See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/267569134

PADRAM: Parametric Design and Rapid Meshing System for Turbomachinery

Optimisation

Conference Paper · January 2003

DOI: 10.1115/GT2003-38698

CITATIONS READS
144 6,184

2 authors:

Shahrokh Shahpar Leigh Lapworth

Rolls-Royce Rolls-Royce
170 PUBLICATIONS   1,603 CITATIONS    34 PUBLICATIONS   533 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Surrogate modelling techniques for large-scale optimisation and applications in aerospace engineering View project

AMEDEO View project

All content following this page was uploaded by Shahrokh Shahpar on 07 November 2014.

The user has requested enhancement of the downloaded file.

T TOC
Proceedings of ASME Turbo Expo 2003
Proceedings of IGTI:
Power for Land, Sea, and Air
ASME Turbo Expo
June 16–19, 2003, Atlanta, Georgia, USA
June 16-19, 2003 – Atlanta, Georgia.

GT2003-38698
GT2003-

PADRAM: PARAMETRIC DESIGN AND RAPID MESHING SYSTEM FOR

TURBOMACHINERY OPTIMISATION
Shahrokh Shahpar— and Leigh Lapworth˜
Aerothermal Methods,
Rolls-Royce plc
Derby, U.K.

ABSTRACT
A parametric design system suitable for inclusion in an INTRODUCTION
automatic optimization process is presented. The system makes As Computational Fluid Dynamics (CFD) codes have
use of a multi-block structured grid generation system specially steadily evolved into everyday analysis tools, so attention is
designed for the rapid meshing of two-dimensional, quasi- now focused on integrating the analysis codes with design tools
three-dimensional, and three-dimensional single passage as paving the way to carry-out automatic optimization. Automatic
well as multi-passage, multi-row turbomachinery blades. Full optimization tools and methodologies have proved able to
annulus viscous meshes of the order of five to ten million mesh significantly reduce the manual design time and simultaneously
points for the complete bypass assembly of the low pressure improve the quality of the designs, e.g. see references [1] and
compression (LPC) system can be generated in a matter of [2]. Traditionally the CFD pre-processing tools are embedded
minutes. PADRAM offers a major new design capability where within a Graphical User Interface (GUI) environment. The
the optimisation of multi-passage three-dimensional blades and user-friendly environment which a GUI provides to engineers
its circumferential pattern is done simultaneously in one is deemed a necessary and essential part of the CFD process.
system. Successful usage of PADRAM in a number of design, However, automatic optimization strategies require a
optimisation and analysis applications has recently been parametric definition of the geometry and the transfer of
demonstrated and reported herein. geometry from CAD or a database, such as blade definition
file, to the mesh generator to be seamless with no user
NOMENCLATURE intervention. GUIs must be scriptable, otherwise they become
a, b, c elliptic partial-differential-equation coefficient bottlenecks in the optimization process. In the design by
m’ non-dimensional streamwise coordinate analysis approach, spending a few days to set up a new
r, θ, z physical, polar co-ordinates: geometry interactively and spending a couple of hours, using a
radial, circumferential and axial direction very good mesh generator to produce a suitable computational
P, Q elliptic grid generator forcing functions grid has been regarded as reasonable. However, this is totally
y normal to the wall distance unacceptable for inclusion in an optimisation loop, as the
1/ 2 design process cannot be automated and the overall
y+ non-dimensional wall distance, = y τ w  optimisation time would be too long to be of use to the
ν 2 ρ 
  designer.
x, y, z Cartesian coordinates The grid generator presented in this paper arose not only
α, β clustering function coefficient from the need to perform rapid mesh generation within an
ξ, η coordinates in the computational plane optimisation scheme; but, also the need to mesh rings of vanes
θ azimuthal angle where each vane can potentially be different from all the others.
ν kinematic viscosity Figure 1 shows an engine cut-away along with a CFD
ρ density representation of the bypass duct. Downstream of the Bypass
τw wall shear stress Outlet Guide Vanes (OGVs) there are the engine pylon and
Radial Drive Faring. These create large asymmetric pressure
— Specialist – Aerothermal Design, C.Eng., MAIAA, MRAerSoc. distortions which propagate upstream through the OGVs and
˜ Specialist – Numerical Methods, C.Math FIMA interact with the fan rotor. The level of pressure distortion has
a direct influence on the forced response levels of the fan rotor.

1 Copyright © 2003 by ASME

 (a) Engine cut away with bypass OGVs highlighted (b) CFD solution for bypass duct (static pressure)
Figure 1. Bypass duct components, Outlet Guide Vane (OGV), Pylon and Radial Drive Faring (RDF).

