Predicting
uncertainty
Bahram Khalighi and GR Shevare, K Sudhakar, BK Gupta and Irshad Khan,
IIT Bombay present a case study based on CFD
CFD
simulations carried out by General Motors (GM) in the recent past were
adequate for predicting the drag coefficient of its vehicles within
5-10 per cent compared to the values obtained from wind tunnel tests.
Unfortunately, CFD simulations for lift coefficient were found to be
off by as much as 20 per cent. The project under discussion was one
of the steps towards investigating and understanding the sensitivity
of parameters involved, and correcting this discrepancy between the
computational and experimental results in the prediction of lift coefficient.
In order to eliminate the effect of geometric infidelity, the exact
geometry being tested in the wind tunnel was used for the CFD simulation.
The Initial Graphics Exchange Specification (IGES) files supplied were
translated without loss of information into the surface grids. The number
of test cases for validating CFD simulations based on structured grids
outnumbered CFD simulations based on unstructured grids. Thus the natural
choice for the simulation is structured grids, even though producing
structured grids is time consuming. Since the geometry is nontrivial,
only multi-block grids are feasible so that the quality of grids is
maintained throughout the computational domain. Structured surface grids
adequately clustered in the regions of high surface curvatures were
produced. H-H topology volume grids are generated as the van is like
a box. Grids were clustered close to the body using exponential function.
The outer boundary of the domain was set to the walls of the tunnel.
A sufficiently long length of the tunnel ahead and behind the model
was simulated so that boundary conditions were inflow/outflow, rather
than free stream boundary conditions. Boundary conditions for turbulence
kinetic energy, 'k' has been taken from wind tunnel tests. Due to all
these inputs, no wind tunnel solid/wake blockage corrections were necessary
and the uncertainties arising out of these corrections were taken care
of automatically. An in-house density based code has been used for simulation.
This code is a multi-block, second order accurate in space, finite volume,
parallelised software with multiple schemes for convective terms and
turbulence models. The scheme proposed by Weiss and Smith, with Jones
Launder k-e turbulence model has been used.
Methodology
Mesh:
The geometry of the vehicle and the outer boundary of the
domain (wind tunnel geometry) were read as IGES files by the in-house
CFD package, CFDExpert. Two triangular patches near the sidewalls of
the box were removed after discussions with GM. The domain had an inlet
and outlet at positions as provided in the IGES file. A tool in CFDExpert
generated surface triangulation from given the Non Uniform Rational
B-Splines (NURB) surfaces. A total of 81,000 grids were created in this
process. The generation of multi-block structured grids for simulation
requires the topology of the domain, ...
....CONTD
