Turbulent Flow and Endwall Heat Transfer Analysis in a 90 8 Turning Duct and Comparisons with Measured Data Part II : In ̄ uence of Secondary Flow , Vorticity , Turbulent Kinetic Energy , and Thermal Boundary Conditions on Endwall Heat Transfer

Fundamental character of forced convection heat transfer to the endwall surface of a gas turbine passage can be simulated by using a 90 turning duct. Elevated operating temperatures in gas turbines require a thorough understanding of the turbulent thermal transport process in the three-dimensional end-wall boundary layers. The current study uses an in-house developed three-dimensional viscous ̄ow solver to computationally investigate the heat transfer character near the endwall surfaces. Extensive heat transfer experiments also illuminate the local heat transfer features near the endwall surface and form a baseline data set to evaluate the computational method used. Present experimental eort at Re5 342,190 employs a prescribed heat ̄ux method to measure convective heat transfer coecients on the end-wall surface. Local wall temperatures are measured with liquid crystal thermography. Local convective heat ̄ux is determined by solving the electric potential equation with constant uniform potential that is applied along the inlet and exit boundaries of the endwall. The viscous ̄ow and heat transfer computation algorithms are based on the same methods presented in Part I. The in ̄uence of the two primary counter-rotating vortices developing as a result of the balance of inertia and pressure forces on the measured and computed endwall heat transfer coecients is demonstrated. It is shown that the local heat transfer on the endwall surface is closely related to the structure of the three-dimensional mean ̄ow and the associated turbulent ̄ow ®eld. A comparison of the molecular/ turbulent heat diusion, wall shear stress, visualization of local turbulent kinetic energy are a few of the tasks that can be performed using a numerical approach in a relatively time ecient manner. The current numerical approach is capable of visualizing the near wall features that can not be easily measured in the ̄ow ®eld near the end-wall surfaces.

