report-pierrot.pdf

download report-pierrot.pdf

of 13

Transcript of report-pierrot.pdf

  • 8/13/2019 report-pierrot.pdf

    1/13

    The Spalart Allmaras turbulence model

    The main equation

    The Spallart Allmaras turbulence model is a one equation model designed especially foraerospace applications; it solves a modelled transport equation for kinematic eddy viscosity

    without calculating the length scale related to the shear layer thickness. The variable

    transported in the Spalart Allmaras model is which is assimilated, in the regions which are

    not affected by strong viscous effects such as the near wall region, to the turbulent kinematic

    viscosity. This equation has four versions, the simplest one is only applicable to free shear flows

    and the most complicated, which is written below, can treat turbulent flow past a body with

    laminar regions.

    This transport equation bring together the turbulent viscosity production term and the

    destruction term . The physics behind the destruction of turbulence occurs in the near wall

    region, where viscous damping and wall blocking effects are dominants. The other terms or

    factors are constants calibrated for each physical effect which needs to be modelled. This

    equation allows to determinate for the computation of the turbulent viscosity , which is

    for interest for us, from:

    The production terms

    In the transport equation for kinematic eddy viscosity, the production term is modelled in this

    way:

    Where

    is a scalar measure of the deformation tensor. During the development of the formula, was

    thought to depend only on the vorticity magnitude and was expressed in this way:

  • 8/13/2019 report-pierrot.pdf

    2/13

    Where is the mean rate-of-rotation tensor and is defined by

    In Fluent this formulation is used when the option Vorticity based is selected in the turbulence

    model definition box.

    Nowadays it is known that it is necessary to take into account the effect of the mean strain on the

    turbulence production and has been modify by J. Dacles-Mariani, G. G. Zilliac, J. S. Chow, and

    P. Bradshaw and incorporated to FLUENT:

    Where

    With the mean strain rate, , defined as

    Those formulations are used in FLUENT when the Strain Vorticity based option is selected in the

    turbulence model definition box. Its effect is to reduce the eddy viscosity in region where

    measure of vorticity exceeds that of strain rate.

    Destruction terms

    As it has already been pointed out, the destruction terms are only active in the region where

    shear is present and hence where viscosity effects are strong. The destruction terms are

    modelled in this way:

  • 8/13/2019 report-pierrot.pdf

    3/13

    The constants of the models are , and . There values have

    been calibrated and can be summarized as follow:

    0.1355 0.622 2/3 7.1 0.3 2 0.4187

    Boundary conditions

    The boundary conditions have been set and tested to match theory and experiments with a good

    convergence of the code. The wall condition is , it have been tested and it results that the

    turbulence viscosity term begins at the transition trips and spreads genteelly. The ideal value for

    the turbulence viscosity in the free stream is zero but some codes might have some difficulties toconverge because of round-off errors and it is often used:

    This is in the case of a calculation initialize with the trip term but setting up in the free

    stream allows a fully turbulent behaviour in any region where shear is present.

  • 8/13/2019 report-pierrot.pdf

    4/13

    Convergence

    The solution has achieved 26000 iterations for both cases, the Roe and the AUSM numerical method.

    The baseline model has been set in first order with a low courant number of about 0.25 until it has

    converged and then it has been implemented to a second order scheme and to the third order

    scheme. The convergence strategy for rotating applications advised in the FLUENT users manual has

    been used. The rotational speed has then been slowly increased until the case speed was reached.

    The implementation to a higher order scheme has been done at 13000 and 18000 iterations

    respectively. Unfortunately the residual plot is not available because it has not been saved but the

    residuals have converged to values of 0.001 and 0.01 before stabilising. The above graph show a

    comparison of the symmetric planes in terms of static pressure and it is possible to see that the

    solution is very similar, however it is not identical and the solution might not be converged enough.

  • 8/13/2019 report-pierrot.pdf

    5/13

    Results

    Pressure distribution and skin friction coefficient

    A comparison of the pressure coefficient on the five sections of the blades located respectively at 50,

    68, 80, 88 and 96 percent of the span and for a blade tip Mach number of M=0.827 has been carried

    out. This comparison investigates the effects of two numerical methods, the Roe and the AUSM

    method, both in third order MUSCL.

    The calculations yield results in accordance with the experimental data, even though the suction

    peak is not accurately captured and is slightly under-estimated for both solutions. As moving toward

    the tip the local velocity of the flow increase and the section peak increases until the flow over the

    blade becomes transonic. A shock forms at 89% of the span and is located at 23% of the chord. The

    AUSM numerical method tends to predict with more accuracy the shock position. However the

    prediction for the last section seems not to predict the shock observed in the experiment, the

    solution calculated is more diffusive.

