Jtam.dvi JOURNAL OF THEORETICAL AND APPLIED MECHANICS 47, 2, pp. 321-342, Warsaw 2009 INVESTIGATION OF ENDWALL FLOWS AND LOSSES IN AXIAL TURBINES. PART I. FORMATION OF ENDWALL FLOWS AND LOSSES Piotr Lampart The Szewalski Institute of Fluid-Flow Machinery, Polisch Academy of Science, Gdańsk, Poland e-mail:lampart@imp.gda.pl Endwall flows are among the most important sources of loss in turbi- nes. The process of formation of endwall flows and evolution of vorticity from the endwall boundary layers is briefly described in the paper. The resulting endwall losses are discussed. The endwall boundary layer los- ses are evaluated theoretically by integrating the entropy increase in the boundary layer and assuming a certainmodel of blade profile load. The endwall losses are also calculated with the help of CFD in a 3D straight turbine cascade of Durham and compared with the results of available experimental data of ERCOFTAC. Key words: axial turbine, endwall flow, secondary flow, enthalpy loss 1. Introduction Endwall flows in turbineblade-to-bladepassages areflows in endwall boundary layers and secondary flows. The secondary flows originate from specifically developing endwall boundary layers and are associated with the presence of longitudinal vortices with a dominant streamwise component of the vorticity. They are driven by transverse static pressure gradients andmass forces acting on fluid elements in curvilinear motion through the blade-to-blade passage. The secondary flows also modify the shape of endwall boundary layers from which theyoriginate.Understanding thecomplexdevelopment of endwall flows is a part of understanding turbomachinery flows. The problem of endwall flows, especially secondary flows, is discussed in the literature. There are a number of secondary flowmodels documenting the progress in our understanding of the secondary flows over the years. Some of these models will briefly be presented in Section 2. First, the secondary 322 P. Lampart flows in cascades without a tip clearance and relative motion of the blade tips and endwall will be considered. Also, the formation of secondary flows in tip clearance cascades will be discussed. Endwall flows are also an important source of losses in turbines, espe- cially in cascades with short-height blading and high flow turning. Due to the complex nature of endwall boundary layer flows and secondary flows, the evaluation of endwall losses is not an easy task. New formulas and methods are needed to calculate endwall losses. It is also expected that CFDmethods can provide a powerful tool for the evaluation of endwall flows and losses in turbines. 2. Formation of endwall flows The picture of endwall flows in turbine blade-to-blade passages is extremely complex, dominated by the presence of secondary flows. A few landmarkmo- dels explaining the development of secondary flows are illustrated in Fig.1. Themain type of secondary flow is the induced recirculating flow, which leads to the formation of a passage vortex. The source of the induced recirculating flow is the cross flow in the endwall boundary layer that forms as a result of force equilibrium in curvilinear motion. The momentum equation in the cross-stream direction can be written in the form ρv2 R = ∂p ∂n (2.1) where v denotes the velocity, p is the pressure, ρ – density, R – streamline curvature radius, n –normal coordinate.With adecrease of the velocity in the boundary layer, a reduction of the streamline curvature radius in the boun- dary layer flow is required in order to balance the pitchwise pressure gradient formed in the channel. As a consequence, the boundary layer flow is turned more than themain flow in the blade-to-blade channel, leading to a crossflow from the pressure to suction surface in the endwall boundary layer. A com- pensating return flowmust then occur at a certain distance from the endwall, giving rise to the recirculating flow described by e.g. Hawthorne (1951), Pu- zyrewski (1963), which can be seen in Fig.1a. From this recirculating flow, a passage vortex is formed. Downstream in the blade-to-blade passage, due to the pressure-to-suction side pressure difference, the passage vortex locates near the blade suction surface. As a result of the recirculating flow in the Investigation of endwall flows... 323 Fig. 1. Secondary flowmodels in turbine cascades: (a) –model of Hawthorne (1955), (b) – model of Langston (1980), (c) – model of Sharma and Butler (1987), (d) – model of Goldstein and Spores (1988), (e) – model of Doerffer and Amecke (1994), (f) – model ofWang et al. (1997) 324 P. Lampart neighbouring blade-to-blade passages, a vortex layer is formed at the trailing edge, which is quickly rolled-up downstream into a shed trailing edge vortex. Another element of secondary flows is a horse-shoe vortex. The process of formation of the horse-shoe vortex upstream of the leading edge and its downstream transport was explained by Langston et al. (1980), Marchal and Sieverding (1977), Hodson and Dominy (1987), Eckerle and Langston (1987). Themodels of this process presented in these papers differ fromone another in details only. Themodel illustrated in Fig.2 comes from the paper byMarchal andSieverding (1977).Theboundary layer fluidupstreamof the leadingedge is deceleratedbytheadversepressuregradientandseparates ata saddlepoint s1. The boundary layer fluid elements forma reverse recirculating flow just before the leading edge. This reverse flow separates at another saddle point s2. The upstream boundary layer rolled-up in the