Lecture1 PDF

COMPUTATIONAL FLUID DYNAMICS: Important Aspects of CFD Analysis
Lecture 40
IMPORTANT ASPECTS OF REAL LIFE CFD ANALYSIS
40.1 INTRODUCTION
CFD analysis provides an approximate solution of the underlying flow problem. An analyst
must be aware of the approximations made at different stages (modelling as well as
simulation stages) and their impact on the solution accuracy. Ideally, the following aspects
should be addressed:
• A quantitative bound on the solution accuracy should be obtained as part of the
simulation process.
• Numerical solution must be verified, and validated with available experimental data if
possible.
In this lecture, we look at some of these aspects. Further, CFD analysis of real life problems
invariably requires modelling of complex geometries and solution of large scale algebraic
systems which cannot be solved on a serial workstation. We would outline some
discretization methods for complex geometries and solution techniques for parallel
computing.
40.2 VERIFICATION AND VALIDATION

CFD simulation represents an approximation to a real world problem. Approximations are
involved in each stage of numerical simulation process: in mathematical modelling, boundary
conditions, discretization and computer solution of the discretized system of equations. Errors
in the solutions must be identified and quantified for use of CFD results in engineering
analysis and design. This quantification process should be done in two steps: (a) verification
and (b) validation (AIAA, 1998).
40.2.1 Verification
Verification stands for quantitative estimation of the closeness to the numerical simulation
results to the exact solution of the mathematical model. The verification process requires
comparison of the computational solution with known analytical solutions OR high-accuracy
benchmark solutions. CFD simulation of any flow problem must be verified as follows:
• Grid independence test: Perform a careful grid independence test in spatial as well
as temporal domain by systematically refining grid size and time step used in CFD
simulation. This step would yield a grid-independent solution.
• Bench-mark comparison: To establish the accuracy of the CFD solution, compare it
with analytical solution (if available) or high-accuracy benchmark solution (which
may have been obtained using some other numerical scheme and a high-resolution
grid).
Note that the verification process provides a quantitative measure of accuracy of CFD
solution with respect to the mathematical model used in CFD analysis.
Dr K M Singh, Indian Institute of Technology Roorkee NPTEL 40.1

40.2.2 Validation
Validation is the process of determining the closeness of the approximate numerical
simulation to actual real world problem (AIAA, 1998). CFD simulations should ideally be
validated with experimental measurements performed on the real system (or its physical
model). In real life applications, it may not be possible to obtain detailed experimental data
for the physical system. In such cases, validation is normally done by comparing the
numerical simulation with experimental data obtained for a sub-system. For further details on
the validation process, refer AIAA(1998) and Versteeg and Malalasekera (2007) .
40.3 METHODS FOR COMPLEX GEOMETRIES

Practical flow problems invariably involve complex geometries. For these problems,
generation of the suitable grid(s) for CFD simulation is the most demanding aspect of CFD
modelling. The type of grid is dictated by the numerical method used for numerical
simulation as well as requirements of the flow physics (e.g. need to provide fine mesh close
to the wall to resolve boundary layer).
40.3.1 Unstructured grid techniques

With finite volume and finite element based CFD analysis of flow in complex geometries can
be modelled using unstructured grids. The following guidelines should be helpful in
generation of a good quality grid for CFD analysis:
• Instead of attempting to generate volume mesh in the entire geometry, decompose the
complex problem domain in many small sub-domains. Mesh each sub-domain
separately.
• For a properly graded grid, first mesh the edges, then the surfaces of a sub-volume.
Thereafter, generate the volume mesh.
• Check quality of the mesh before using it in CFD simulation.
• Care should be exercised in merging the grids of the subdomain. One of the most
common errors is creation of an artificial wall boundary between two sub-domains
while merging. This must be taken care of.
• In finite volume analysis, hexahedral finite volumes are usually preferred over
tetrahedral (due to better accuracy obtained in interpolation and integration with the
former). Hence, tetrahedral and wedge elements should be confined to regions too
complex to mesh using hexahedral elements.
40.3.2 Structured grid methods

