWO-DIMENSIONAL) Once the throughflow design has been completed. but now taking into account the endwall effects in more detail. the blading that is intended to achieve that air angle design can now be defined. etc. Similar correlations to those in the preliminary design process can now be used to refine the performance prediction of the compressor. taken from reference [16 ]. Dp/D. By changing the radial variation in stator exit angles and indeed the radial distribution of the stage pressure ratio. Typical values for blockage can reach over 10 per cent of the annulus area in the rear stages of a multistage compressor.

deviation and the boundary layer parameters. reference [22]).AXIAL FLOW COMPRESSOR DESIGN 445 blade-to-blade program is essentially a two-dimensional approach to blade design and suffers from similar problems to those described earlier near the endwall regions where the flow field is certainly not two-dimensional. For the integral methods the basic flow solver is inviscid and the effects of the blade surface boundary layers are incorporated through the displacement thickness of the boundary layer which is imposed on the blade shape either by moving the blade surface or by using a transpiration model. In this case a double circular arc (DCA) blade has been modified to have a parabolic arc camber line. the predicted blade losses and air exit angles can be fed back into the throughflow calculation to give updated aerodynamics for a further refinement of the blade profiles [23]. The essential parameters that need to be calculated in order to complete the two-dimensional design of the blading are the section incidence. The calculated boundary layer displacement thickness and shape factor have been considerably reduced at the trailing edge for the parabolic blade. In these cases the designer specifies the blade surface pressure distribution or desired boundary layer parameters along the blade chord and the methods calculate the blade profile that will achieve the desired result. A prime use of these calculations has been to allow the use of non-standard blade sections where the predicted boundary layer parameters can be controlled to give reduced losses by allowing the designer freedom to change the blade camber and thickness distributions. indicating a more firmly attached boundary layer with reduced profile loss. It was the intention of these designs to minimize the blade profile loss by controlling the blade boundary layer growth along the blade chord. Typically. An example of the improvements that can be made with these methods is given by Ginder [23] and reproduced in Fig. This gives a quasi-three-dimensional design system for the compressor with an iterative link established between the throughflow and blade-to-blade calculation. Further refinements can be made to the basic methodology just described. As time has passed these initial attempts at designing the blade profiles in detail have developed into proprietary rules and approaches with the same philosophy of minimizing blade losses but with adequate off-design operating range. Towards the rear of multistage compressors.g. particularly in the endwall regions). the whole operation can proceed with only the minimum amount of input from the designer. it is reasonable to say that the majority of the blade is essentially two-dimensional with the threedimensional endwall effects being restricted to close to the annulus walls. with a series of specified design rules. For instance. In the front stages. such as steady point or length transition. The designer does not have complete freedom in the Proc Instn Mech Engrs Vol 213 Part C . For reasonably loaded compressors it can be anticipated that reblading from conventional standard blade profiles to modern ones will give some 1 per cent improvement in efficiency without any change in the operating range of the machine. the first rotor could be designed to have about 3 per cent of choke margin. The blade section deviation is now generally accepted to be that calculated by the blade-to-blade prediction programs (of course the real deviations may well be different from the section deviations discussed in this part of the paper because of secondary flow effects. so it was essential to be able to calculate these accurately. Additional modifications to the blades may be made in these endwall regions using the threedimensional techniques described in the next section. where the aspect ratios can be of an order of unity or even less. This type of design also had significantly lower deviations than the standard sections. Clearly it is essential to calculate the boundary layer in some way in order to calculate the change in loss as the blade profile is designed. However. The boundary layer parameters are calculated using the integral boundary layer equations and the blade surface pressure distribution calculated by the Euler solver. it is doubtful that any of the blade aerodynamics is really two-dimensional. where the annulus wall boundary layers are relatively thin and the aspect ratios are higher. requiring a blade-to-blade calculation with the boundary layer modelled. It is recognized that the blade surface boundary layers are not twodimensional in reality. C02198 © IMechE 1999 although more complex unsteady criteria based on the real transition phenomena described earlier are now becoming available. It is also common for relatively simple boundary layer transition criteria to be used in these calculations. Choke margin is another parameter that is influenced by incidence setting. This is important for the front stages of machines because if the blade throats are designed too small the high-speed flow capacity and efficiency of the compressor will be reduced. expressed as a percentage of the choke flow. 6. so freeing time for other tasks. The design of the blade profiles themselves can be automated so that. Generally speaking the incidence setting will be decided upon by experience of how the compressor needs to be matched in order to meet a range of operating points. with significant radial flows being present owing to centrifugal effects on rotors for example. These programs calculate the flow through the blade passage and include the effect of the blade surface boundary layers either by integral methods or through full two-dimensional Navier– Stokes calculations. Generally these methods are used in the design by analysis mode. An early example of this new type of profile was the ‘supercritical’ airfoil designs borrowed from aeroplane wing development (e. although some methods are available that allow true inverse design to be used. This parameter is the difference between the flow passing through the blade section and the theoretical choke flow for that section. it has been found that the predictions given are sufficiently accurate to give good compressor performance.