recirculating zone flows past the leading edge and is transported downstream in two legs – pressure-side and suction-side leg of the horse-shoe vortex. The suction-side leg of the horse- shoe vortexmoves near the suction surface of the blade. The pressure-side leg subject to the pressure gradient towards the suction surface moves across the blade-to-blade passage towards this surface. The legs of the horse-shoe vortex move along the lift-off lines that are lines of the saddle points as illustrated in Fig.2. The location of the horse-shoe vortex lift-off lines, especially that of the Fig. 2. Separation of the endwall boundary layer upstream of the blade leading edge and formation of the horse-shoe vortex; s1, s2 – saddle points, Marchal and Sieverding (1977) pressure-side legdependson the load of the frontpart of theblade.For the case of front-loaded profileswith high flow turning in the front part of the blade-to- bladepassage, the lift-off line of the horse-shoe vortex pressure-side leg reaches earlier the vicinity of the suction surface than for the case of the aft-loaded profiles. All main forms of secondary flows meet at the suction surface of the blade. Themodel that explains the transport of the horse-shoe vortex, where Investigation of endwall flows... 325 the pressure-side leg of the horse-shoe vortex together with the endwall cross flow formthemain recirculatingflowand the resultingpassage vortex,whereas the suction-side leg of the horse-shoe vortex stays apart counter-rotating with respect to the passage vortex, comes from the work of Langston (1980) shown in Fig.1b. In themodel of Sharma andButler (1987) (Fig.1c), the suction-side leg of the horse-shoe vortex is wrapped around the passage vortex, whereas in the model of Goldstein and Spores (1988) (Fig.1d), the suction-side leg locates above the passage vortex andmoves togetherwith it.Thepicture looks similar in the model of Doerffer and Amecke (1994) (Fig.1e), with a dividing stream surface between the passage vortex and the suction-side leg of the horse-shoe vortex. In point of the suction-side leg of the horse-shoe vortex, Wang et al. (1997) (Fig.1f) return to the concept of SharmaandButler (1987) – this leg of the horse-shoe vortex remainswrappedaround the passage vortex. In addition to that, both legs of the horse-shoe vortex are formednot froma single vortical structure, but from a pair of alternatelly dissipating vortices. At the trailing edge, the secondary kinetic energy of the suction-side leg of the horse-shoe vortex can be entirely dissipated as a result of shear interac- tionwith the strongerpassagevortex,Moore andSmith (1984), alsoSieverding (1985). Thus, subregions of the sense of rotation opposite to that of the passa- ge vortex may not be observed at the trailing edge section. Most papers also suggest that there will be no distinctionwithin the passage vortex as towhich part of it is due to the main recirculating flow or due to the pressure-side leg of the horse-shoe vortex. The investigations of Doerffer and Amecke (1994) indicate that the pressure-side leg of the horse-shoe vortex can be located in the core of the passage vortex. Results of the above investigations (all papers cited so far in this paper are experimental works) are also confirmedby results of visualisation of RANS-based numerical calculations of the subsonic linear Durham cascade (Gregory-Smith, 1993-2002) made by Doerffer et al. (2001). The visualisation was performed bymeans of streaklines originating from the regions of formation of various types of secondary flows and passing through the loss centres at the cascade exit section downstream of the trailing edge. It follows from this visualisation that the pressure-side leg of the horse-shoe vortex of the sense of rotation the same as that of the passage vortex con- centrates inside the passage vortex region. The fluid elements coming from the suction-side leg of the horse-shoe vortex are distributed at the borders of the passage vortex, losing their rotation (originally opossite to that of the passage vortex) as a result of interaction with it. The mutual location and interaction of the passage vortex, pressure-side leg and suction-side leg of the 326 P. Lampart horse-shoe vortex are main elements differentiating the presented secondary flow models. On the one hand, the differences between the models reflect the progress in understanding of the endwall flows in the course of time. On the other hand, they suggest that depending on the cascade geometry and flow thermodynamics, the picture of secondary flowsmay be not uniform but sub- ject to change, especially in point of mutual relations between the intensities of particular secondary flow structures. Themodels of Goldstein and Spores (1988), Doerffer and Amecke (1994), Wang et al. (1997) also illustrate the presence of a number of tertiary vortex structures, including corner vortices. They are shear-driven secondary flows or structures formed as a result of separation of the main secondary flows in the corners between the endwall and pressure or suction side of the blade. More corner vortices, called the leading edge corner vortices, appear in the endwall/pressure or suction side corners at the leading edge. The model of Wang et al. (1997) features also a wall vortex induced above the passage vortex at the suction surface. This vortex appears at a location, which in the earlier models was reserved for the suction-side leg of the horse-shoe vortex (Goldstein and Spores, 1988; Doerffer and Amecke, 1994). It should