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

net/publication/303902710

Comparative Analysis of the Arbitrary Mesh Interface(AMI) and Overset

Methods for Dynamic Body Motions in OpenFOAM

Conference Paper · June 2016

DOI: 10.2514/6.2016-3324

CITATIONS READS

4 2,614

2 authors:

Dominic Chandar Harish Gopalan

Queen's University Belfast Institute Of High Performance Computing
41 PUBLICATIONS   230 CITATIONS    59 PUBLICATIONS   337 CITATIONS   

SEE PROFILE SEE PROFILE

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

A Comparitive Study of Arbitrary Mesh Interface (AMI) and Overlapping Meshes in OpenFOAM View project

Urban Flow Simulations in Heterogeneous Systems View project

All content following this page was uploaded by Harish Gopalan on 04 July 2016.

The user has requested enhancement of the downloaded file.

 Comparative Analysis of the Arbitrary Mesh
Interface(AMI), Generalized Grid Interface (GGI) and
Overset Methods for Dynamic Body Motions in
OpenFOAM

Dominic Chandar∗ and Harish Gopalan∗

Fluid Dynamics, Institute of High Performance Computing, Singapore

This paper focuses on the comparative performance of the well known Arbitrary Mesh
Interface(AMI) and Generalized Grid Interface (GGI) methods in OpenFOAM and our
in-house Overset grid tool (OPErA), all of which are used to compute the flow past moving
bodies.To evaluate the performance of both approaches, the flow past a spinning cylinder at
Re 50000 and a canonical wind turbine are considered. Force coefficients, mass conservation
and parallel performance of the above approaches are illustrated. It is seen that the overset
approach is able to reproduce similar results compared to AMI and GGI approaches with
acceptable parallel overhead. While the GGI method is the fastest among the three, the
parallel efficiency of all the methods fall in the same range.

I. Introduction
Complex flows past rotating/moving objects can be computed using several approaches - The Multiple
Reference Frame(MRF) method, Overset/Overlapping grid method and by a Re-Meshing method. The
MRF model is usually applied for steady-state simulations and is not capable of reproducing unsteady
effects especially when the flow is turbulent. The Overset grid method is very flexible in terms of handling
arbitrary complicated body motions and generally suffers from mass conservation errors. Mass conserving
methods do exist, but they are computationally expensive due to regeneration of grid data structures at
the overset interface. There are currently no available Overset implementations within the OpenFOAM1
distribution. However, it comes with a different method known as the Arbitrary Mesh Interface(AMI)
method from the OpenFOAM main distribution and the Generalized Grid Interface(GGI) technique from
the Extend distribution. The AMI/GGI method requires meshes that rotate, to be in flush with stationary
meshes, but the cells do not have to match across the moving interface. Interpolation weights at the interface
are constructed such that the interpolation results in conservation.
The Overset module for OpenFOAM, the Overset Parallel Engine for Aerodynamics Applications(OPErA)2, 3
is an in-house implementation that can compute arbitrary complex motions. It is built as an external library
and can be linked to OpenFOAM with minor modifications to the source code. OPErA2 uses a face-based
marching algorithm to perform the donor search in parallel.
All the above cited methods, the AMI, GGI and Overset can be used for rotating objects. It is however
not known how these methods compare with each other in terms of accuracy and parallel performance. This
paper focuses on a comparative analysis of these methods in terms of force coefficients, mass conservative
properties and more importantly, parallel scalability. The paper is organized as follows: In Sec.II, the
governing equations for turbulent flow are described. Sec.III describes the computational test case with
comparative results and Sec. IV concludes the paper with future directions.
∗ Scientist, Fluid Dynamics, Institute of High Performance Computing, Singapore, AIAA Member

1 of 12

American Institute of Aeronautics and Astronautics

 II. Governing Equations
The governing equations for the flow variables are given by

∂U
ei
=0 (1)
∂xj

∂Uei ∂U
ei U
ej ∂ pe ∂2Uei Dij
+ =− +ν + (2)
∂t ∂xj ∂xi ∂xk ∂xk ∂xj
where U
ei is the velocity, pe is the pressure, ν is the kinematic viscosity and Dij is the turbulent stress tensor.
The turbulent stress tensor is specified as follows
2
Dij = kδij − 2νt Seij (3)
3

Here k is the turbulent kinetic energy (TKE), νt is the turbulent viscosity, and Seij is the symmetric part
of the velocity gradient tensor. K-ω SST model4 is used as the baseline linear RANS model. The model
solves transport equations for the TKE and turbulent dissipation ω,4 respectively. The model equations are
given by  
∂k ∂k ∂ ∂k
+u ei = Pk − β ∗ ωk + (ν + σk νt ) (4)
∂t ∂xi ∂xi ∂xi
 