(After Ginder [23]) blade profile design. and hence the total amount of boundary layer growth and loss for the blade profile. endbend C02198 © IMechE 1999 .g. Previous attempts at this had relied upon the experimental approach being based on observations from low-speed compressor rigs (e. Other methods such as that of Denton [21] rely on the more simple mixing plane approach. but recently the use of multistage calculations has become a practical proposition in the design process [24]. Single blade row calculations have been available for some time. particularly on the suction surface. 6 Example of improved blade section performance using a blade-to-blade prediction program. If there are discontinuities in the blade surface they can alter the transition behaviour of the boundary layer. In both cases there is the issue of including the effect of the circumferential non-uniform flow field of the upstream and downstream blade rows on the blade row under consideration. This may be achieved by using the approach of Adamczyk [25] whose analysis results in the inclusion of these effects as deterministic stresses (analogous to the Reynolds stresses derived in turbulence modelling) in the flow equations.g. 6 BLADING DESIGN ( THREE-DIMENSIONAL) With ever-increasing computing power it is possible to use fully three-dimensional Navier–Stokes CFD methods in multistage axial flow compressor design. The use of three-dimensional CFD allows two major advances in the design process. 7. The approach adopted by LeJambre et al. The relative importance of these effects is still the subject of some debate in this new topic and it is perhaps still too early to state which approach gives the best balance between accuracy and computational cost. which shows a computational mesh for the six-stage high-pressure compressor used in the Rolls-Royce Trent engine. [24] was to model the upstream and downstream blade rows using an overlapping grid technique.446 S J GALLIMORE Fig. This latter is still the subject of conProc Instn Mech Engrs Vol 213 Part C siderable debate. Firstly. Among the issues that need to be addressed are the use of an appropriate turbulence model and the modelling of the mixing plane. The leading edge thickness and shape can have an important effect on the profile losses and off-design performance and this should be controlled as accurately as the manufacturing process allows. illustrated in Fig. There are many details associated with the multistage CFD methods that are outside the scope of this paper. The blades must be mechanically sound and this leads to restrictions on minimum blade thicknesses as well as blade leading and trailing edge thickness. reference [26 ]) or on semi-empirical methods based on annulus wall boundary layer theory (e. it is possible to design the blades in the three-dimensional flow field and therefore reduce the secondary losses.

a recent re-analysis of an isolated rotor has shown that only a small axial gap with zero net flow just upstream of the hub leading edge can alter the pressure rise across the rotor over the bottom half of the blade span [28]. Bowed stators have been used by Pratt and Whitney [24]. fillet radii. For example.g. The other main advantage of these CFD methods is that they allow the effect of real geometries on the flow field now to be included in the design. Examples of these are the effects of tip clearance. The authors noticed that the effective incidence on the endwall sections of a blade was reduced in a three-dimensional calculation of the flow through a blade row compared with that which would have been deduced from a purely two-dimensional analysis along each streamline separately. because its accuracy still leaves something to be desired. In these techniques the blades are modified in the endwall regions by recognizing the effects of the endwall boundary layers. it is now recognized that the three-dimensional nature of the flow field is important in achieving successful three-dimensional designs and it is not sufficient to consider sections of the blades at different radial heights independently from each other. The use of CFD allows the design of three-dimensional blading that reduces endwall losses. This has the effect of reducing the loading on the endwall Proc Instn Mech Engrs Vol 213 Part C . references [3] and [28]). it is likely that some of the discrepancies between prediction and experiment are not due fundamentally to the CFD modelling but rather to the lack of realism in the mechanical arrangement used in the CFD model (e. This fact was first recognized by Wadia and Beacher [27] who introduced the concept of radial relief. it is now reliable enough to predict the direction of any changes correctly while not predicting the magnitude correctly. 7 Example of a computational mesh for a six-stage compressor designs). It has recently been realized that these can have a significant influence on the flow field in a compressor and hence its performance. These approaches suffered by not having the physical modelling included in the design process. shroud leakages and any other leaks and axial gaps that may be a consequence of the mechanical design of the compressor. for example by increasing the blade inlet angles in these regions where the reduced axial velocities cause the air inlet angles to be greater than they are towards mid-height and by adjusting the blade exit angles to take into account the inevitable overand underturning of the endwall flow as it passes through the blade row. Indeed. It is clear that including these real effects will be essential in future designs as well as focusing attention on the detrimental effects of such features on the flow field.AXIAL FLOW COMPRESSOR DESIGN 447 Fig. For these designs the endwall sections of the stator blades were shifted tangentially relative to the mid-height sections. This still means that the designer needs to exercise the C02198 © IMechE 1999 same type of skill and judgement as has been required throughout the rest of the design process. so it was quite possible to design one compressor successfully and the next one unsuccessfully. For example. While it is still not possible to rely on the CFD exclusively. bleeds.