be noted that real vortex structures have the rotating intensity much lower than that presented in Fig.1, intentionally augmented for the sake of clarity. The vorticity contours in the cascade exit section typically exhibit two strong peaks of the opposite signs, that is two counter-rotating vortex struc- tures located close to each other at a some distance from the endwall. These vortex structures can be identified as the passage vortex and trailing shed vor- tex, Hawthorne (1951), Gregory-Smith and Cleak (1992). The trailing shed vortex originates from the vortex layer that is formed at the trailing edge due to the suction-to-pressure side velocity difference. Especially important is he- re the fact that the suction side velocity in the endwall region is affected by formation of the passage vortex. The vortex layer shed into the flow domain downstream of the trailing edge is quickly rolled up and assumes the sense of rotation opposite to that of the passage vortex. There is strong shear between the passage vortex and shed trailing edge vortex. The centres of secondary vortices are regions of a high turbulence level. The turbulence level at the exit section of the Durham cascade in the pas- sage vortex and trailing shed vortex measured by Gregory-Smith and Cleak (1992) was estimated at 35% with respect to the inlet velocity, yielding 16- 17% with respect to the local downstream velocity. The turbulence level in the suction-side corner vortex is nearly as high as above. The increase of tur- bulent fluctuations in the region of secondary vortices can be attributed to Investigation of endwall flows... 327 the process of deformation of the endwall boundary layer under conditions of high streamline curvature and acceleration of the main flow in the casca- de. On the other hand, these vortex flows wash away the endwall boundary layer towards the suction surface and give rise to relaminarisation of the do- wnstream endwall boundary layer. The newly formed endwall boundary layer becomes thin. It is gradually increasing in thickness but is constantly washed away. As a result, the endwall boundary layer has a highly three-dimensional character. Measurements of pressure pulsations in the endwall boundary layer made byHarrison (1989) and reproduced in this paper in Fig.3 show that this boundary layer is in the major part laminar (downstream of the horse shoe vortex lift-off lines) or intermittent, and becomes turbulent only in the rear part of the blade-to-blade passage at the suction surface. For amore detailed description of formation of secondary vortices and flows in endwall regions the reader is requested to refer to the work by Gregory-Smith (1997). Fig. 3. Division of the endwall boundary layer into the regions of laminar, intermittent and turbulent flow, Harrison (1989) Let us also consider the case of a cascade with a clearance. A schematic development of passage vortices and a tip leakage vortex in this case is de- picted in Fig.4 after Sjolander (1997). The presence of the tip gap over the blade usually eliminates the stagnation at the endwall, which typically occurs in the corner between the endwall and blade leading edge in the no-tip-gap configuration. Therefore, a horse-shoe vortex does not feature at the tip en- dwall unless the tip gap is very small. The flow at the tip endwall approaching 328 P. Lampart Fig. 4. Tip leakage and passage vortices at the tip endwall with a clearance, Sjolander (1997) the tip region above the leading edge of the blade is divided into two streams aiming towards low pressure regions at the suction surfaces of the neighbo- uring blades – amain stream of the tip leakage flow going through the tip gap over the blade and a stream of cross-flow going across the blade-to-blade pas- sage.The tip leakage flow leaving the tip gap separates fromthe endwall under conditions of adverse pressure gradient and forms a tip leakage vortex. The cross-flow blocked by the development of the tip leakage vortex also separates from the endwall and rolls up into a passage vortex. The stream dividing line between the tip leakage and cross-flow lies at the pressure side of the blade tip. The tip leakage andpassage vortices are characterised by the opposite sense of rotation. Usually, a dominant structure is the tip leakage vortex. The relations between the circulation and size of the two structures depend onmany factors such as the tip gap size and flow turning angle. 3. Evolution of vorticity from the endwall boundary layer Theprincipalmathematical descriptionof secondaryflowsdrawson the spatial evolution of the vorticity vector in the cascade. This description enables better understanding of the phenomena associated with the secondary flows. At the inlet to the cascade in a non-skewed endwall boundary layer, the vorticity can be assumed to have a direction normal to the main stream ve- Investigation of endwall flows... 