By tailoring the OGVs (i.e. making them non-circumferentially structured [3-4], unstructured [5-6], overset or the so-called
uniform) they can be used to attenuate the pressure distortion Chimera [7-9], hybrid [10-11] and Cartesian grids [12]. In the
and reduce/eliminate fan forcing. If the tailoring of the OGV following discussion, we concentrate on mesh generation for
geometry is to be performed by an optimisation system (see turbomachinery blades. A more detailed discussion of various
Shahpar et al., [2]), particularly one that uses heuristic grid generation techniques is beyond the scope of this paper,
optimisers, then there can be strong variations in the geometry interested readers should refer to a number of good books
of adjacent OGVs during the design cycle. published in this field, e.g. Thompson et al. [13]. Among the
Grid generation for turbomachinery is relatively well grid generation techniques, the block-structured is the most
evolved for generating simple blade-to-blade mesh sections that powerful and is the most established, in particular for external
are then stacked radially to create a 3D mesh. The typical types aerodynamics CFD. Complex geometries can either by grided
of meshes used in the field of turbomachinery CFD are block by unstructured grids or by structured multi-block grids.

(b) Leading edge detail

(b) Leading edge detail

(a) Blade to blade mesh (a) Blade to blade mesh

Figure 2. Sheared H style mesh for a compressor blade, with leading Figure 3. Letter-box style mesh for a compressor blade, with leading
edge detail. edge detail.

2 Copyright © 2003 by ASME

 (b) Leading edge detail

Blade to blade mesh (a) Blade to blade mesh

Figure 4. C-type mesh for a Turbine blade. Figure 5. H-O-H grid for a compressor blade with leading edge detail
of the O-mesh.
Unstructured grids typically have memory and CPU overheads An unstructured mesh topology [18] is shown in Figure 6.
due to the need store mesh connectivity data, but offer the This has a structured O-mesh around the vane and an
greatest geometrical flexibility. On-the-other-hand multi-block unstructured mesh of triangles in the freestream and wake
structured grids are very efficient and can fill-up topologically regions. The triangles become prisms when the mesh is
complex domain by decomposing the geometry into simple stacked in 3D. This style of mesh can easily accommodate a
blocks. Multi-block meshes have successfully been used in the wide range of perturbations to the OGV geometry around the
context of MDO (Multi-Disciplinary Optimization) sensitivity annulus. The difficulty of this approach is that to preserve the
study [14]. prismatic topology of the unstructured region, the blade-to-
The single–block structured H-mesh topology is illustrated blade mesh must be generated on a single section and then
in figures 2 and 3. The sheared H-mesh [15] shown in figure 2 cloned to the other sections. This can lead to significant
leads to a very fast CFD solver because the solver exploits the shearing of the prisms if the optimiser generates strongly three
fact that the circumferential mesh lines have a constant axial dimensional blades. When generating new meshes not only the
co-ordinate. This approach is clearly too restrictive for the topology but also the grid density needs to be kept as consistent
current work where the axial position of each OGV may vary as possible to the original datum geometry in order to minimise
circumferentially around the annulus. The letter-box H-mesh
[16] shown in figure 3 uses a curvilinear mesh topology and is
more-flexible than the sheared H-mesh, but the high aspect
ratio cells in the cross-passage direction at the leading and
trailing edge would also restrict the geometric flexibility of the
optimiser. A single–block structured C-mesh topology is
shown in figure 4. For single passage meshes this has good
resolution of the leading and trailing edges and also the vane
wake; but, has a mesh singularity at the junction of the
upstream and periodic boundaries. The C-mesh has more
flexibility for movement of the OGVs than the H-meshes, but is
difficult to interface with the downstream meshes for the pylon
and RDF. The multi-block structured H-O-H mesh topology (b) Leading edge detail
[17] is shown in figure 5. Here, the flexibility to vary the axial
position of the OGVs around the annulus is limited by the
requirement to have degenerate nodes at the junction of the O-
and H-meshes with the periodic boundaries. The periodicity (a) Blade to blade mesh
means one O-block cannot be moved independently of its
neighbour. Figure 6. Hybrid-mesh for a compressor blade with
leading edge detail.

