The significance of coherent flow structures for the turbulent mixing ...

5 Investigation of the xy-plane
diameter, surrounded by a dark background. The loss of light associated with this setting was
compensated by combining the output beams of four independent Nd:YAG laser cavities (BMI
model 5013 DNS 10), each approximately 255 mJ output energy per pulse ×XØ È¿Â0Û at nm (see
figure 4.2 on page 52) and by triggering the system in such a way that always two differently
polarised pulses leave the laser-heads simultaneously. It is clear that always a delay of the
order of the pulse length must exist between the two pulses to avoid laser induced damage
of the frequency doubler crystal due to the high peak intensity, which might occur when the
two beams interfere. Usually, this is automatically realized because the time accuracy of the
PIV triggering systems is not suited to superimpose exactly two ÆÁÌÈ m long light pulses (È ns
pulse duration assumed) travelling with speed of †ÄÐƾzy m/s. The ¾ÁÃß mm light-sheet was
formed by three appropriate cylindrical and spherical lenses mounted in an optical bench and
directed into the test-section by using a properly coated mirror, see section 4.6. To ensure
that the measurement plane is aligned with respect to the stream-wise wall-normal plane, the
cylindrical lens was rotated until the light-sheet reflection at the ceiling was parallel with a
mark indicating the centreline of the wind-tunnel and the mirror was tilted in such a way that
the weak light-sheet reflections, coming from the ceiling- and floor-window of the test-section,
coincide on the mirror with the light-sheet coming from the optical system.
The tracer particles generated for this experiment were delivered from a smoke generator
which produces high concentrations of monodisperse poly-ethylene-glycol particles with
a mean diameter of Û@Õ m according to figure 2.4 on page 18 or [47]. To obtain ideal conditions
for accurate PIV measurements, the closed circuit wind-tunnel was completely seeded
and continuously operated until no seeding inhomogeneities could be observed at all. During
this procedure all diffuse wall reflections, caused by the interaction of the laser-light with
tracer particles sticked on the wall, were removed with appropriate liquids and pressurised air
while the wind-tunnel was running at a moderate speed. The operation of the wind-tunnel is
essential, and should never be stopped before the experiment is finished in order to avoid any
settling of the tracer particles which might deteriorate the performance of the measurements.
As these particles follow the flow precisely, no settling of the particles or other diffuse reflections
could be observed on the acquired data while the experiment was running. Only a tiny
bright line of the wall reflection (Â pixel wide) was visible due to the glass / air interface. After
the experiment, this line was detected by using correlation algorithms in order to determine
the position of the wall with sub-pixel accuracy.
For the evaluation of the stereo-scopic images the second order warping technique, outlined
in section 3.2, was applied along with the calibration validation procedure to ensure that
the interrogation spots from each of a pair of stereo-scopic images correspond exactly to the
same region of the flow, see section 3.3. The interrogation of the data was performed with
the FFT-based free shape cross-correlation and for the peak detection with sub-pixel accuracy
the two-dimensional Gaussian fit was applied. This peak finding method is less sensitive
to sub-pixel displacements compared with the three point Gaussian peak fit, see figure 2.13
and 2.14. For the calculation of the velocity Â0ÛÓÄœÂ0Û vectors pixel ß¿ËFÄ Æß and pixel interrogation
windows were selected. Because of the increased spatial resolution of the stretched
windows in wall-normal direction spatial averaging effects could be easily estimated from the
performed evaluation. It should be noted that an aspect ratio of 4 between the stream-wise
and wall-normal dimensions of the measurement volume is relatively moderate regarding the
typical values which are applied in numerical flow simulations. The evaluation of the data was
performed in parallel on an eight processor high performance SGI computer, and a raid sys-
74