have not been discussed. but this will be countered by the increased sophistication required to produce a competitive product. with the proviso that the product must retain a competitive performance.448 S J GALLIMORE sections and reducing the separations there. thus reducing losses. This may lead to further loss reductions in blade designs by optimization of the blade and endwall boundary layer behaviour. predicted by multistage CFD Proc Instn Mech Engrs Vol 213 Part C C02198 © IMechE 1999 . 7 CONCLUSIONS AND FUTURE TRENDS This paper has been. A better understanding of stall will allow higher blade loadings to be used. Further improvements in computational power will allow more automation and optimization in the design Fig. As computational power grows the certainty of compressor design increases. although efficiency predictions accurate to within a couple of percentage points are still some way away. To obtain a successful design these two opposing changes have to be balanced to give a net improvement and the only methods that allow this judgement to be made on a scientific basis are three-dimensional ones. This type of work is the latest advance in compressor design and there is little doubt that fully threedimensional designs using multistage CFD are now the standard. that many of the areas that need to be considered during the design process have been highlighted and that the interested reader will be able to explore specific topics in more depth. of course. a brief survey of axial flow compressor design. Crucial will be the development of a method that accurately predicts the stalling of a compressor. where the results of multistage CFD are presented for a conventional stator blade and a three-dimensional redesigned one. particularly at the casing. it must be questionable as to how much effort is worth expending on improving the efficiency of a compressor that has a polytropic efficiency of 92 per cent. An example of such a design is shown in Fig. This effect is a three-dimensional one but also has some detrimental effects. advances in CFD have produced efficiency gains of at least 2 per cent in the last decade owing to improved blade profile and three-dimensional design. 8. with the rest of the design process described earlier becoming just part of the preliminary design system and used before the three-dimensional modifications to the designs are incorporated. 8 Axial velocity contours (m/s) close to the stator suction surface. with a consequent reduction in parts count and hence cost as well as perhaps greater efficiencies. This reflects the current drivers in the commercial world of engine manufacture of cost and reliability but. threedimensional calculations. While such calculations are available now and are being used. however. None of the issues have been explored very deeply and even now significant areas. it will be some time before sufficient experience and understanding have been gained for them to be regarded as mainstream design tools. such as casing or tip treatments. It will require time-accurate. The predicted axial velocities just off the stator suction surface show that the amount of separation on the redesigned blade has been considerably reduced. Because the CFD incorporates more of the real physics of the flow it should also give better predictions of the compressor performance. unsteady. so it is likely that future improvements in compressor design practice will be used to reduce cost and lead times and improve reliability. by necessity. However. For example. It is hoped. one of which occurs towards mid-height where the loadings are increased somewhat. Future developments would include improved understanding of the unsteady nature of the flow field and its impact on boundary layer transition and secondary flows. This is still some way off because of the complex flow phenomena involved. leading to reduced losses.