∂ω ∂ω γ ∂ ∂ω
+u
ei = Pk − βω 2 + (ν + σw νt ) (5)
∂t ∂xi νt ∂xi ∂xi
σw2 ∂k ∂ω
+ 2 (1 − F1 )
ω ∂xi ∂xi
where Pk = max(Dij ∂e ui /∂xj , 10β ∗ ωk), is the kinetic energy production term, β ∗ = 0.09 is a model constant
and the last term in the ω equation is the cross-diffusion term. F1 is a blending function which has a value
of one inside the boundary layer and zero outside. The limiting of the production term is an alternative to
the use of damping function in the near-wall region. The turbulent stress tensor and viscosity are computed
in this model as follows
2
Dij = kδij − 2νt Seij (6)
3
a1 k
νt =  q  (7)
max a1 ω, 2Seij Seij F2

Here a1 = 0.31 is a model constant and F2 is a blending function similar to F1 . The expressions for the
blending functions are given by
 ( " √ # )!4 
k 500ν 4σw2 k
F1 = tanh  min max ∗ , 2 ,  (8)
β ωd d ω CDkw d2
 
σw2 ∂k ∂ω
CDkw = max 2 , 10−10 (9)
ω ∂xi ∂xi
 √ !2 
k 500ν
F2 = tanh  2 ∗ , 2  (10)
β ωd d ω

The model constants are calculated by blending K-ω model near the wall and K- away from the wall using
the blending function F1 . The form is given by

φ = φ1 F1 + (1 − F1 )φ2 (11)

The model constants are given in Table 1.

2 of 12

American Institute of Aeronautics and Astronautics

 φ1
σk = 0.85 γ = 5/9 β = 0.075 σw1 = 0.5
φ2
σk = 1.0 γ = 0.44 β = 0.0828 σw2 = 0.856

Table 1. Model constants for the SST model.

III. Results
A. Flow Past a Spinning Cylinder in Cross flow at Re 50000 - AMI and Overset
We consider the flow past a spinning cylinder with a spin ratio (U tip/U∞ = 1.0), where U tip = rωr , r
being the radius of the cylinder and ωr the rotational speed. On a mesh of 15000 cells(7500 on a rotor mesh
and 7500 on a stator mesh) with a mean y+ = 5, the flow is initialized with potential flow conditions using
potentialF oam. Using a k −ωSST model described in Sec. II, computations are carried out for 250 turn-over
times (t̄ = 2tU∞ /r). Both solvers use the multigrid for the pressure equation with identical convergence
criteria. The current case is run on a single processor for the AMI implementation and on two processors
for the Overset implementation. This is due to the fact that the overset implementation needs at least one
processor per computational grid. Figures 1,2 compare the drag and lift coefficients for both cases. The
mean drag values compare well with existing 2D RANS data from,5 but there is a discrepancy with respect
to the mean lift coefficient. In figures 3,4 a comparison of the velocity contours is made at the interface of
both the meshes. Both approaches give reasonable continuity in the contours. Due to a phase shift between
the time histories(which can be seen from the force histories later on), the wake is not positioned at the
same location at this physical time. The computed global mass conservation div(U ) for the AMI and overset
are 5.83 × 10−11 and 5.04 × 10−10 respectively. The overset method has an order of magnitude higher error
than the AMI, but both are small in magnitude and are acceptable for most practical purposes. Note that
there are roughly two cells of the near-body(rotor) for every cell in the background(stator) for the overset
and AMI configurations.

B. Spacing Ratio Analysis

In this section, we consider varying the relative cell size(or perhaps the edge size) at the interface of the rotor
and stator. We consider three cases where the spacing on the stator mesh at the rotor-stator interface is 1.5,
2.0 and 3.0 times that of the spacing on the rotor mesh at the same location. The extent of the rotor mesh
is roughly twice that of rotor mesh considered in the previous section. Figures 5,6 show a part of the mesh
at the interface for the overset and AMI/GGI cases respectively for the 3.0 spacing ratio. There are roughly
three edges of the stator mesh for every edge of the rotor mesh at the interface. Table2 shows a summary
of the force coefficients for all possible cases. The drag coefficients are on an increasing and decreasing trend
for the AMI/GGI and the overset cases respectively. However, all GGI cases have relatively higher drag
coefficients compared to the AMI and Overset cases. The data presented for the 2.0 spacing ratio case is
considerably different than that presented in the previous section which was also for a 2.0 spacing ratio. This
is due to the fact that, although the previous setup had a smaller outer distance on the rotor(which can cause
interpolation errors to affect the behavior near the wall), the meshes at the interface were relatively finer
than the case presented here. Another observation that can be made from these spacing tests on AMI/GGI
and overset computations is that, a phase shift is introduced in the force coefficients as the spacing is varied
(Figures 7-9).