    From the skin friction plots it is possible to analyse the shock boundary layer interaction and see if

    any separation is induced. No experimental data is provided to compare so only the Roe and AUSM

    data are plotted. No separation seems to be predicted for both method but it is possible to see that

    the skin friction coefficient is very low, about 0.002, for all section from 60% of the chord to the

    trailing edge. This illustrates the fact that the flow is at the edge of separation. At the shock location

    the skin friction decreases meaning that the velocity gradient is lower in the boundary layer and that

    there is an increase of the boundary layer thickness after the sock. The numerical method which

    seems to predict separation the more accurately is the AUSM method.

    Table 1: Lift coefficient comparison for the five sections with the Roe and AUSM numerical method

    Section location

    on the span (%)

    Cl for

    AUSM

    Errors (%) Cl for Roe Errors (%) Cl experimental

    50 0,216 8,94 0,19 19,90 0,2372

    68 0,2789 0,50 0,2767 1,28 0,2803

    80 0,3074 -9,32 0,3307 -17,60 0,281289 0,3157 -4,95 0,339 -12,70 0,3008

    96 0,2937 8,05 0,2881 9,80 0,3194

    The above table compare the lift coefficient from the experimental data with the calculation done

    with the Roe and AUSM numerical method. The lift coefficient for those two method have been

    calculated with a trapezoidal rule inducing great errors (errors for the trapezoidal rules can be of

    about 60%) and have to be analysed with care. However the lift coefficients predicted by the AUSM

    method are much more accurate with errors of about 9% for the worst case whereas the Roe

    calculation yields errors of 20%. Those errors can come from the mist prediction of the pressure

    distribution which would be reflected in the lift coefficient calculation.

  • 8/13/2019 report-pierrot.pdf

    6/13

    Figure 1: Y+ distribution for the five sections of the blade

    The differences between the CFD calculation and the experimental data could be due to the grid

    resolution. Indeed a analysis of the Y+ distribution on the five section of the blade show that a large

    portion of the blade has Y+ comprised in-between 30 and 5. It is known that a Y+ located in this

    region is not the optimum configuration of the grid for accurate results because the turbulent models

    are then not resolving the entire boundary layer, hence mist calculate the velocity distribution within

    the boundary layer. This affects the pressure distribution, and the skin friction distribution which are

    dependent on the velocity gradients in the boundary layer. This would explain why the recovery in

    pressure at the leading edge is not very well predicted; indeed the displacement thickness of the

    boundary layer must be under-estimated affecting the actual camber of the airfoil. For better results

    the grid must be modify so the centroid of the near wall cells does not lay in the region of 5