3 Copyright © 2003 by ASME

 (a) Optimum turbine nozzle guide vane produced (b) FAITH non-axisymetric endwall design and mesh perturbation
by the FAITH inverse design system [1]. capability – Contours of radial mesh movement [21].
Figure 7. Application of mesh perturbation within design.

the errors in computing the flow sensitivities. This is define default mesh generation parameters that are then fixed
particularly difficult to control when using unstructured grid throughout the optimisation process. Tschirner et al. [20] have
generators as the total number of grid points cannot be fixed a recently reported viscous 2D analysis of 31 OGVs, 6 fairings
priori. and a pylon. They have reported an automated procedure using
Fully unstructured meshes are too time consuming to be the FLUENT grid generator GAMBIT and several UNIX
generated within the optimisation cycle, if only because they scripts. The manual grid generations necessary to perform the
are working in 3D and not in a stacked 2D mode. design and analysis with GAMBIT was reported to be
Chimera or overset meshes [7-9] are a flexible way to extremely time consuming. Hence, a combination of several
apply structured mesh solution techniques to geometrically UNIX shell scripts and libraries were used in conjunction with
complicated multi-body configurations. Novel applications an interactive user interface for data input by the user. Re-
have been reported for dynamic systems like store release or staggering and re-blading of the OGVs were done in a
helicopter rotor flow simulation. Overset meshes can easily different, in-house system that designs blades in a single-
accommodate large variations in the OGV geometry around the passage environment.
annulus, but require care with interpolation schemes in the If mesh generation is time consuming, then an alternative
overlap regions. Wadia et al. [19] reported that the Chimera approach is to use a mesh movement algorithm to morph the
grid generation process was extremely complicated for the mesh onto each new geometry generated by the optimiser.
analysis of their 16 vanes and 1 strut configuration. The Figures 7a and 7b show a successful application mesh
Chimera technique was used to create a mesh for the sixteen perturbation techniques to both aerofoil [1] and endwall design
variably-staggered OGVs, strut and splitter. Their process [21]. Although, this approach is quite popular, it is the authors’
started with untrimmed initial surface grids for each component
that was generated from CAD data using the ICEM CFD
program. A series of Fortran programs, shell scripts and
various grid tools were used to rotate components to their
respective positions, find surface-to-surface intersections, and
construct the final surface grids. Volume grids for the strut and
vanes were generated using a hyperbolic grid generator while
an algebraic grid generator was developed for the other
components. An additional program was used to perform the
Chimera hole-generation and determine the grid-to-grid
boundary interpolation coefficients. The final grid
configuration consisted of sixteen-vane grids overlaid on an
outer grid, a strut grid, and core and bypass section grids,
comprising of 2.7 million grid points.
The majority of optimisers begin with a CFD solution for
base or reference geometry. Since this is upstream of the Figure 8. Erroneous geometry and mesh generated by
optimisation cycle, more time can be taken in generating the perturbing the datum 3D BOGV in a ring of 52 OGVs.
base mesh. Indeed, time is generally taken at this stage to

4 Copyright © 2003 by ASME

experience that producing good quality viscous meshes in rs = r (m, θ ),
three-dimensions for complicated geometry is very difficult and
θ s = θ (m, θ ),
time consuming. This is particularly true when the geometry is
changing significantly from its datum shape for example during z s = z (m, θ ). (1)
a heuristic optimisation run. It has been found [22] that the Where,
approach of changing the geometry gradually and perturbing z
dr 2 + dz 2
the meshes can lead to erroneous geometry (see figure 8) and
meshes (if geometry changes abruptly); and, also can take up to
m' = ∫
z0
r ( z)
,
ten hours of CPU on a compute server which clearly makes the
strategy of the mesh perturbation impractical to include in an
r = r ( z ). (2)
automatic optimisation cycle. Where z0 is an arbitrary reference and m’ is a non-
In this paper, a unique design system is presented that dimensional distance along a stream section which will be zero
allows the design of multi-passage 3D blades and their at z0.
circumferential pattern to be carried out simultaneously in one The starting point for PADRAM system is to transform the
system. A novel method is presented that allows consistent radial stream sections into parametric two-dimensional planes,
good quality viscous meshes to be generated very rapidly for using the co-ordinates θ and m’. As r is greater than zero for
multiple passages of three-dimensional blades requiring no any z coordinates, m’ is a monotonic function of z, hence a
more than a few minutes of CPU time. Since the mesh unique inverse function exists to map the computational
generation is extremely fast and occur in real time, the users coordinates back to the physical, three-dimensional polar
can tune the default parameters that control the quality of the coordinates. The advantage of the above transformation is that
3D mesh upstream of the optimisation runs interactively and the angles are preserved and the mesh-generation procedure
easily. This system is called PADRAM (Parametric Design deals with plane sections only.
and Rapid Meshing System).
PADRAM can automatically and parametrically change the Grid generation
2D, quasi-3D and 3D blade geometry and produce very rapidly PADRAM grid generation starts by dividing the
good quality viscous mesh for the multi-passage, multi-stage computational blocks into sub-blocks for the purpose of
turbomachinery passages. PADRAM makes uses of both generation of the algebraic grid and the control functions. O-
transfinite interpolation and elliptic grid generators to generate type grid is used for the blades and H-type grids near the
hybrid C-O-H meshes. An orthogonal body-fitted O mesh is periodic boundary, upstream and downstream blocks, C-type
used to capture the viscous region in the vicinity of the aerofoil grids are used for semi-infinite boundaries such as the splitter,
whilst an H mesh is used near the periodic boundaries and pylon and RDF (Radial Drive Fairing). Transfinite
where stretched cells are required, for example in the wake. interpolation (TFI) [13] is used to generate the initial grid based
The H-meshes also take up the slack from the relative on a linear interpolation of the specified boundaries. TFI is
movement in adjacent O-meshes as the OGVs are being re- very easy to program, computationally efficient and the grid
designed. The mesh is independently generated for every spacing is under direct control. PADRAM uses the following
stream section, hence three-dimensional meshes are produced double clustering functions:
η −α
easily from stacked two-dimensional blade section meshes with
(2α + β ) β + 1 + 2α − β
1−α
no mapping required to transfer the meshes radially. This
avoids the mesh morphing, and ensures that good quality y=  β − 1 ,
 1−α 
η −α
meshes are created at every height even if the geometry is
changing considerably from hub to tip. Mesh generation is (2α + 1)1 +  (β + 1) 
extremely fast, requiring no more than a few seconds to   (β − 1)   (3)
generate 2D meshes for the complete bypass OGV or a single  
passage 3D mesh; and, a few minutes to generate a mesh of the 1 < β < ∞,
order of 6 million points for the complete bypass assembly (on 0 ≤ α ≤ 1.
Sun Ultra 10 workstations). Where η is the non-dimensional grid point distribution in
the computational plane which varies between zero and 1, β is
METHODOLOGY
the clustering function, more clustering is achieved by letting β
Geometry Modelling
to approach 1 and α is a non-dimensional quantity that
Turbomachinery blade geometry is defined at a number of
indicates which grid location the clustering should be attracted
radial stream sections These radial sections lie on a three-
to, e.g. a value of 0.5 ensures clustering is uniformly done at
dimensional surfaces defined in polar coordinates as Si(r, θ, z),
both ends.
where i is the section index. Using non-dimensional parametric
PADRAM solves a system of elliptic partial differential
coordinates, m’ and θ, a typical surface can be defined as: equations (PDEs) to smooth the algebraically generated grids in