329 locity. During the flow through the cascade, the shear flow is turned, which gives rise to the production of longitudinal (streamwise) vorticity, whereas the normal component of the vorticity is also changed. The starting point to de- scribe the evolution of the vorticity vector is the vorticity transport equation, which can be obtained from themomentum equation subject to the curl ope- rator, e.g. Sherman (1990). The vorticity transport equation for a steady flow of a viscous compressible gas can be written as below (with the effect of mass forces neglected) (u ·∇)ω=(ω ·∇)u−ω(∇·u)−∇× (∇p ρ ) +∇× (∇·τ ρ ) (3.1) where u is the velocity vector, ω – vorticity vector, p – pressure, ρ – density, τ – viscous stress tensor. Besides the left-hand-side convective term, the first right-hand-side term represents the vorticity production due to the velocity gradient (stretching and curving of vortex lines), second term – describes the vorticity production in a compressible flow (reduction in an expanding flow, increase in a compression region), third term – accounts for the baroclinic effect (in the field of spatial changes of pressure and density), fourth term – due to interaction of viscous forces in a compressible flow. LakshminarayanaandHorlock (1973) presenteda solution to this equation, which is of interest in turbomachineryapplications.Theyassumedacoordinate system (s,n,b), where s is the unit vector tangent to the streamline u= sq (q – velocity magnitude), n is the unit normal vector directed towards the streamline curvature centre n/R = s ·∇s (R – streamline curvature radius), whereas b is the unit binormal vector b= s×n so that (s,n,b) is a right- handed set of vectors. The differential coefficients of the unit vectors of the system fulfil the following relations ∂s ∂s = n R ∂s ∂n = n an ∂an ∂s ∂s ∂b = b ab ∂ab ∂s ∂n ∂s = b τ − s R ∂b ∂s =− n τ (3.2) where τ is the radius of torsion of the streamline, an – distance in the n direction between neighbouring streamlines, ab – distance in the b direction between neighbouring streamlines. In the assumed coordinate system, changes of the pressure distribution in an inviscid flow are ∂p ∂s =− 1 2 ρ ∂q2 ∂s ∂p ∂n =− ρq2 R ∂p ∂b =0 (3.3) 330 P. Lampart Changes of the streamwise andnormal component of the vorticity vector in the cascade, ωs, ωn, where ω= sωs+nωn+bωb, derived in Lakshminarayana and Horlock (1973) can be rewritten in the form ρq ∂ ∂s (ωs ρq ) = 2ωn R − 1 qρ2 (∂p ∂n ∂ρ ∂b − ∂ρ ∂n ∂p ∂b ) +[terms with viscosity] (3.4) 1 q ∂ ∂s (ωnq)= ωb τ − ωn ab ∂ab ∂s + 1 qρ2 (∂p ∂s ∂ρ ∂b − ∂ρ ∂s ∂p ∂b ) +[terms with viscosity] In practical applications, Eq. (3.4)1 for the streamwise vorticity seems to be of the main significance. Equation (3.4)2 for the normal vorticity is not that widely used.Values of the normal vorticity can be found in an alternative way. The above expressions do not expose terms connected with interaction of viscous forces, which in general have a very complex form. Although fluid elements that make up the passage vortex originate from endwall boundary layers, the dynamics of secondary flows within the blade-to-blade passage is often regarded in terms of interaction of the pressure and inertia forces, ne- glecting the viscous forces. The interaction of viscous forces is then limited to the formation of the inlet boundary layer and destruction of secondary vortex structures in the process of their mixing with the main flow downstream of the blade trailing edges. Lakshminarayana and Horlock (1973), also Horlock and Lakshminarayana (1973) obtained detailed solutions to Eqs. (3.4) for a number of model flow cases, including an inviscid constant-density flow, invi- scid incompressible stratified flow (with constant density along streamlines), inviscid compressible flow of a barotropic fluid or perfect gas, viscous incom- pressible flow aswell as inviscid incompressible flow in the rotating coordinate system. In themost simple case of an inviscid incompressible flowwith ρ = const, Eq. (3.4)1 yields ρq ∂ ∂s (ωs ρq ) = 2ωn R (3.5) Equation (3.5) indicates the effect of normalvorticity and streamline curvature on the process of generation of streamwise vorticity. Making use of an appro- ximate assumption that the normal vorticity and velocity magnitude remain unchanged along the streamlines and integrating by substitution dα = ds/R, yields the following equation (Squire andWinter, 1951) ωs−ωs0 =2ωn0(α−α0) (3.6) where ωs is the current streamwise vorticity, ωs0 and ωn0 are the streamwise and normal vorticities at the inlet to the cascade (with ωs0 = 0 for non- Investigation of endwall flows... 331 skewed boundary layers), whereas the difference α−α0 denotes the current flow turning in the cascade. In a steady inviscid flow, streamlines and vorticity lines lie on surfaces of constant total pressure (Bernoulli surfaces), and the normal vorticity can be obtained from scalar multiplication of the Gromeko-Lamb equation by the unit binormal vector b · (u×ω)= b · 1 ρ ∇ ( p+ q2 2 ) ⇒ ωnq = 1 ρ ∂p∗ ∂b = ∣ ∣ ∣ ∇p∗ ρ ∣ ∣ ∣ cosφ (3.7) which turns Eq. (3.5) into the form (Hawthorne, 1955) ρq ∂ ∂s (ωs ρq ) = 2 Rρq ∂p∗ ∂b or (ωs ρq ) 2 − (ωs ρq ) 1 = 2 ∫ 1 2 Rρq2 ∣ ∣ ∣ ∇p∗ ρ ∣ ∣ ∣ cosφ ds (3.8) where p∗ is the total pressure, φdenotes theangle between thebinormalversor and the normal to the Bernoulli surface (or direction of ∇p∗). Equation (3.7) exhibits a relationship between the streamwise vorticity and total pressure loss in the binormal direction. Equations (3.5) and (3.8) also remain valid for the case of an inviscid compressible flow of a barotropic fluid (ρ = f(p) – which follows from the fact that in the assumed coordinate system ∂p/∂b =0, from which ∂ρ/∂b = 0) and for the case of homoentropic flow of a perfect gas (p/ρκ = const). As shown in Shermann (1990), for the case of flow of a perfect gas with entropy gradients Eq. (3.8) reads as ρq ∂ ∂s (ωs ρq ) = 2 Rρ∗q ∂p∗ ∂b (3.9) where ρ∗ denotes the density calculated from the state equation using stagna- tion parameters. Lakshminarayana and Horlock (1973) also explain the effect of rotation on the development of secondary flows. The equation for the evolution of the vorticity along