high quality ¯uid mechanics and heat transfer data from carefully executed experiments.The current study is a detailed assessment of a general computational heat transfer method that can be used in the prediction of heat transfer rates in thermal turbomachinery.Numerical heat transfer performance of the method is tested against measured data.The turning duct eectively simulates the streamwise acceleration characteristics, passage vortices and endwall boundary layers of a turbomachinery passage.The turning duct is also an ideal environment to obtain ¯uid mechanics/heat transfer data with high accuracy and resolution in a controlled laboratory environment.
In thermal turbomachinery applications, a variation in heat transfer rate due to a small ¯ow disturbance can lead to an increase in thermal stress and a decrease in component life.This is true for both hot gas side and coolant side of turbomachinery passages.On a highly curved wall, the change in heat transfer rate is mainly due to an increase or decrease of the turbulent mixing by the eect of the streamline curvature.It has been indicated in von Karman's stability argument (1934) that the convex wall has a stabilizing eect on ¯uid particles, while the concave wall has a de-stabilizing eect with respect to an equivalent reference ¯at plate.The measurement and prediction of heat transfer rates for a two-dimensional boundary layer on a concave and convex surface have been presented by Mayle et al. (1979).It was found that the heat transfer on the convex surface was less than a ¯at surface having the same free stream Reynolds number and turbulence.Concave surface heat transfer was augmented when compared to the ¯at surface.Good agreement between the numerical results and heat transfer experiments was noted for the convex surface when a twodimensional dierential boundary layer code was used with modi®ed curvature model.For the concave surface, the agreement with the measured heat transfer data was poor due to the uncertainties in the turbulence model.Camci (1985) obtained similar conclusions in heat transfer experiments performed on a gas turbine rotor blade under realistic free stream conditions.
Secondary ¯ow in a duct has been generally classi®ed into two categories.Secondary ¯ows of the ®rst kind were derived from the unbalanced body force caused by the change of streamline curvature.Secondary ¯ows of the second kind in a straight duct of rectangular cross-section were observed in early twenties.It was not until the work of Brundrett and Baines (1964); Perkins (1970); and Launder and Ying (1972) that a complete description was provided.From an analysis of the vorticity transport equation in the streamwise direction, they deduced that the secondary velocities were generated by the Reynolds stress gradients in the plane of duct cross-section.Later, Melling and Whitelaw (1976) gave a quantitative description of Reynolds stress, kinetic energy and secondary velocity in the fully developed ¯ow.Although the magnitude of turbulence driven secondary ¯ow motion is only the order of 2 ± 3% of the streamwise mean velocity, the motion causes the streamwise mean velocity and temperature ®elds to be distorted considerably toward the corners, and it can produce a substantial modi®cation in the heat transfer coecient.The study of the in¯uence of secondary ¯ow on heat transfer dates back to 1967.Early measurements of heat transfer coecient and secondary ¯ow were performed in a vertical straight duct.Numerical prediction of this ¯ow with algebraic and anisotropic two-equation turbulent models was performed by Emery et al. (1980) and Myong (1991).Their investigation indicated that the wall shear stress/wall heat ¯ux ®rst arises from the symmetry plane toward the corner, with a peak shear stress/heat ¯ux about midway between the corner and midpoint of the duct sides, and then falls again near the corner, approaching zero at the corner.In a curved duct ¯ow, the major secondary ¯ow is of the ®rst kind secondary ¯ow but little is known about its eect on the heat transfer coecient.Only a few experimental measurements in this area were performed in the last few decades and the numerical predictions have not appeared in the literature.Mori et al. (1971) presented an analytical and experimental study for a fully developed laminar ¯ow in a 220 square duct with a constant-wall-heat-¯ux boundary condition.The ¯ow and temperature ®elds were divided into two regions as the core and boundary layer regions.The analytical results were obtained by considering the balance of kinetic energy and entropy production in the boundary layer.Temperature distributions and local Nusselt numbers on the wall of a 180 bend have been measured by Johnson and Launder (1985).The experimental results at a Reynolds number of 56,030 showed that the ratio of heat transfer coecient of the concave and convex surface of 90 section was about 2 : 1 at mid-span.Other similar measurements for a strongly curved duct ¯ow were documented in Graziani et al. (1980) and Mayle et al. (1979).Recently, a comprehensive description of streamwise mean ¯ow, secondary ¯ow, and endwall heat transfer coecient in a 90 turning duct was presented by Wiedner (1994) and Wiedner andCamci (1996a, b, 1997).Several distinct regions of the heat transfer coecient on the endwall were characterized as the result of the interactions between turbulent ¯ow ®eld and endwall thermal boundary layer.The non-uniform endwall heat ¯ux boundary condition was determined accurately by applying electric ®eld theory and a ®nite element method, Wiedner and Camci (1986a).
At the present time, comparisons between the calculated and measured local endwall heat transfer in a curved duct of square cross-section have not appeared in the open literature.It is the intent of this paper to provide a better understanding of the in¯uence of the secondary ¯ow as well as the local turbulent ¯ow characteristics on the endwall heat transfer coecient.In the present study, interactions of the vorticity, turbulent kinetic energy, and turbulent shear stress with the endwall heat transfer were investigated.The data for comparison were based on recent measurements in a 90 turning duct by Wiedner and Camci (1997).The accurate endwall thermal boundary condition of nonuniform heat ¯ux was prescribed during the solution of thermal energy.The computational technique used for the solution of the turbulent ¯ow ®eld in Part I is extended to include the thermal energy equation in this paper.

NUMERICAL MODEL AND THERMAL BOUNDARY CONDITIONS
The Energy Equation For three-dimensional steady, incompressible, turbulent ¯ows, the governing equations and the solution method for the momentum, energy and turbulent kinetic energy equations were described in Part I. Since the ¯ow is assumed incompressible, the energy equation can be decoupled from the solution procedure.The conservative form of convection-diusion equation for the energy equation is as follows: @ @x U x T ÿ ÿ @T @x @ @y U y T ÿ ÿ @T @y @ @z U z T ÿ ÿ @T @z 0 1 where ÿ =Pr t =Pr t .Equation [1] was obtained under the following assumptions: (1) C p is a constant; (2) the molecular thermal diusion is modeled with the Fourier's law; (3) the turbulent heat ¯ux is modeled in Boussinesq form; and (4) the dissipation term and pressure gradient are negligible when compared to the convective terms.The laminar and turbulent Prandtl numbers, are chosen to be 0.7 and 0.9 for air respectively.The discretized equation and the solution method for the energy equation are similar to those for the momentum and turbulence equations.The details of the transformations from the Cartesian coordinates (x, y, z) to the generalized coordinates (, , ) were discussed in Part I. Near the wall, the following logarithmic temperature pro®le of Launder and Spalding (1974) is employed to build the relationship between the wall temperature and the temperature at the ®rst grid point, provided that the ®rst grid point is located in the fully turbulent region.where constant E and are 9.739 and 0.4187.A is the Van Driest's constant, equal to 26.0.The viscous sub-layer thickness was determined from the condition y p 4. The energy equation was considered as converged when the sum of the energy ¯ux residuals at all nodes was less than 0.01% of the inlet energy ¯ux.All the computations were performed with 81 2 49 2 49 grid points in the streamwise, vertical, and radial directions to obtain grid independent solutions.