5 Copyright © 2003 by ASME

 Figure 9. PADRAM mesh for a compressor blade H-O-H
topology.
Figure 11. PADRAM’s multi-passage tip clearance model
each block. For each radial height, the following equations are (5% gap for demonstration).
solved:
amξξ − 2bmξη + cmηη = P (ξ ,η ), Tip Gaps
Tip clearance is often required for shroudless rotor blades
aθξξ − 2bθξη + cθηη = Q(ξ ,η ).
such as fans, compressors and some turbines as well as hub-
Where, (4)
clearance for cantilevered stator blades. Tip leakage flows are
a = m +θ ,
2
η
2
η often poorly resolved. If the number of points in the tip
b = mξ mη + θ ξθη , clearance is too low, then the complex physical phenomena that
occur there cannot be accurately modelled. It has been found
c = mξ2 + θ ξ2 . that the flow solution can be sensitive how the gap is modelled
P and Q are the forcing functions to ensure orthogonality and in some solvers the geometry is changed to alleviate the
and grid clustering at the surfaces. The grids presented in this problem, e.g. sheared H-meshes often use the so called
paper were generated with the forcing functions switched off “pinched” tip gap shown in figure 10. However, PADRAM
which saves significant computational time (typically 50%). models the gap in a consistent way that meshes the blade
However, the grid orthogonality near the surfaces was ensured geometry. Once the O-grid is generated with the outer domain
by using a hyperbolic-type grid generation to generate the O- of the blade, the grid corresponding to the solid part of the
grids, that is using a marching procedure to grow the grid domain is constructed using the same boundary node
normal to the boundaries. Equation 4 can be solved by an distribution. It is important to keep the mesh spacing in the
iterative solver such as the Gauss-Siedel scheme or a point inner and outer part of the wall as close as possible to each
successive over relaxation scheme (PSO). PADRAM makes use other. Figure 11 shows a typical tip gap mesh for a compressor
of the latter scheme to elliptically smooth the TFI generated rotor generated in PADRAM (the tip gap has been increased to
grids. 5% span for demonstration). The work of Van Zante [23] has
Figure 9 shows the PADRAM mesh topology for a single
OGV mesh. Note that the upstream and downstream H-mesh
can also be rotated to align the mesh with the blade inlet and
exit metal angles.

(a) Sheared H-Mesh (b) Hybrid unstructured mesh

Figure 10. Tip clearance models.
Figure 12. Single passage PADRAM mesh for the swept back
OGV.