streamlines can be presented in the form (with viscous terms neglected) ρq ∂ ∂s (ωs ρq ) = 2ωn R + 2s · (Ω×ω) q − 1 qρ2 (∂p ∂n ∂ρ ∂b − ∂ρ ∂n ∂p ∂b ) (3.10) In the above equation, the vorticity is written in the absolute reference fra- me, whereas the streamlines are assumed in the rotating frame, thus s is the streamwise unit vector in the relative frame, q – relative velocity magnitu- de; Ω – vector of the angular velocity of the rotating reference system. The 332 P. Lampart rotating term disappears as long as the Bernoulli surfaces remain cylindrical, which can be assumed approxiomate for the case of axial turbines. Therefore, rotation of the blade-to-blade channels does not influence secondary flows in axial turbines. The situation looks different in radial (radial/axial) turboma- chines.More information on the physical nature of evolution of the streamwise vorticity in the secondary flow region can be found in Gregory-Smith (1997). Based on the known distribution of the streamwise vorticity in the exit section, the exit velocity of the induced flow and the secondary kinetic energy can be evaluated. Assuming that this energy is lost in the process of mixing with the main flow, one can estimate the level of mixing losses. The velocity induced at the trailing edge section can be numerically calculated from the Poisson equation for the stream function ψ ∆ψ =−ωs1 (3.11) The components of the induced secondary velocity vr1, vn1 are then found as vr1 =− ∂ψ ∂n vn1 = ∂ψ ∂r (3.12) Themass-averaged mixing loss coefficient due to secondary flows can be writ- ten as ξsec = hp 1 ∫ 0 1 ∫ 0 ρ1v1(v 2 r1+v 2 n1)cosα1 dxdy hp 1 ∫ 0 1 ∫ 0 ρ1v 3 1 cosα1 dxdy (3.13) 4. Endwall loss sources Production of endwall losses is a complex problem. Among basic loss me- chanisms in the endwall and secondary flow regions are (Sieverding, (1985); Gregory-Smith, 1997): • formation of the inlet boundary layer upstreamof theblade leading edge, • formation of the boundary layer downstream of the horse-shoe vortex lift-off lines, • shear effects along the horse-shoe vortex lift-off lines, separation lines and along dividing surfaces between the passage vortex, other vortices, main flow and blade surfaces, especially at the suction surface, Investigation of endwall flows... 333 • dissipation of the passage vortex, trailing shed vortex, corner vortices and other vortex flows in the process of their mixing with themain flow (it can be assumed that the secondary kinetic energy of the relative motion in the exit section is lost duringmixing). The processes of mixing due to secondary flows are usually not accompli- shed in the blade rowwhere they originate and are continued in the downstre- am blade-to-blade passages. Considerable non-uniformities in the distribution ofmagnitude and direction of the velocity at the inlet to the subsequent blade row may lead to local separations and upstream relocation of the laminar- turbulent transition at the blade in the secondary flow dominated region. Themost decisive for the level of endwall losses are theblade span-to-chord ratio, flow turning in the cascade and inlet boundary layer thickness, which is accounted for in all experiment-based loss correlations for turbine cascades, e.g. Craig and Cox (1971), Traupel (1977). In the correlation of Traupel, the endwall losses are reverselyproportional to the span-to-chord ratio in theblade span range in which there is no interference between secondary flow vortices from the opposite endwalls. For short blades this corelation is more complex. The endwall losses increasewith the increased flow turning and increased inlet boundary layer thickness. 5. Endwall loss diagram Denton (1999) evaluated the loss coefficient in the profile boundary layer by integrating the entropy production along the boundary layer. For an isolated turbine cascade, the profile loss coefficient can be expressed in the form ζ = T∆s 1 2 V 2 δ,te = TṠ 1 2 V 2 δ,te m = ∑ C p 2 cosα1 1 ∫ 0 CD ( Vδ Vδ,te )3 d ( x C ) (5.1) where the integration extends on both sides of the profile, T denotes tem- perature, ∆s – entropy rise, Ṡ – entropy production in the boundary layer, Vδ is the velocity at the edge of the boundary layer, Vδ,te – this velocity at the trailing edge, CD – dissipation coefficient for a turbulent boundary lay- er (CD = 0.002 – Schlichting (1968)), p – cascade pitch, C – profile chord, α1 – exit angle (measured from the normal to the cascade front). The above formula exhibits the importance of the cube from the surface velocity V 3δ (re- ferred to the exit isentropic velocity), state of the boundary layer, pitch/chord ratio and exit angle in generation of profile boundary layer losses. 334 P. Lampart Themethods of evaluation of boundary layer losses in the profile boundary layersmaynot in general apply to endwall boundary layers.However, Harrison (1989) and Denton (1999) indicate that the losses in a specifically evolving endwall boundary layer can roughlybeevaluated ina similarwayassuming the dissipation coefficient as for turbulent flow. Thus, the entropy production in the endwall boundary layer upstream of the leading edge Ṡ− and downstream of the trailing edge Ṡ+ can be found as Ṡ− =2CDp∆x−ρ V 3δ,le Tδ,le Ṡ+ =2CDp∆x+ρ V 3δ,te Tδ,te (5.2) where Vδ,le, Tδ,le, Vδ,te, Tδ,te denote velocities and temperatures at the edge of the boundary layer; ∆x−, ∆x+ are lengths of the endwall boundary layers, that is the distances along the endwalls from a referential inlet section to the leading edge and from the trailing edge to the exit