Thermal Boundary Conditions
The geometry of the duct with heated endwall surface is shown in Figure 1.For the unheated walls and the inlet condition, the temperature is set to the ambient temperature; i.e., 298 K.At the exit of the duct, zero temperature gradient in the streamwise direction is prescribed.Prescription of the heat ¯ux ®eld on the top wall requires the solution of an electrostatic boundary value problem described in Wiedner and Camci (1996).The numerically determined wall heat ¯ux distribution on the endwall surface is used as a boundary condition both in computations and experiments.For a two-dimensional, isotropic, homogeneous conducting medium with zero free charge, the electrical potential V must satisfy, r 2 Vx; y 0 3 The electric boundary conditions for the heated surface are uniform potential at the beginning (line AB) and the end (line CD) of the medium along Y direction and zero current ¯ow normal to the unbounded streamwise edges.An electric ®eld vector can be de®ned as the gradient of the scalar potential ®eld, V(x, y).

Ẽ ÿ rV 4
Once the electric potential and electric ®eld vector are obtained, a scalar multiplication of the electric ®eld and electrical conductivity of the medium provide the current density ®eld, J Ẽ 5 Finally, the heat ¯ux is determined from q 00 gen Ẽ 1 J 6 where is the thickness of the electrically conducting medium.This general local heat ¯ux formulation allows us to obtain an accurate thermal endwall boundary condition for the solution of the energy equation.

Prescribed Heat Flux on the Endwall
The geometry of the 90 turning duct with mean radius to duct width ratio of (r i r o )/2 2.3 is shown in Figure 1.The top endwall surface shown as the shaded wall was electrically heated by applying DC current to bus bars attached along the inlet line AB and exit line CD.The rest of the Plexiglas walls were treated as constant temperature surfaces kept at ambient temperature of 298 K.A 5 VDC equi-potential line at the leading edge of the top wall (AB) and 0 VDC equi-potential line at X 90 (CD) were used as the prescribed electrical potential boundary conditions.Along the curved inner and outer wall boundaries, no cross-current electrical ¯ow boundary condition was imposed.Electrical current was only allowed to ¯ow tangential to the curved boundaries.The prescribed local heating in the conducting foil that was made of Inconel was computed from Eqs. [3] ± [6] by using a ®nite elements based solver described in Wiedner and Camci (1996b).The endwall heat ¯ux distribution obtained from the numerical solution of the electrostatic boundary value problem is shown in Figure 2. A correction on the generated heat ¯ux was needed due to the conduction and radiation losses from the endwall surface.In the far upstream region, the prescribed heat ¯ux was relatively uniform.The heater foil shape in the inlet section was close to a rectangular shape that has the capability to generate almost uniform local heat ¯ux.The curved section of the duct started at about 2.5 D downstream of the inlet section.As soon as the turning section is encountered, an almost elliptic heat ¯ux distribution was observed.The ®nal section of the straight part of the inlet duct was also slightly in¯uenced from the distributions in the curved section because of the elliptic nature of electrical potential ®eld.The overall electrical current ¯ow in the thin foil was comparable to the inviscid, irrotational, two-dimensional ¯uid ¯ow through the endwall surface.The electrical potential V(x, y) corresponded to potential function of the inviscid ¯ow (x, y).The current density ®eld vector was also analogous to the velocity vector.Figure 3 shows the endwall static temperature distribution as numerically calculated from the solution of Eq. [1].In the entrance region, the endwall showed a continuous temperature increase from 298 to about 307 K within the ®rst duct width (D) distance downstream.Higher temperatures near the convex surface were consistent with the existence of large magnitude current density vectors in this area.