6 Copyright © 2003 by ASME

shown the importance of extra mesh clustering close to the
shroud in meshes with a tip gap. This can be achieved in
PADRAM using specified clustering parameters.

LP COMPRESSION DESIGN SYSTEM

Design Considerations
In order to reduce or eliminate the effects of downstream
struts or pylons, the bypass outlet guide vanes (BOGVs) can be
reconfigured, e.g. circumferentially restaggered or recambered.
A detailed review of various design methods is reported in
reference [2] and is not repeated here. However, most of the
reported analyses in the literature are 2D or 3D single passage
calculations. Long computational time and generating suitable
CFD meshes for the complex three-dimensional full annulus
geometry are two main factors deterring engineers to carry out
routine analysis of reconfigured BOGVs.
PADRAM provides a unique design system that allows the
Figure 13. PADRAM mesh for the 52 bypass OGVs, pylon, RDF,
design of multi-passage 3D blades and their circumferential
and splitter consisting of 119 blocks.
pattern to be carried out simultaneously in one system.
the OGVs.
Bypass Duct Meshing PADRAM can construct a viscous mesh for the complete
One key feature of PADRAM is the fact that it is intimately bypass assembly consisting of 52 different staggered and three
linked with the geometry definition. Within the ring of the types over and under cambered OGVS, pylon, RDF and a
BOGVs the geometry of each OGV can be specified splitter in a matter of minutes. The details of the mesh near the
independently of the others either by using a different geometry OGV, Pylon, RDF and the splitter are shown in figures 12 to
definition file or by using one of the several design options 15. Figure 1b shows a typical CFD solution.
within PADRAM (see below). Hence, PADRAM is easily able
to produce meshes for variable stagger OGVs with a fixed base Design/Optimisation System
geometry; or, to design for sophisticated OGV patterns. In order to perform design iterations, PADRAM requires
Another important feature of PADRAM is that it is a true multi- the reference geometry for the OGVs, pylon/RDF and
passage meshing system. The geometry of each OGV in the endwalls. The reference OGV geometry may include one or
annulus can be varied independently and a good viscous mesh more distinct blade shapes in a prescribed circumferential
for the complete ring of OGVs is generated in one step. This is pattern. Each OGV is then assigned a set of design parameters
both more efficient and more robust than the practice of (see next section) which can be varied as a function of %span.
generating a symmetric ring of identical OGVs and then Hence, the design system is able to, simultaneously, construct
morphing the mesh to match the actual asymmetric geometry of new 3D blade shapes and new circumferential patterns.

RDF

Pylon

Figure 14. Detail of the C-mesh near Figure 15. Detail of the splitter C-mesh, hub mesh and the
the Pylon and RDF – top View. engine-core exit mesh.

7 Copyright © 2003 by ASME

 The mesh generation parameters are tuned for the
reference geometry and then held fixed for the design
iterations. Since, the mesh generation parameters are
parametrically related to the blade geometry, this system
ensures consistent meshes are generated even if the geometry
changes considerably during the optimisation.

Parametric Design Space

The PADRAM design space includes standard

Figure 18. PADRAM mesh for a single stage fan with

rotor TE and stator LE detail near the hub.

turbomachinery design parameters such as lean (or pitch

variation in 2D multi-passage simulation), sweep (or axial
variation of sections), leading-edge and trailing-edge re-
camber. A sample of design parameters and PADRAM meshing
(a) (b) capability are shown in Figures 16 and 17. The mesh is
o
Figure 16. Differential 0 to 2 lean, (a) perturbed mesh with generated after the new leant geometry has been constructed,
Fixed periodic boundaries, (b) PADRAM mesh. this will produce a better mesh than the mesh perturbation
approach. Figure 16b shows a differential 0 to 2o leant
geometry and mesh generated by PADRAM for a bypass OGV.
Successful automatic design and optimisation of the
bypass OGVs in the presence of the Pylon and RDF using
PADRAM is reported in reference [2] and is therefore not
repeated here.

Remote Optimisation
Another feature of the PADRAM system is that it presents
a means of making use of remote computer resources.
PADRAM is written in ANSI ‘C’ for complete portability. The
mesh generation parameters can be defined locally, where the
ability to view the mesh is essential. The geometry files,
PADRAM input file and pre-processing scripts, which are only
a few Kbytes in size, can be sent electronically to the remote
machine. Since PADRAM is extremely fast, the mesh
generation and pre-processing can then be done remotely,
avoiding the need to send mesh files of many tens or hundreds
of megabytes to the remote machine via CD or DAT tape.
Salient details from the solution can also be extracted remotely
before deciding whether the full CFD solution should be
returned on CD/DAT tape to be examined locally.
Figure 17. An example of the PADRAM’s flexible design system to
vary an OGV’s lean in a row of four OGVs.