J. J. Trans. and Gallimore. 393–401. 354.. C. C. R. S. H. G. Turbomachinery. October 1965. and Strazisar. 1993. ASME. 18 Gallimore. 1992. and Cumpsty. J. E. R. R. 296. 354. D. J. 1985. J. B. Biederman. B. paper C423/028. J. C. Von Karman Institute for Fluid Dynamics. J. 1991. K. to be published in J. The calculation of three dimensional viscous flow through multistage turbomachines. Turbomachinery. 1981. Turbomachinery. Zacharias.. S. Engng for Power. 120. A. Wake recovery performance benefit in a highspeed axial compressor. Endwall effects at two tip clearances in a multi-stage axial flow compressor with controlled diffusion blading. 109. J. 1987. Trans. 32. J. Adamczyk. R. Trans. A. 1985. 1997. Turbomachinery. ASME. Spanwise mixing in multi-stage axial flow compressors: Part I—Experimental investigation. P. 3 Shabbir. J. Trans. turbulent diffusion and mixing in axial-flow compressors. D. Trans. B.. 26 Wisler. M. Loss reduction in axial flow compressors through low speed model testing. 1990. the views expressed herein are those of the author and are not necessarily those of Rolls-Royce plc. Off-design prediction of compressor blade losses. ASME paper 95-GT-343. Trans. J. R. N.. The development of a second generation of controlled diffusion airfoils for multistage compressors. A. J. IMechE Conference Proceedings CP 1991-3. D. However. E. Effect of tip clearance flow on compressor stability and engine performance.. and Miller. 115. Trans. P. Bauer. and Day. H. J. P. ASME. Lecture Series 1985-05. 10 Koch. ASME paper 91-GT-374. N. A. F. IMechE Conference Proceedings CP 1987-6. ASME paper 85-GT-226. D.. Okiishi. H. 1991. 4 Howard. J. Trans. H. ASME.. Engng for Gas Turbines and Power. ASME. 13 Wennestrom. 112. P.AXIAL FLOW COMPRESSOR DESIGN 449 process with a consequent reduction in costs and lead times as well as better designs. A study of spike and modal stall phenomena in a low-speed axial compressor. 116. T. A. ASME paper 95-GT-461 to 464. Compressor Aerodynamics.. Ph. ASME paper 97-GT-346. 455. C. Effect of a mismatch between the buttons of variable stator vanes and the flowpath in a highly loaded transonic compressor stage. Trans. 16 Gallimore. C. D. 635.. Hodson.. and Yekta. Turbomachinery. 7 Stratford. Barton. 5 Camp. Y. Adamczyk. M. J. P. The use of boundary layer techniques to calculate the blockage from the annulus wall boundary layer in a compressor. J. J. H. T. F. 1967. 109. S.. Secondary flow. paper C279/87. and Beacher. ASME. Non-axisymmetric flows in axial compressors. and Young. and Okiishi. Wisler.. Spanwise mixing in axial flow turbomachines. 104. P. F. P. 10. 552. A. 2. 22 Behlke. G. 6 Freeman. 20 Wisler. 1997. and Smith Jr. 114. L. 662.. ASME. 1998. G. I. S. H. Bois.. 120. C. 1987. 1986. Engng for Power. 28 Escuret. 24 LeJambre. Gallimore. J. 411. REFERENCES 1 Cumpsty. J. 15 Howard. C. J. J. Design and performance of advanced blading for a high-speed HP compressor. Veysseyre. 2 Halstead. S. 587. 1984. Turbomachinery. 108. 98. A. 1997. J. Walker. ASME. Engng for Gas Turbines and Power. 11 Van Zante. Trans. Three-dimensional relief in turbomachinery blading. and Shin. F. ASME paper 97-GT-471. J. 12 Dunham. A. H. J. 17 Adkins Jr. C. 108. D. 1986. The effect of hub leakage flow on two high speed axial flow compressor rotors. Trans. J. An improved compressor performance prediction model. C02198 © IMechE 1999 Proc Instn Mech Engrs Vol 213 Part C . Boundary layer development in axial compressors and turbines: Parts 1 to 4. R. 1995.. C. ASME paper 67-WA/GT-7. Trans. Turbomachinery. 14 Dong. G. Celestina. J. M. Spanwise mixing in multi-stage axial flow compressors: Part II—Throughflow calculations including mixing. Monograph 3. D. Viscous throughflow modelling of axial compressor bladerows using a tangential blade force hypothesis. S. 25 Adamczyk. R. Viscous throughflow modelling for multi-stage compressor design. J. Mechanical Engineering Science. ACKNOWLEDGEMENT The author would like to thank Rolls-Royce plc for permission to publish this paper. J. A. 9 Wright. Villain. T. Model equation for simulating flows in multi-stage turbomachinery. C. 108. J. 19 Gallimore. J. A. ASME. 1994. Turbomachinery. Turbomachinery. J. 97. ASME. D. Trans. Gleixner. ASME. ASME. Ivey. I. and Hodson. 8 Miller. Turbomachinery. Turbomachinery. 1995. and Wasdell. J. Stalling pressure rise capability of axial flow compressor stages. D. T. Strazisar. Turbomachinery. J. J. 1982. C. Savarese. L. J. S. Experimental study of a high throughflow transonic axial compressor stage. Three-dimensional flows and loss reduction in axial compressors.. M. B. 27 Wadia. 1985. ASME. 107. ASME. 23 Ginder.. L. H. J. Trans. Development and application of a multistage Navier–Stokes solver: Part II— Application to a high pressure compressor design. 1998. ASME.. 106. 1986. and Okiishi. and Naviere. 1989 (Longman). C. Trans.. 1987. 21 Denton. ASME paper 97-GT-535. M.-W.

Sign up to vote on this title
UsefulNot useful