Table 2. Computed force coefficients for different spacing ratios

CD CL
Spacing Ratio -> 1.5 2.0 3.0 1.5 2.0 3.0
AMI 0.751 0.762 0.773 1.876 1.876 1.875
GGI 0.802 0.811 0.818 1.853 1.851 1.849
Overset 0.730 0.714 0.701 1.866 1.849 1.829

3 of 12

American Institute of Aeronautics and Astronautics

Figure 1. Comparison of drag coefficients between AMI and Overset. Mean value AMI=0.66, Overset=0.67,
Reference=0.65

C. Comparative Parallel Performance

A strong scaling of all the methods has been performed for different spacing ratios on the recent National
Supercomputing Centre. There are roughly 800K and 400K cells on the rotor and stator respectively.
Computations are performed only for ten steps to gauge the performance from 18 to 576 processes. Although
GGI and Overset computations were able to run beyond 144 processes, the AMI computations had issues
running on higher processor counts. On 144 processes, and for a spacing ratio of 1.5, AMI, GGI and Overset
consume about 3.5s, 3.25s and 4.4s respectively. The GGI is marginally faster than other methods and
this observation is consistent among all cases considered. In terms of parallel efficiency, all methods are
comparable as seen from Figure10 with the overset case running with a marginally higher efficiency. In
Figure 11, the fraction of the overset overhead compared to the total solution time for a given time step
is plotted as a function of processor count for different spacing ratios. It is encouraging to note that this
fraction is only about 1-2%. This fraction however increases gradually since the number of search cells
involved in the donor search process is very small, contributing to more communication.

IV. Conclusion and Future Directions

In this paper, a comparative analysis of different moving mesh methods in OpenFOAM are performed to
study its accuracy and scalability. The following are the brief conclusions:
• Overset runs demonstrate comparable solutions with existing AMI and GGI implementations.
• All methods display a consistent phase shift in the force coefficients for different spacing ratios between
the rotor and the stator
• Overset runs have marginally higher parallel efficiency compared to the GGI implementation.
• The fraction of the time spent for the overset part stays low for a given mesh for the span of the
processor count considered.

4 of 12

American Institute of Aeronautics and Astronautics

Figure 2. Comparison of lift coefficients between AMI and Overset. Mean value AMI=1.96, Overset 1.9,
Reference=1.5

Acknowledgments
The computational work for this article was partially done on resources of the National Supercomputing
Centre, Singapore (https://www.nscc.sg)

References
1 Jasak, H., Aleksandar J., and Zeljko T., ”OpenFOAM: A C++ library for complex physics simulations.” In International

workshop on coupled methods in numerical dynamics, vol. 1000, 2007, pp. 1-20.
2 Chandar, D., Development of a Parallel Overset Grid Framework for Moving Body Simulations in OpenFOAM, Journal

of Applied Computer Science and Mathematics, Vol. 20, No. 9, 2015, pp. 22-30.
3 Chandar, D., Nguyen, V.T., Gopalan, H., and Jaiman, R., Flow Past Tandem Circular Cylinders at High Reynolds

numbers using Overset Grids in OpenFOAM, 53rd Aerospace Sciences Meeting, Scitech 2015, AIAA-2015-0315, Kissimmee,
FL.
4 Menter, F. R., ”Improved two-equation k-omega turbulence models for aerodynamic flows.” NASA STI/Recon Technical

Report N 9, 1992, 22809.

5 Elmiligui, A., Abdol-Hamid, K. S., Massey, S. J., and Pao, S. P., ”Numerical study of flow past a circular cylinder using

RANS, Hybrid RANS/LES and PANS formulations”, In 22nd Applied Aerodynamics Conference and Exhibit, Rhode Island,
2004.

5 of 12

American Institute of Aeronautics and Astronautics

 Figure 3. Velocity magnitude contours near the AMI interface

Figure 4. Velocity magnitude contours near the Overset interface

6 of 12

American Institute of Aeronautics and Astronautics

Figure 5. The rotor-stator interface for the overset case with a 3.0 spacing ratio

7 of 12

American Institute of Aeronautics and Astronautics

Figure 6. The rotor-stator interface for the AMI/GGI case with a 3.0 spacing ratio

8 of 12

American Institute of Aeronautics and Astronautics

Figure 7. Comparison of drag coefficients for different spacing ratios using AMI

Figure 8. Comparison of drag coefficients for different spacing ratios using GGI

9 of 12

American Institute of Aeronautics and Astronautics

Figure 9. Comparison of drag coefficients for different spacing ratios using Overset

10 of 12

American Institute of Aeronautics and Astronautics

 1.8
Overset spacing 1.5
1.6 Overset spacing 2.0
Ovserset spacing 3.0
AMI spacing 1.5
1.4
AMI spacing 2.0
AMI spacing 3.0
1.2
GGI spacing 1.5
Parallel Efficiency

GGI spacing 2.0

1 GGI spacing 3.0

0.8

0.6

0.4

0.2

0
0 100 200 300 400 500 600
Number of Processes

Figure 10. Parallel efficiency for AMI, GGI and Overset runs

11 of 12

American Institute of Aeronautics and Astronautics

 3
1.5 spacing
2.0 spacing
3.0 spacing
% of Overset Time relative to Total Time

2.5

1.5

0.5
0 100 200 300 400 500 600
Number of Processes

Figure 11. Fraction of the overset overhead with respect to the total solution time for a time step

12 of 12

American Institute of Aeronautics and Astronautics

View publication stats