8 Copyright © 2003 by ASME

 Figure 19. PADRAM mesh for a 5 stage compressor, showing contours of static pressure.
the clustering parameters linearly interpolated between the hub,
FURTHER APPLICATIONS OF PADRAM mid and tip values) to get a fully satisfactory mesh. The final
Although PADRAM was initially developed for the mesh is shown in figure 18 and consists of 334k nodes in the
parametric design and rapid meshing of the LP compression rotor mesh and 212k nodes in the stator mesh. Figure 18 also
system, its robust mesh generator and its design space have shows the mesh detail at the hub in the vicinity of the mixing
been used on a number of other project applications. The direct plane. The very tight control over the O-meshes on the rotor
link between the geometry and mesh make it an ideal tool for and stator is clearly shown.
design studies whether those are based on an optimiser or on
design by analysis. Multi-Stage Research Compressor
PADRAM has also been used to generate meshes for
Single Stage Research Fan multistage compressors and turbines. Figure 19 shows the
PADRAM has also been used to set up a suitable mesh for CFD solution from a PADRAM mesh for a 5 stage research
a single stage research fan to demonstrate its ability to provide compressor. Of interest is the fact that the 2nd, 3rd and 4th stage
meshes for mixing and sliding plane analyses. This single stage stators are cantilevered and the PADRAM tip gap model has
fan is a very severe test for a mixing plane analysis because the been successfully applied to both stator hubs and rotor casings.
rotor and stator are closely coupled at the hub, but have a very
wide spacing at the tip. This represents problems both in terms Waterjet Impeller
of mesh generation and stability of the CFD calculation. Using Figure 20 shows a PADRAM mesh for a 6-bladed waterjet
typical civil fan default values for the PADRAM mesh impeller. This is a challenging case to mesh since there is a
generation parameter produced a rotor mesh where the outer considerable change in the geometry from hub to tip; and, the
boundary for the O-mesh extended past the exit boundary. The radius changes significantly from leading to trailing edge. The
capability in PADRAM to independently set the mesh geometry has been represented using the same style of blade
clustering parameters at hub and casing was crucial to getting geometry definition as used above and so all of the blade
an acceptable mesh for this test case. In fact, it was necessary design parameters in PADRAM are available for a redesign of
to add independent control of the clustering at mid-height (with this component.

(a) Perspective view (b) View from front (c) Leading edge detail
Figure 20. PADRAM mesh for a 6 bladed waterjet

9 Copyright © 2003 by ASME

MESH QUALITY
The need for the meshing system to be both rapid and
based on parametric geometry has been described above. For it
to be used successfully within an optimisation cycle it must
also be robust when presented with large geometry
perturbations. This is particularly likely to occur with heuristic
optimisers. The robustness of the mesh generator must also be
matched by the robustness of the CFD solver. For example, as
the optimiser tries to push up incidence angle, the CFD solver
must respond with a realistic loss loop in order that the
optimiser converges to a design with acceptable incidence. In
this section, two examples of the importance of mesh quality
(a) Letter-box mesh (~1.12M nodes)
within the design system are presented.

CFD Solver
The CFD solver used with the PADRAM system is the
Rolls-Royce plc. HYDRA. HYDRA is a suite of non-linear,
linear and adjoint CFD solvers developed collaboratively by
Rolls-Royce plc. and its University partners. HYDRA
represents the PADRAM meshes as hybrid unstructured meshes
using an efficient edge-based data structure [24]. Convergence
to steady state is accelerated through the use of an element-
collapsing multigrid algorithm [25]. The HYDRA solver has
been parallelised using the domain decomposition method and
runs efficiently on both shared and distributed memory
computers. The parallel multigrid approach is essential to (b) PADRAM mesh (~1.37M nodes)
generating CFD solutions in the elapsed time needed for
Figure 22. Letter-box and PADRAM meshes for
effective use within a design/optimisation system.
the bypass OGV noise calculation.
Leading Edge Separations
of the boundary-layer downstream. The PADRAM mesh gives
It is almost inevitable in an optimisation cycle that the
much more reasonable leading edge behaviour – due to the fact
sensitivity to incidence will be explored. Here, both the mesh
that the O-mesh around the blade does not contain corner
topology and CFD solution must respond physically to changes
points. The PADRAM mesh does contain singularities in the
in incidence. Figure 21 shows the Mach number contours near
mesh, where the O- and H-meshes meet, but these are away
the leading-edge of a high-pressure compressor stator. A
from the high gradient regions at the leading and trailing edges.
relatively coarse grid was used in this study. It has been found
Both meshes in figure 21 have a relatively coarse
that the boundary-layer is better represented by the PADRAM
resolution of the leading edge and it is likely that the
mesh than the letter-box style mesh. The corner grid points in
consistency of the solution of the 2 meshes will improve as the
the letter-box mesh, near the blade leading-edge, thicken the
mesh gets finer. However, the successful application of an
boundary-layer leading to excessive thickening and separation
optimiser is often a judicious choice between the level of mesh
resolution for rapid run times and that needed for good quality
CFD solutions. Care in needed to ensure good quality mesh is
used in the area of high-gradient flow regions.