section. The factor 2 in Eq. (5.2) indicates that both endwalls are considered.Any change of radius for an- nular cascades is neglected assuming the samevalue of pitch for both endwalls. Assuming that the mass flow rate is equal to m = Vxρph, where h denotes the channel height, Vx is the axial velocity Vx = Vδ,lecosα0 = Vδ,teα1 (α0, α1 – inlet and exit angles), and making use of the entropy definition of the loss coefficient ξ = TṠ/[0.5V 2δ,tem], the following relations for the loss coefficient upstream of the leading edge ξ− and downstream of the trailing edge ξ+ can be derived ξ− =4CD ∆x− h cos2α1 cos3α0 ξ+ =4CD ∆x+ h 1 cosα1 (5.3) The contribution of the first term is negligible for a low inlet-to-outlet velocity ratio in an expanding cascade, whereas the second term can be important for an extended boundary layer region downstream of the trailing edge ∆x+. The loss coefficient remains inversely proportional to the channel height and also to the flow turning in the second term. The loss coefficient in theblade-to-blade region, that is between the leading and trailing edgewill nowbeevaluated.The entropyproduction in the endwall boundary layerwithin theblade-to-bladepassage ṠM canbe foundasa surface integral ṠM =2CDCap 1 ∫ 0 1 ∫ 0 ρV 3δ Tδ d ( x Ca ) d (y p ) (5.4) where Ca is the axial chord (Ca = C cosγ, where γ denotes the stagger angle of the profile). The integral appearing above can be evaluated assuming a Investigation of endwall flows... 335 linear velocity profile across thepitch (that is between the suctionandpressure surface Vδ = Vp +(Vs −Vp)y, y ∈ 〈0,1〉, where Vs, Vp are not functions of the coordinate y), and also assuming a model of a thin profile evenly loaded along the chordwith a constant suction-to-pressure surface velocity difference (that is Vs−Vp =2∆V = const, Vs+Vp =2V , V cosα = Vx, where α is the current flow angle in the cascade changing between the inlet and outlet angles α0, α1). The whole calculation procedure was described by Lampart (2006) andwill not be repeatedhere.The following formula for the loss coefficient ζM accounting for the endwall boundary layer within the blade-to-blade passage was derived there ζM = TṠM 1 2 V 2 δ,te m = 4C2Dα1cosγ α1− tanα0 C h 4 3 c1 (5.5) This formula was obtained for the velocity ratio ∆V/Vx = √ c1/3c2 corre- sponding to the minimum value of loss coefficient in the profile boundary layer, which is equivalent to the optimum pitch/chord ratio, and also assu- ming that the function tanα changes linearly with the chord between tanα0 and tanα1. The functions c1 and c2 are c1 = 1 4 ( tanα1 cos3α1 − tanα0 cos3α0 +3c2 ) (5.6) c2 = 1 2 ( tanα1 cosα1 − tanα0 cosα0 +ln tanα1+ 1 cosα1 tanα0+ 1 cosα0 ) With the surface velocity cubed in the loss formula as well as with the linear surface velocity profile across the pitch postulated, and additionally assuming that Vp =0, it can be shown that themean endwall boundary layer losses per unit area formonly 1/4 of themean profile boundary layer losses. In fact, the surface pressure velocity can not be neglected and the endwall losses are larger than that. The sum of the endwall loss coefficients expressed by Eqs. (5.3) and (5.5), that is ξ−+ξM+ξ+ is plotted in Fig.5 as a function of cascade inlet and exit angles for the chord/span ratio C/h =1, stagger angle γ =30◦ (which gives the axial chord/span ratio Ca/h = 0.866) and for a typical axial extention of the inlet and exit region ∆x± = C/4. The boundary layer losses clearly increase with the increased exit angle. For a combination of inlet and exit angles typical for subsonic stator cascades of impulse turbines (e.g. with PŁK profiles – see part II of the paper) α0 = (−10 ◦,+10◦), α1 = (75 ◦,80◦), the 336 P. Lampart endwall boundary layer loss coefficient calculated from Eqs. (5.3) and (5.5) changes between 1.8-2.5%. As indicated by Denton (1993), 2/3 of the overall endwall loss falls on endwall boundary layers. The other part comes from mixing of the inlet boundary layer in the region of secondary flow formation, and fromdissipationof thekinetic energyof secondaryvortices, typically 1/4of the overall endwall loss. For a combination of the inlet and exit angles typical for rotor cascades of impulse turbines (e.g. with R2 profiles – also see part II) α0 =(−70 ◦,−60◦), α1 = (70 ◦,75◦), the endwall boundary layer loss coefficient calculated from Eqs. (5.3) and (5.5) changes between 1.5-2.0%, which is less than for the considered stator cascade. This can be explained by the lower cascade exit angle and, consequently, lower velocities in impulse stage rotor throats. But the overall endwall losses are by no means lower. In this case, the losses due dissipation of secondary vortices considerably increase. Fig. 5. Endwall boundary layer loss coefficient ξ − + ξ++ ξM for a given set of cascade inlet and exit angles calculated fromEqs. (5.3) and (5.5); C/h=1, Ca/h =0.866 The evaluation of losses due to mixing of secondary flows with the main streamusingelementarymethods is especially difficult since the secondaryflow patterns have a very individual character depending on the cascade geometry andflow thermodynamics.Thesemixing losseswere evaluated theoretically for an inviscid flow in the paper by Puzyrewski (1963). Based on the convection of vortex lines in a flow turning channel, the streamwise vorticity in the exit section was calculated there, from which the secondary kinetic energy was found. With the assumption that all this kinetic energy is dissipated, the Investigation of endwall flows... 