  • 8/13/2019 report-pierrot.pdf

    7/13

    Figure 2: Cp distribution for the section situated at 50% of

    the span

    Figure 3: Cp distribution for the section situated at 68% of

    the span

    Figure 4: Cp distribution for the section situated at 80% of

    the span

    Figure 5: Cp distribution for the section situated at 89% of

    the span

    Figure 6: Cp distribution for the section situated at 96% of the span

    x/c

    Cp

    0 0.2 0.4 0.6 0.8 1

    -0.5

    0

    0.5

    1

    1.5

    Experimental dataAUSM data

    Roe data

    x/c

    Cp

    0 0.2 0.4 0.6 0.8 1

    -0.5

    0

    0.5

    1

    1.5

    Experimental dataAUSM data

    Roe data

    x/c

    Cp

    0 0.2 0.4 0.6 0.8 1

    -1

    -0.5

    0

    0.5

    1

    1.5

    Experimental dataAUSM data

    Roe data

    x/c

    Cp

    0 0.2 0.4 0.6 0.8 1

    -1

    -0.5

    0

    0.5

    1

    Experimental dataAUSM data

    Roe data

    x/c

    Cp

    0 0.2 0.4 0.6 0.8 1

    -1

    -0.5

    0

    0.5

    1

    Experimental data

    AUSM data

    Roe data

  • 8/13/2019 report-pierrot.pdf

    8/13

    Figure 7: Skin friction coefficient for the section located at

    50% of the span

    Figure 8: Skin friction coefficient for the section located at

    68% of the span

    Figure 9: Skin friction coefficient for the section located at

    80% of the span

    Figure 10: Skin friction coefficient for the section located at

    89% of the span

    Figure 11: Skin friction coefficient for the section located at 96% of the span

    x/c

    Skin

    friction

    coeffic

    ient

    0 0.2 0.4 0.6 0.8 1

    0.002

    0.004

    0.006

    0.008

    0.01

    0.012

    Roe

    AUSM

    x/c

    Skin

    friction

    coeffic

    ient

    0 0.2 0.4 0.6 0.8 1

    0.002

    0.004

    0.006

    0.008

    0.01 Roe

    AUSM

    x/c

    Skin

    friction

    coefficient

    0 0.2 0.4 0.6 0.8 1

    0.002

    0.004

    0.006

    0.008

    0.01 Roe

    AUSM

    x/c

    Skin

    friction

    coefficient

    0 0.2 0.4 0.6 0.8 1

    0.002

    0.004

    0.006

    0.008

    0.01

    Roe

    AUSM

    x/c

    Skin

    friction

    coefficient

    0 0.2 0.4 0.6 0.8 1

    0.002

    0.004

    0.006

    0.008

    Roe

    AUSM

  • 8/13/2019 report-pierrot.pdf

    9/13

    Vortex analysis

    The above figure present the visualizations of the computed wake structure using iso-surfaces ofvorticity, for the case with a tip Mach number of 0.827. It is possible to see the wake generated from

    the blades and the vortex generated at the tip. The solution has considerable noise because it is not

    converged enough yet, even though the tip vortex is resolved until 120. However, the wake of the

    blade is not yet well resolved and the calculation must be carried on for more iteration to be able to

    capture it with accuracy.

    Figure 12: Vorticity magnitude visualisation for a rotational speed of 2350 rpm

    X

    Y

    Z

  • 8/13/2019 report-pierrot.pdf

    10/13

    The above figures show the vortex shading for different ages going from 0 to 80 with a step of 10

    for the Roe and the AUSM numerical method calculation. At an age of 0 it is possible to see the

    vortex been created at the tip of the blade. As the vortex becomes older it grows, becoming less

    strong and migrates down toward the root of the blade. Both numerical method yield similar results

    and a deeper analysis is done in Figure 31 which plots the age of the vortex versus its Z and Y

    direction non dimensionalised by the radius of the blade R=1.142m.

    Vortex shading for the Roe calculation:

    Figure 13: Tip vortex shading for 0 Figure 14: Tip vortex shading for -10 Figure 15: Tip vortex shading for -20

    Figure 16: Tip vortex shading for -30Figure 17: Tip vortex shading for -40

    Figure 18: Tip vortex shading for -50

    Figure 19: Tip vortex shading for -60Figure 20: Tip vortex shading for -70

    Figure 21: Tip vortex shading for -80

  • 8/13/2019 report-pierrot.pdf

    11/13

    Vortex shading for the AUSM calculation:

    Figure 22: Tip vortex shading for 0 Figure 23: Tip vortex shading for -10 Figure 24: Tip vortex shading for -20

    Figure 25: Tip vortex shading for -30Figure 26: Tip vortex shading for -40

    Figure 27: Tip vortex shading for -50

    Figure 28: Tip vortex shading for -60Figure 29: Tip vortex shading for -70

    Figure 30: Tip vortex shading for -80

  • 8/13/2019 report-pierrot.pdf

    12/13

    Figure 31: Wake geometry measurements for a rotor speed of 2350 rpm and comparison with classical data

    A comparison with the data from F. X. Caradonna and C. Tung, 1981 is carried out for both solution

    calculated with the Roe and the AUSM method. Vortex ages from 0 to 120 have been plotted sincethe solution was not fully converged to predict the vortex location at a further age. The results are

    not very satisfactory and diverge for high vortex ages. This might be due to a convergence problem

    and the solution might be run for several more iteration in order to predict the vortex migration with

    more accuracy. However, the AUSM solution is much closer to the experimental data and seems to

    predict the vortex migration with more accuracy.

  • 8/13/2019 report-pierrot.pdf

    13/13

    Conclusion

    This report has focused on the simulation of hovering rotor tip vortices and rotor wake convection

    using the Spalart and Allmaras one equation turbulence model for two numerical methods, the Roe

    and the AUSM method. The vortex sheet is a relatively weak feature of the flow that descends in a

    tightening helical pattern below the rotor. The root and tip vortices follow contracting helical

    trajectories below the rotor disc. This behaviour has been observed for both calculation carried out

    but the AUSM model tend to be closer to the experimental solution. The tip vortices and the wake

    influence strongly the pressure distribution of the blades in a hovering rotor generating vibration and

    noise. The result of this analysis tend to show that the results for the pressure distribution are in

    accordance with the experimental data but that the resolution of the mesh is of importance and that

    further calculations must be carried out with a better grid resolution for a better accuracy of the

    results. For a better capture of the vortex trajectory and wake calculation the actual solution is not

    converged enough and it is a critical parameter to analyse the vortex migration for advanced ages. In

    any case, the numerical method which seems to predict with the more accuracy this type of

    problems is the AUSM numerical method but a particular care must be taken toward the Y+

    distribution on the wing to avoid near wall cells comprised in the region of 5