Bypass OGV Noise Simulation

PADRAM has been used to generate suitable meshes for
the CFD analysis of a representative Fan/bypass OGV
interaction. The OGV noise generation calculation uses the
linearised unsteady HYDRA solver, with the incoming wake
amplitude and phase defined from a separate non-linear steady
HYDRA calculation. Initial calculations for the bypass OGV
(a) letter-box mesh (b) PADRAM mesh used a letter box style mesh. Figure 22a shows the letterbox
mesh that was sized to 1.12M to maintain the propagation of
Figure 21. HYDRA solution for a compressor blade – contours the expected acoustic waves at 2BPF (blade passing frequency)
of Mach number shown.

10 Copyright © 2003 by ASME

 Figure 24 compares the amplitude of the predicted mid-
span unsteady pressure in 2BPF for the two mesh styles and
illustrates the reductions obtained with the PADRAM mesh.
Using linear HYDRA and the PADRAM system the effects of
different numbers off (65 vs. 58) and different OGV geometries
are being studied.

CONCLUSIONS
A flexible, integrated multi-passage blading design system
is presented for inclusion in an automatic optimisation cycle
(a) Linear HYDRA solution on letter-box mesh. using high fidelity CFD codes. Examples are included in this
paper that illustrate the range of capabilities of the new
parametric design and rapid meshing system (PADRAM) to
modify blade’s geometry and configuration and to generate a
good quality viscous mesh very rapidly. PADRAM can be used
in both design and analysis of 2D, 3D, single-passage, multi-
(b) Linear HYDRA solution on PADRAM mesh. passage, full annulus, multi-stage environment.
Figure 23. Bypass OGV Instantaneous Axial Velocity
ACKNOWLEDGMENTS
Perturbation (-8m/s purple to +8m/s red) at 2BPF.
The authors would like to acknowledge the permission of
Rolls-Royce plc to publish this paper. This work was
OGV Surface Unsteady Pressure Amplitude at ~50% Span in 2BPF
performed under the UK DTI CARAD Programme (Contract
Linear HYDRA - PADRAM Mesh No. CKBB/C/016/00029). The bypass OGV noise simulation
170
Linear HYDRA - Letter Box Mesh was performed under the European Union 5th framework
160 project TurboNoiseCFD. The authors also wish to thank their
colleague Mr John Coupland for useful discussions.
150
Pressure SPL (dB)

140 REFERENCES
130
[1] Shahpar, S., 2001, “Three-dimensional Design and
120 Optimisation of Turbomachinery Blades using the Navier-
Stokes Equations”, ISABE 2001-1053, proceedings of 15th
110
ISABE conference, Bangalore, India.
100
-100 -80 -60 -40
Pressure Surface
-20 0 20
- % Axial Chord from LE
40 60
- Suction Surface
80 100 [2] Shahpar, S. Giacchi, D., and Lapworth, L., 2003, “Multi-
objective Design and Optimisation off Bypass Outlet-guide
Figure 24. Bypass OGV Unsteady Surface Pressure Vanes”, ASME paper GT2003-38700, Atlanta.
Response in 2BPF.
[3] Shaw, J.A. and Weatherill, N.P., 1992, “Automatic topology
and 3BPF in the linear unsteady calculation. With this mesh the generation for multiblock grids”, Applied Mathematics and
non-linear steady HYDRA calculation yielded a substantial
Computation, 53, pp 335-388.
pressure-surface flow separation. The problem was traced to
excessive thickening of the leading edge boundary layer due to
[4] Dannenhoffer, J.F., 1993, “A New method for creating grid
the distorted cells in that region of the letterbox mesh. This
abstractions for complex configurations”, AIAA paper 93-
was a case run at part shaft speed with significant negative
0428.
incidence onto the OGV. For the linear unsteady calculation
the separated flow led to substantial flow instability and very
[5] Lohner, R., and Parikh, P., 1988, “Generation of three-
high unsteady blade response to the incoming wake.
dimensional unstructured grids by the advancing front
Figure 22b shows the alternative PADRAM O-H mesh
method”, AIAA paper 88-0515.
generated for this OGV, which has 1.37M nodes. With this
mesh the steady flow was now well attached on the OGV and
[6] Baker, T.J., 1995, “Prospects and expectations for
the linear unsteady blade response and resulting noise
unstructured methods”, Proceedings of the Surface Modelling,
generation were much reduced. Figure 23 illustrates the levels
Grid Generation and Related Issues in Computational Fluid
of unsteady velocity response for the two mesh styles.
Dynamics Workshop, NASA-CP-3291.

