Anda di halaman 1dari 18

Comparison between two and three-dimensional POD in a turbulent boundary layer using multi-plane stereoscopic PIV

Alex Liberzon*, Roi Gurka and Gad Hetsroni


Tel Aviv University, Ben Gurion University, Technion ETC13, Sep 12-15, 2011

Outline

Background and motivation Description of the experimental setup Description of POD in 2D and in 3D Comparison Summary and conclusions

Background and motivation

Coherent structures in turbulent ows Objective (!) detection and characterization (geometric, dynamic, etc.) Lumley (1970) introduction to Proper Orthogonal Decomposition Broad use and potential abuse of POD detecting coherent motions One of the main aspects is due to the 2D cross-sections as compared to full 3D analysis

Experimental setup

10

11

12

13

5 4

XPIV method

Fig. 5. Original imag

with particles in the

Fig. 6. Original (left)

XPIV algorithm

image (right)

Fig. 8. Gradient image (l

XPIV algorithm contd.

enhanced gradient map (

Fig. 9. Defocus particles

the identied objects in a (right)

Fig. 10. Size distribution

of the binary image (left)

Proper orthogonal decomposition

POD for turbulence

Lumley 1970 Holmes, P. and Lumley, J.L. and Berkooz, G 1996 Sirovich snapshots method 1987

R (x , x )(x )dx
f k (x ) =

= (x )

(1)

K n=1

k an n (x )

(2)

Snapshots method (Sirovich 1987)


For the three-dimensional analysis, the based on a set of vector of the length

Nx Ny Nz grid points and expressed as a n = Nx Ny Nz :


= k . . . k n

kth snapshot is dened


T

(3)

= ( )T ( )

(4)

C =
(n) =
Here

(5)

M j =1

(n)

(6)

(n)

denotes the

corresponding to the

jth element of the eigenvector , jth eigenvalue (n) .

Our POD approach

Vorticity is used instead of velocity Liner combination of the modes in the spirit of characteristic eddy (Lumley 1970)

(x) =

K n=1

(n) (n) ((x )),

(7)

2D vs 3D: each plane is analysed separately or all the planes together

Results and discussion


1 0.9 0.8

Sum of slice-POD 3D - POD

Cumulative energy

0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 0 50 100

middle plane

lower plane

higher plane
150

# modes

Figure: Cumulative energy of the POD modes using the 3D-POD (solid line) and the slice-POD for separate planes and their summation (dashed-dotted line).

Dierent rst modes

Dierent higher modes

Dierent combination of modes

(a)

(b)

2 Figure: Comparison of the wall-normal enstrophy component y obtained using a 3D POD (a) versus the eld obtained three single-plane slice-POD (b).

Or dierent structures?

Figure: Comparison between the reconstruction of a single snapshot

Summary

The multi-plane stereoscopic particle image velocimetry measurement system (XPIV) was utilized to measure the boundary layer in a ume. Three-component velocity vector elds at three parallel streamwise - spanwise planes with a xed distance between them were obtained.

The proposed analysis emphasizes the necessity of the three-dimensional measurements in turbulent boundary layers. We calculate the POD modes of the wall normal component of the vorticity elds. This procedure was performed using two methods: A) two-dimensional slices of the data were used separately in a so-called slice-POD, and B) the results are utilized as a volumetric dataset to obtain the 3D-POD modes.

Conclusions

Both decompositions capture all the essential information of the underlying ow features, as shown by the snapshots reconstruction.

However, the ow patterns based on the combination of the lower order modes shows discrepancies between the slice- and 3D-POD.

The interpretation should be very careful The 2D POD analysis is valid and is important in order to emphasize the patterns which are limited in span to the cross-section, but could be misleading without the comparison with the 3D-POD analysis.

Turbulent boundary layer: streaky patterns in 2D are absent in the 3D-POD modes. This is due to the energy distribution of the boundary layer which has a maximum at the buer region, presented here as a middle plane.

Anda mungkin juga menyukai