Measurement Uncertainties on Heat Transfer Coecient h
The experimental uncertainty of the convective heat transfer coecient was estimated according to the procedures given by Kline and McClintock (1953).The uncertainties were via the ®nite element method used in the prescription of generated heat ¯ux results in reasonably low h uncertainty.and Camci (1994).The free-stream temperature was ®xed at an ambient temperature of 298 K.
Several distinct regions of interest on the endwall have been labeled A ± G and Paths I and II, Figure 4b.The entry region (A) prior to the 90 turn shows a streamwise decrease in the heat transfer coecient from the inlet to the 0 cross section.The streamwise decline in heat transfer coecient is controlled by the growth of hydrodynamic and thermal boundary layers.Narrow regions of high heat transfer with strong gradients in the cross-stream direction are shown near the inner and outer radius corners (B).Speci®c corner ¯ows, as described by Brundett and Baines (1964) and Gessner (1973), in these regions may be responsible for higher heat transfer levels.Figure 4b shows that the strong gradient regions of h in the straight inlet section of the duct coincide with the regions that have the highest turbulent velocity ¯uctuations.
An enhancement in heat transfer (D) followed by a low heat transfer region (E ) occurs near the inner radius surface between 0 and 45 cross sections where the initial development of cross stream velocities takes place.Similar patterns were observed in the cascade results of Goldstein and Spores (1988).They attributed the high heat transfer zone to a local highly turbulent ¯ow that resulted from the transverse pressure gradient turning the endwall boundary layer.In regions D and E, shown in Figure 4b, a highly accelerated boundary layer on the convex inner surface and the endwall boundary layer merge together in the corner region.The mean ¯ow in this corner region after the 0 section is highly three-dimensional.The endwall boundary layer ¯uid starts to develop the secondary velocities in a direction from the concave outer wall to convex inner wall between the 0 and 45 locations.Between the inlet section and the 0 section, the inlet endwall boundary layer on the heat transfer surface is not subject to streamwise curvature.The inlet ¯ow away from the corners has almost a twodimensional core ¯ow structure as shown in Part I of this paper.Immediately after 0 section, near region D, this structure is somewhat disturbed by the immediate development of centrifugal forces, static pressure gradient in the Y direction, stagnation pressure gradient in the Z direction, and the highly accelerated boundary layer on the inner radius side wall.Region D experiences a locally enhanced mode of turbulent heat transport (77 ± 82 W/m 2 K ) due to its highly strained mean ¯ow structure.Immediately downstream of region D, a relatively reduced heat transfer coecient island (E ) (67 ± 72 W/m 2 K ) exists.Turbulent momentum and heat exchange in this region may be reduced due to further acceleration of the ¯uid near the inner radius wall between the 20 and 45 cross sections.
At the 0 cross section, a gradient in the heat transfer coecient exists with higher levels near the inner radius surface and lower levels near the outer radius surface (F).The favorable and adverse pressure gradients near the inner and outer surfaces, respectively, resulted in accelerated heat transfer level (Path I).This path forms the characteristic wedge shape that has been described by Graziani et al. (1980) and can also be seen in the cascade results of Gaugler and Russell (1984); Goldstein and Spores (1988); and Boyle and Russel (1989).The path crosses the 45 cross section at an approximate value of Y/ D 0.25.Detailed ¯ow data shows that the highest secondary velocity vectors adjacent to the endwall occur at this position.Similar circumstances are apparent at the 90 cross section.Although the streamwise component of the mean velocity is relatively reduced in the core (G ) of the passage vortices, the related local turbulence enhancement due to enhanced levels of secondary kinetic energy produces relatively high convective heat transfer levels comparable to the inlet section.
Traversing the endwall radially inward, a decrease in the heat transfer coecient occurs as the inner radius surface and endwall corner is approached.This decrease in heat transfer may be the result of the secondary velocity reduction followed by the abrupt turning of the ¯ow onto the inner radius surface.The secondary ¯ow conditions are similar to those present in a longitudinal vortex.Eibeck and Eaton (1987) showed a decrease in the Stanton number on the up-wash side of the vortex.
Finally in the outer radius region of the passage endwall between the 45 and 90 cross sections, the streamwise pressure gradient changes from adverse to favorable.This pressure gradient reversal results in an acceleration of the streamwise velocity.The endwall heat transfer coecient distribution in this region indicates a path of increasing heat transfer that continues through the 90 cross section, Path II.The elevated heat transfer coecient values (87 ± 92 W/m2K) measured near the outer surface corner at the 90 section lie in this path of increasing heat transfer.