337 mixing loss coefficient was derived there as (after adaptation of the original formula to the denotations used in this paper) ξsec = c2 ( α1+ h C C p )2 sin2(α1−α0) (cosα1+cosα0)2 cos4α1 cos4α0 (5.7) The above formula contains a constant c2 whose value was obtained from correlationwithexperimentaldata for a fewtypesof cascades, c2 =0.159.This formula explains large contribution of mixing losses due to secondary flows in cascades with front-loaded profiles having high negative inlet angles and high flow turning (typical in impulse stage rotor cascades, where α0 < −60 ◦), whereas the mixing losses are relatively small for cascades with the zero inlet angle (as in stator cascades). The increased possibilities of determination ofmixing losses are connected with the development of 3D flow solvers. 6. Comparison of numerical results with ERCOFTAC data A series of CFD computations of a 3D turbine cascade of Durham (Gregory- Smith, 1993-2002) were made with the help of 3D RANS solvers FlowER (fi- nite volumemethod, upwind differencing, thrid-order ENO scheme, Baldwin- Lomax or Menter k − ω SST turbulence model, see Yershov and Rusanov (1997), Yershov et al. (1998)) andFluent (finite volumemethod, second-order upwind,Reynolds StressModel – having basic features of the Launder-Reece- Rodi model, see Fluent Inc. (2000)). The calculations converging to a steady- state were made in a one blade-to-blade passage of the rotor cascade on H or O-H type grids refined near the blade walls and endwalls (y+ = 1-2), le- ading and trailing edges of the blades. Typically, there were 12-16 grid cells in each boundary layer. The number of cells in one blade row reached 500000 for H-type grids (92 axially, 76 radially, 72 pitchwise) and 1000000 for O-H grids. Regions of accumulation of secondary flow losses are located at a some distance from the endwalls, Hodson and Dominy (1987), Gregory-Smith and Cleak (1992). The area distribution of total pressure downstream of the tra- iling edge in a Durham linear cascade (pitch/chord – 0.85, span/chord – 1.79, inlet angle 43◦, exit angle −69◦) shown in Fig.6 exhibits a strong loss cen- tre containing two peaks due to the passage vortex and trailing shed vortex. Another loss centre is placed near the endwall in the region of the suction 338 P. Lampart side corner vortex. The total pressure peaks coincidewith places characterised by the highest viscous stresses and are located near the places of maximum vorticity. The picture of secondary flows evolves downstream as a result of dissipative processes. In other cascades, depending on the distance from the trailing edge, the area distribution of total pressure in the exit section features only one peak in the secondary flow loss centre due to the passage vortex, see Langston et al. (1977), Yamamoto (1987). Fig. 6. Durham cascade – total pressure contours at the slot 10 (28% axial chord downstream of the trailing edge); (a) experimental (Gregory-Smith, 1993-2002), (b) computed by FlowERwith k−ω SST, (c) computed by Fluent with RSM-LRR The pitch-averaged spanwise distribution of the total pressure coefficient (referred to the inlet dynamic head) in the exit section of theDurhamcascade illustrated in Fig.6 shows a maximum at 20% of the blade span from the endwall. The secondary flows give rise to non-uniformities in the velocity field. In the secondary flowdominated region, the flow turning angle is decreased on themid-span side of the secondary flow vortices, whereas on the endwall side it is increased. This flow underturning and overturning can also be observed in Fig.7. The comparison of experimental and computational results for the Dur- ham cascade presented in Figs. 6 and 7 can be considered a validation of the computational methods used in this paper. The turbulent viscosity models Investigation of endwall flows... 339 Fig. 7. Durham cascade – spanwise distribution of the loss coefficient (a) and exit swirl angle (b) at the slot 10 (28% axial chord downstream of the trailing edge); experimental (Gregory-Smith, 1993-2002), computed by FlowERwith Baldwin-Lomax orMenter k−ω SST and computed by Fluent with RSM-LRR seem to be able to qualitatively predict basic features of the secondary flows. However, they overpredict the level of losses in the wake and in the secondary flow region. Quantitative predictions are improved with the Reynolds stress model, but they are connected with a significant increase of computational costs. 7. Conclusions The formation of endwall flows and evolution of vorticity from the endwall boundary layers was explained in the paper. Severalmodels of secondary flows in turbine cascades were presented. An analytical expression was derived to approximately describe the level of losses in the endwall boundary layers by integrating the entropy increase in the endwall boundary layer and assuming a model of a thin profile evenly loaded allong the chord. The endwall losses were also calculated with the help of CFD in a 3D planar cascade of Durham and comparedwith the results of available experimental data of ERCOFTAC.The 340 P. Lampart calculations based on turbulent viscositymodels are capable of capturingbasic 3D flow effects in cascade flows. However, they overpredict the level of losses in the wake and in the secondary flow region. The quantitative predictions are improved with the Reynolds stress model, but they are connected with a significant increase of computational costs. References 1. Craig H.R.M., Cox H.J.A., 1971, Performance estimation of axial flow tur- bines,Proc. Inst. Mech. Eng., 