We have three options with structured grids for complex geometries:
1. Cartesian grids in which use a stair-case approximation for the curved parts of the
complex domains.
2. Body-fitted structured grids which are based on mapping of the complex physical
domain onto a rectangular computational domain. These require transformation of
governing equations in curvilinear co-ordinates.
3. Block-structured Cartesian grids are based on the division of the problem domain
into smaller sub-domain with each sub-domain having a different Cartesian grid.
Thus, finer grids can be used in regions near solid boundaries providing a closer
approximation of the problem geometry.
Initially, use of stair-case approximation for the curved parts of the complex domains was
only option with Cartesian grids. However, recent developments in immersed boundary

methods (IBM) have brought the Cartesian grids back in favour for arbitrarily complex
geometries. IBMs permit a much better representation of actual geometries of problem
domain while still employing a Cartesian grid (which may be block-structured). For further
details, please see Peskin (2002), Mittal and Iaccarino(2005).
40.4 PARALLEL IMPLEMENTATION

Discretization of a complex real life problem domain (say flow around an aircraft or an
automobile) usually requires a very large number of grid nodes/elements (order of a few
millions or even billions). The resulting system of equations of such a large order cannot be
solved on serial workstations, and one must use a parallel cluster which could be
• A shared memory machine (wherein all processors share a large random access
memory), or
• A distributed memory cluster where each processor (or node) has its own dedicated
memory.
Parallel implementation of a CFD code on each type of machine would be somewhat
different. However, in either context, the bigger problem must be broken into smaller sub-
problems which can be solved in parallel on a set of processors. Two popular approaches for
this break-up are
• Domain decomposition: Divide the problem domain into a set of non-overlapping
subdomains, generate grid on each subdomain, and proceed with the solution process.
Each processor (or processor core) is assigned a subset of these subdomains. This
approach can be used with any discretization scheme.
• Grid partitioning: Use graph-partitioning tools (such as METIS and JOSTLE) to
divide the overall discretized problem into a set of sub-grids.
In parallel implementation, care must be taken to balance the computing load on each
processor. Further, iterative schemes which can be easily parallelized have to be chosen for
solution of the discrete algebraic system.
REFERENCES/FURTHER READING
AIAA (1998). Guide for the Verification and Validation of Computational Fluid Dynamics
Simulations, AIAA Guide G-077-1998.
Chung, T. J. (2010). Computational Fluid Dynamics. 2nd Ed., Cambridge University Press,
Cambridge, UK.
Ferziger, J. H. And Perić, M. (2003). Computational Methods for Fluid Dynamics. Springer.
Mittal, R. and Iaccarino, G. (2005). Immersed Boundary Methods, Annual Review of Fluid
Mechanics, vol. 37, pp. 239–261.
Peskin, C. S. (2002). The immersed boundary method, Acta Numerica, 11, pp. 1–39.
Versteeg, H. K. and Malalasekera, W. M. G. (2007). Introduction to Computational Fluid

Dynamics: The Finite Volume Method. Second Edition (Indian Reprint) Pearson Education.

Lecture1 PDF

Uploaded by

Lecture1 PDF

Uploaded by

COMPUTATIONAL FLUID DYNAMICS: Important Aspects of CFD Analysis

40.2 VERIFICATION AND VALIDATION

Dr K M Singh, Indian Institute of Technology Roorkee NPTEL 40.1

40.3 METHODS FOR COMPLEX GEOMETRIES

40.3.1 Unstructured grid techniques

40.3.2 Structured grid methods

Dr K M Singh, Indian Institute of Technology Roorkee NPTEL 40.2

40.4 PARALLEL IMPLEMENTATION

Versteeg, H. K. and Malalasekera, W. M. G. (2007). Introduction to Computational Fluid

Dr K M Singh, Indian Institute of Technology Roorkee NPTEL 40.3

You might also like