y 1 Distribution Near the Endwall Surface
The viscous sub-layer in the present computations was de®ned as the ¯uid layers that occupied the region where y 4. Figure 5 is a plot of non-dimensional distance y w = p 1 y= for the ®rst two planes parallel to the endwall (Y, Z plane) passing from the ®rst three grid points.In the plane passing from the ®rst grid point, all contour lines indicated that the viscous sub-layer is below the ®rst grid point.The minimum observed y value was always greater than 7 in the ®rst plane.The highest y values were encountered in the region where passage vortex clearly aects the endwall heat transfer.This region as surrounded by contour line D, contained the highest wall shear stress magnitudes.Similar qualitative trends were observed in the plane passing from the second grid point.Figure 6 shows the distribution of predicted TKE on the endwall surface from the computational model discussed previously.Turbulent kinetic energy is plotted in three horizontal planes passing from the ®rst second and third grid points in the domain.There is a strong similarity between the measured/computed heat transfer contours and the computed TKE distribution near the endwall surface.The computational simulations predict higher levels of turbulent transport in the third (highest TKE) and second (higher TKE) planes than the ®rst plane near the endwall.detailed measured hydrodynamic data presented in Wiedner (1994).Vorticity augmentation is much more signi®cant near the inner radius corner than near the outer radius corner.The augmentation of the magnitude of the mean vorticity vector near the inner radius corner is also shown from measurements and computations, in Figure 7 of Part I. Current computations were capable of showing the direct relationship between heat transfer coecient (Figure 4), TKE (Figure 6) and vorticity magnitude (Figure 7).Figures 6 and 7 indicate that the computed mean vorticity magnitude coincides with the regions of enhanced turbulent ¯ow activity.The points on the endwall surface under the in¯uence of these high TKE regions could experience enhanced turbulent heat transport from the endwall.The highest heat transfer rates are at the entrance section of the straight section of the duct and near the convex side of the endwall surface that is exposed to the passage vortex system.
Comparison of measured heat transfer coecients with numerical computations showed about 10% to 20% underprediction on the heated endwall surface.Using a wall function approach may not be the most accurate method of resolving the ¯uid mechanics and heat transfer characteristics of near wall regions.However, the use of wall functions is a useful approximation that reduces the overall time consumption of the computation.This approach also eliminates the need for an extremely ®ne computational grid that may not be practical for the current viscous ¯ow calculations.Computational results suciently simulate the general features of the measured heat transfer coecients on the endwall surface.The characteristic wedge shape (Path I) observed in the measured heat transfer distribution is fully simulated by the computations as shown in Figure 4a.The heat transfer in¯uence of the secondary ¯ows as indicated by region (G) of measurements is also recovered in the numerical simulations.The contour levels 8, 9, A and B of the computed distribution as shown in Figure 4a are clearly visible in the measured distribution presented Figure 4b.

Independence of h from Thermal Boundary Conditions
In forced convection heat transfer, the heat transfer coecient h is de®ned as a hydrodynamic indicator that should be independent of thermal boundary conditions.As long as the local Reynolds number distribution in the ¯ow is preserved, variations in wall temperature or in free stream temperature does not in¯uence the magnitude of h.The endwall heating rate can be easily changed by varying the applied voltage to the electrical bus bars located along line AB and CD.Two dierent endwall heating schemes resulted in very similar heat transfer coecient distributions con®rming the idea that local wall heat ¯ux in forced convection problems is linearly dependent on the temperature dierence between the free stream and the wall.This idea is closely related to the fact that momentum and thermal energy equations are not coupled in incompressible ¯ow.