11 Copyright © 2003 by ASME

[7] Steger, J., Dougherty, F.C., and Benek, J., 1983, “A chimera [20] Tschirner, T., Pfitzner, M., Mertz, R., “Aerodynamic
grid scheme”, Advances in Grid Generation, Ghia, K.N. and Optimisation of An Aeroengine Bypass Duct OGV-Pylon
Ghia, U., Eds., ASME FED, 5, pp 59-69. Configuration”, ASME GT-2002-30493, Netherlands, 2002.

[8] Benek, J.A., Buning, P.G., and Steger, J.L., 1985, “A 3-D [21] Harvey, N.W., Rose, M.G., Taylor, M.D., Shahpar, S. and
Chimera Grid Embedding Technique”, AIAA Paper 85-1523. Gregory-Smith, D.G., 2000, "Nonaxisymmetric Turbine
Endwall Design, Part I -3D Linear Design system", Journal of
[9] Kao, K.H., Liou, M.S., and Chow, C.Y., 1994, “Grid Turbomachinery, 122, number 2, pp286-294.
Adaptation using chimera composite overlapping meshes”,
AIAA Journal, 32, pp 942-949. [22] Parry, A.B., 2002, private communication, Fan System
Engineering, Rolls-Royce plc.
[10] Kallinderis, Y., Khawaja, A., and McMorris, H., 1996,
“Hybrid prismatic/tetrahedral grid generation for viscous flows [23] Van Zante, D.E. et al., 2000, “Recommendations for
around complex geometries”, AIAA Journal, 34, pp 291-298. Achieving Accurate Numerical Simulation of Tip Clearance
Flows in Transonic Compressor Rotors”, Journal of
[11] Irmisch, S., Walker, D., Bettelini, M., Haselbacher, A., and Turbomachinery, 122, pp733-742.
Benz, E., and Kallinderis, Y., 1999, “Efficient Use of Hybrid
Grids in Modern Turbomachinery Applications”, AIAA-99- [24] Moinier, P., and Giles, M.B., 1998, “Edge-Based Multigrid
0783. Schemes for Hybrid Grids”, 6th ICFD Conference on
Numerical Methods for Fluid Dynamics, Oxford, UK.
[12] Aftomis, M.J., Melton, J.E., and Berger, M.J., 1995,
“Adaptation and surface modeling for Cartesian mesh [25] Muller, J-D and Giles, M.B., 1998, “Edge-Based Multigrid
methods”, AIAA-95-1725-CP. Schemes for Hybrid Grids”, 6th ICFD Conference on
Numerical Methods for Fluid Dynamics, Oxford, UK.
[13] Thompson, J.F., Soni, B.k., Weatherill, N.P., 1999,
“Handbook of Grid Generation”, CRC Press.

[14] Jones, W.T., Samarah, J., 1995, “A Grid Generation

System For Multi-Disciplinary Design Optimization”, AIAA-
95-1689.

[15] Dawes, W.N., 1986, “A Numerical Method for the

Analysis of 3D Viscous Compressible Flow in a Turbine
Cascade; Application to Secondary Flow in a Cascade with and
without Dihedral”, ASME Paper 86-GT-145.

[16] Moore, J.G., and Moore, J., 1990, “The Moore Elliptic
Flow Program for Turbomachinery Flow Calculations – 1990
User Guide”, VPI & SU Turbomachinery Research Group
Report No. JM/90-3.

[17] Hall, E.J. and Delaney, R.A., 1995, “Investigation of

Advanced Counter Rotation Blade Configuration Concepts for
High Speed Turboprop Systems: Task VII – ADPAC User’s
Manual, NASA Contract NAS3-25270, NASA CR-195472.

[18] Sbardella, L., Sayma, A.I., and Imergun, M., 2000, “Semi-
Unstructured Mesh Generator for Flow Calculations in Axial
Turbomachinery Blading”, Int. Journal for Numerical Methods
in Fluids, 32, no. 5, pp 569-584.

[19] Wadia, A.R., Szucs, P.N., and Gundy-Burlet, K.L., 1999,

“Design and Testing of Swept And Leaned Outlet Guide Vanes
to Reduce Stator-Strut-Splitter Aerodynamic Flow
Interactions”, Transactions of the ASME, 121, pp416-425.

12 Copyright © 2003 by ASME

View publication stats