185, 32/71, 407-424 2. Denton J.D., 1993,Lossmechanisms in turbomachines,Trans.ASMEJ.Tur- bomachinery, 115, 621-656 3. Denton J.D., 1999, Loss mechanisms in turbomachines, Part I – Entropy creation in fluid flows, Part II – Loss generation in turbomachines, VKI LS 1999-02 4. Doerffer P., Amecke J., 1994, Secondary flow control and stream-wise vor- tices formation,ASME Paper 94-GT-376 5. Doerffer P., Rachwalski J., Magagnato F., 2001, Numerical investi- gation of the secondary flow development in turbine cascade, TASK Q., 5, 2, 165-178 6. Eckerle W.A., Langston L.S., 1987,Horseshoe vortex formation around a cylinder,Trans. ASME, J. Turbomachinery, 109, 278-286 7. Fluent/UNS/Rampant, 2000,User’s Guide, Fluent Inc. 8. Goldstein R.J., Spores R.A., 1988, Turbulent transport on the endwall in the region between adjacent turbine blades, Trans. ASME, J. Heat Transfer, 110, 862-869 9. Gregory-SmithD.G., 1993-2002,Durham low-speed turbine cascade,Turbo- machineryworkshoptest caseNo.3,Proc. ERCOFTACSeminar andWorkshop on Turbomachinery Flow Prediction, I-VIII 10. Gregory-Smith D.G., 1997, Secondary and tip-clearance flows in axial tur- bines,VKI LS 1997-01 11. Gregory-SmithD.G.,CleakJ.G.E., 1992,Secondaryflowmeasurements in a turbine cascadewithhigh inlet turbulence,Trans.ASME, J.Turbomachinery, 114, 173-183 12. Harrison S., 1989, Secondary loss generation in a linear cascade of high tur- ning turbine blades,Trans. ASME, J. Turbomachinery, 111, 618-624 Investigation of endwall flows... 341 13. Hawthorne W.R., 1951, Secondary circulation in fluid flow,Proc. Roy. Soc. Ser. A, 206, 374-387 14. Hawthorne W.R., 1955, Rotational flow through cascades,Quart. J. Mech. Appl. Math., 8, 3, 266-279 15. HodsonH.P., DominyR.G., 1987, Three-dimensional flow in a low pressure turbine cascade at its design conditions, Trans. ASME, J. Turbomachinery, 109, 278-286 16. Horlock J.H., Lakshminarayana B., 1973, Secondary flows: theory, expe- riment, and application in turbomachinery aerodynamics,Ann.Rev. FluidMe- chanics, 5, 247-279 17. Lakshminarayana B., Horlock J.H., 1973, Generalised expressions for se- condaryvorticityusing intrinsic co-ordinates,J. FluidMech.,59, Part1, 97-115 18. Lampart P., 2006, CFD investigation of aerodynamics of turbine blading sys- tems, Zeszyty Naukowe IMP PAN, 544/1503/2006 [in Polish] 19. Langston L.S., 1980, Crossflows in a turbine cascade passage,Trans. ASME, J. Eng. Power, 102, 866-874 20. Langston L.S., Nice M.L., Hooper R.M., 1977, Three dimensional flow within a turbine cascade passage,Trans. ASME, J. Eng. Power, 99, 21-28 21. Marchal Ph., Sieverding C.H., 1977, Secondary flows within turbomachi- nery bladings,Proc. AGARD-CP-24 Secondary Flows in Turbomachines, 11 22. Moore J., Smith B.L., 1984, Flow in a turbine cascade: Part 2 – Measure- ments of flow trajectories by ethylene detection, Trans. ASME, J. Eng. Gas Turbines and Power, 106, 409-414 23. Puzyrewski R., 1963, Convection of vortex lines in curved ducts as a means for calculation of endwall losses,Trans. IFFM, 17, 63-108 [in Polish] 24. Schlichting H., 1968,Boundary Layer Theory, McGraw-Hill, NewYork 25. SharmaO.P.,ButlerT.L., 1987,Predictionof endwall losses and secondary flows in axial flow turbine cascades, Trans. ASME J. Turbomachinery, 109, 229-236 26. Shermann F.S., 1990,Viscous Flow, McGraw-Hill, NewYork 27. Sieverding C.H., 1985, Recent progress in understanding of basic aspects of secondaryflows in turbine blade passages,Trans. ASME, J. Eng. Gas Turbines and Power, 107, 248-257 28. Sjolander S.A., 1997, Secondary and tip-clearance flows in axial turbines, VKI LS 1997-01 29. SquireH.B.,WinterK.G., 1951,The secondaryflow inacascadeof aerofoils in a non-uniform stream, J. Aeronaut. Sci., 18, 271-277 342 P. Lampart 30. Traupel W., 1977, Thermische Turbomaschinen, Band I, Springer-Verlag, Berlin 31. Wang H.P, Olson S.J., Goldstein R.J., Eckert E.R.G., 1997, Flow visualisation of a linear turbine cascade of high performance turbine blades, Trans. ASME J. Turbomachinery, 119, 1-8 32. Yamamoto A., 1987, Production and development of secondary flows and losses in two types of straight turbine cascades: Part 1 – A stator case, Part 2 – A rotor case,Trans. ASME J. Turbomachinery, 109, 186-200 33. Yershov S., Rusanov A., 1997, The application package FlowER for the calculation of 3D viscous flows through multi-stage machinery, Certificate of registration of copyright,Ukrainian State Agency of Copyright and Related Ri- ghts, Kiev, February 19 34. Yershov S., Rusanov A., Gardzilewicz A., Lampart P., Świrydczuk J., 1998, Numerical simulation of viscous compressible flows in axial turboma- chinery,TASK Quart., 2, 2, 319-347 Badanie przepływów i strat brzegowych w turbinach osiowych. Część I. Formowanie się przepływów w strefie brzegowej i strat brzegowych Streszczenie Przepływybrzegowe stanowią jedno z głównych źródeł strat w turbinach.Wpra- cy w skrócie wyjaśniono proces formowania się przepływów brzegowych i ewolucję wirowości pochodzącej z brzegowych warstw przyściennych. Przedyskutowano wyni- kające stąd stratyprzepływu.Wyznaczono teoretycznie stratyprzepływuwbrzegowej warstwie przyściennej poprzez całkowanie przyrostu entropii wzdłuż warstwy brzego- wej i zakładając pewienmodel obciążenia profilu łopatkowego. Z pomocą programów komputerowych numerycznej mechaniki płynów wyznaczono także przepływ i straty brzegowe w prostej palisadzie turbinowej Durham i porównano otrzymane rezultaty z danymi eksperymentalnymi ERCOFTAC. Manuscript received January 20, 2008; accepted for print November 7, 2008