Temperature Distribution in the Passage
The in¯uence of secondary ¯ows on local temperature distributions in a 90 turning duct can be numerically visualized for the current conditions with a uniform inlet temperature of 289 K.The local thermal energy released from the top endwall surface is mainly distributed in the duct by forced convection, molecular and turbulent diusion of heat and viscous dissipation.The resulting temperature distributions in 0 , 45 , and 90 cross sections are given in Figure 8.
At 0 , the temperature distribution appears to be symmetric about the plane Y/D 0.5 and upstream eect of the bend is little.Recall that on the top wall, the temperature distribution is not uniform as shown in Figure 3.The steep gradients can only be seen in the near-wall region where pure conduction process exists.At 45 , the ¯uid particles absorb more heat from the top wall and the secondary ¯ow motion convects most of the thermal energy produced near the endwall to the inner wall.The cross ¯ows on the endwall surface and the passage vortex clearly bring more thermal energy into the core of passage vortex from the top endwall surface as shown in Figure 8b.This phenomenon repeats itself in a stronger fashion as one moves from 45 section to 90 section in the streamwise direction.Contour line 1 is the boundary between the thermally undisturbed core ¯ow zone (298 K ) and the thermally contaminated zone.Figure 8c indicates the area coverage of the heated zone in the duct.The in¯uence of the endwall heating can be felt in a non-negligible fashion even at the symmetry plane of the duct near Z/D 0. It can be concluded from Figure 8   that cross ¯ows and secondary vortices can transport mass, momentum and heat from the concave part of the endwall to the convex end at a signi®cant rate.Due to the highly rotational character of the ¯ow near the convex surface, the transport mechanism is not limited to near endwall region.    of the passage vortex.This region may be viewed as a ¯ow zone rearranging the thermal energy distribution in a signi®cant manner because of the existence of the passage vortex system.The contour plot of the magnitude of the turbulent heat ¯ux vector is given in Figure 9c.Endwall cross ¯ow and passage vortex in¯uence on turbulent transport of heat clearly marked.Figure 9c suggests that the highest levels of turbulent heat transfer may appear near the passage vortex core, in addition to the near endwall region.
The magnitude of total streamwise heat ¯ux including molecular and turbulent diusion components is plotted in Figure 9d.This approach is useful in showing what speci®c directions are more dominant in transferring heat by molecular and turbulent diusion process.Two dierent heat ¯ux zones are observed near the convex surface, i.e., positive and negative zones.The negative zone is caused by the position change of the core of the temperature contour, just like the position change of vortex center in the streamwise direction.Since the heat ¯ux is a function of temperature dierence, a negative heat ¯ux may result from the decreasing temperature in the streamwise direction.Moreover, the magnitude of streamwise heat ¯ux was much less than the cross-stream magnitude representing the secondary ¯ow eect.This implied that most of the heat diusion process in the ¯ow passage was in the Y ± Z plane, not in the streamwise direction.

CONCLUDING REMARKS
Convective heat transfer near the endwall surface of a 90 turning duct has been studied by using experimental ¯ow/ heat transfer methods and an in-house developed computational viscous ¯ow/heat transfer solver.
When the endwall boundary layer develops in the curved section of the duct, cross-stream components of the velocity vector are enhanced due to the generation of the passage vortex.The endwall boundary layer is also subject to streamwise curvature.The interaction of the endwall boundary layer with the inner and outer radius sidewall boundary layers creates complex local wall heating rate distributions near the corner regions.
Elevated levels of local heat transfer coecients on the endwall coincide with the ¯ow regions that have high levels of measured vorticity and secondary velocity magnitude.In general, the regions of low heat transfer were associated with the locations where local streamwise velocities are relatively low.
A strong similarity between the measured/computed heat transfer contours and the computed TKE distribution near the endwall surface was observed.
The near wall vorticity distributions are also qualitatively similar to TKE and heat transfer coecient distributions.These results show that the local TKE level and the associated endwall vorticity signi®cantly aect the endwall heat transfer behavior.In general, the computed mean vorticity magnitude coincides with the regions of enhanced turbulent ¯ow activity.The points on the endwall surface under the in¯uence of these high TKE regions could experience enhanced turbulent heat transport from the endwall.
Computational simulations can display the general features of the measured heat transfer coecients on the endwall surface.The characteristic wedge shape (Path I) observed in the measured heat transfer distribution is fully simulated by the computations shown in Figure 4a.The heat transfer in¯uence of the secondary ¯ows as indicated by region (G) of measurements is also recovered from the numerical simulations.
Although at 0 , the temperature distribution in the duct cross section appears to be symmetric about the plane Y/ D 0.5, at 45 , the ¯uid particles absorb more heat from the top wall.The cross ¯ow on the endwall surface and the passage vortex clearly brings more thermal energy into the core of the passage vortex from the top end-wall surface.
Cross-¯ows and secondary vortices can transport mass, momentum and heat from the concave part of the endwall to the convex end of the endwall at a signi®cant rate.Due to the highly rotational character of the ¯ow near the convex surface, the transport mechanism is not limited to near endwall region.
The distinction between the molecular and turbulent diusion process can be made by computationally evaluating the individual terms in the thermal energy equation.The turbulent diusion process removes most of the thermal energy that is released from the endwall surface.The highest levels of turbulent heat transfer may appear near the passage vortex core, in addition to the near endwall region.
Most of the heat diusion process in the ¯ow passage was in the Y ± Z plane, not in the streamwise direction.
A successful interpretation of local heat transfer phenomena in¯uenced by a highly three-dimensional vortical ¯ow requires an extensive evaluation of the local turbulent ¯ow characteristics.
The current numerical approach may not be as accurate as the experimental heat transfer measurements.However, many details of the turbulent heat transport process that can not be measured directly by experimental methods can be visualized by the speci®c computational method.
Observations of the in¯uence of a strongly threedimensional ¯ow on the near wall ¯ow/heat transfer structure are possible via numerical methods.A comparison of the molecular/turbulent heat diusion in selected ¯ow cross sections, surface distributions of wall shear stress, visualization of local turbulent activity are a few of the tasks that can be performed using a numerical approach in a relatively time ecient manner.
High-resolution experimental ¯ow/heat transfer data presented in this paper form a high quality baseline set for future assessments of three-dimensional viscous ¯ow and heat transfer solvers that may originate from new computational strategies and turbulence models.

FIGURE 1
FIGURE 1 90 turning duct geometry, coordinate system, and electrically heated heat transfer surface.

Figure 4 FIGURE 3
Figure4shows (a) calculated and (b) measured endwall convective heat transfer coecient h, that is determined according to, h q 00 conv T w ÿ T 1

FIGURE 4b
FIGURE 4b Measured end-wall heat transfer coecient h in [W/m 2 K] (including measured secondary ¯ow patterns).
FIGURE 5a yp of the ®rst grid point away from top wall.
FIGURE 5b y p of the second grid point away from top wall.
FIGURE 6c TKE at the third grid point away from the end-wall.

FIGURE 7b
FIGURE 7bVorticity at the second grid point away from the end-wall.

FIGURE 7c
FIGURE 7cVorticity at the third grid point away from the end-wall.

FIGURE 8a
FIGURE 8aTemperature distribution at X 0 .

FIGURE 8b
FIGURE 8bTemperature distribution at X 45 .
FIGURE 8cTemperature distribution at X 90 .
molecular and turbulent diusion process can be made by numerically evaluating the individual terms in thermal energy equation.A comparison of the two components suggests that most of the removal of thermal energy released from the endwall surface is achieved by the turbulent diusion process.Strong turbulent heat ¯ux vector components near the endwall as shown in Figure9bare also apparent in the region where passage vortex dominates.The turbulent heat ¯ux vectors become increasingly larger when one moves away from the core

FIGURE 9b
FIGURE 9bTurbulent heat ¯ux vector at X 90 (only cross-stream components).

FIGURE 9c
FIGURE 9cMagnitude of turbulent heat ¯ux vector at X 90 (only cross-stream components).

FIGURE
FIGURE 9dTotal (molecular and turbulent) heat ¯ux at X 90 (only streamwise component).