On the application of image analysis techniques to fluid mechanics experiments John H. Cushman1, Natalie Kleinfelter1, Monica Moroni2 1Department of Earth and Atmospheric Sciences and Department of Mathematics, Purdue University, West Lafayette 2Department of Hydraulics, Transportations and Roads, University of Rome “La Sapienza” Rome (Italy) Purpose • To introduce image analysis techniques – Non-intrusive – Flexible – Lagrangian description->Eulerian description • Quantities can be derived • Phenomena can be investigated Tables of contents Image analysis – digital image processing – different algorithms for image analysis Applications – Porous media – Convective boundary layer – Subduction – Multi-dune channel – Fully developed turbulent channel “Ingredients” for image analysis The fluid under investigation and the test section have to be transparent: mono-phase and multi-phase systems The fluid has to be seeded with tracer particles with the following features: neutrally buoyant and highly reflecting One or more cameras, a high power light source, an acquisition and digitalization system and image analysis system are required Image analysis Methods and applications Images: formation and representation Digital image processing is concerned with the computer processing of pictures or, more generally, images that have been converted into a numeric format. An image is a picture, photograph, display, or other form giving a visual representation of an object or scene. However, in digital image processing, an image is a set of K two dimensional arrays of numbers of N-lines (rows) and Msamples (columns). NxM K K=1 K=3 image resolution defines image bands black and white image color image Images: formation and representation 40 40 40 40 40 40 40 40 40 40 40 40 40 40 40 200 200 200 40 40 40 40 40 40 40 40 200 200 200 40 40 40 40 40 40 40 40 200 200 200 40 40 40 40 40 200 200 200 200 200 200 200 200 200 40 40 200 200 200 200 200 200 200 200 200 40 40 200 200 200 200 200 200 200 200 200 40 40 40 40 40 200 200 200 40 40 40 40 40 40 40 40 200 200 200 40 40 40 40 40 40 40 40 200 200 200 40 40 40 40 40 40 40 40 40 40 40 40 40 40 40 The figure in an image of a simple geometric pattern (left-hand side) and its corresponding digital image (right-hand side). K=1. The digital image has 11 lines and 11 samples per line. Each number in the matrix corresponds to one small area of the visual image and the number gives the level of darkness or lightness of the area Images: formation and representation We assume the higher the number, the lighter the area, so zero is black, the maximum value is white, and intermediate values are shades of grey (this is arbitrary). Each small area to which a number is assigned is called pixel (picture element). The size of the physical area represented by a pixel is called the spatial resolution of the pixel. Each pixel has its value, plus a line coordinate and a sample coordinate. The minimum value a pixel can have is typically zero, and the maximum depends on how the number is stored in the computer. It is common to store each pixel as a byte (8 bits). Maximum grey level value? 255=2^8-1 0255 grey level scale Images: display as a matrix Images: display as a 3D surface Images: display as an ‘image’ High quality image Zoom in previous image Further zoom in previous image Low quality image Sequence of images: file .avi Do other kinds of images exist? Differences with previous images: - color image - no tracer particles Do images contain useful information? The analysis of those images allows the convective boundary to be determined. It is localized at the interface between the image light and dark regions Buffer Red Buffer Green Buffer Blue Ratio pixel/cm 328 pixels pixel/cm=328/1 Particle Tracking Algorithms Particle Tracking algorithms require: • seeding the fluid with small highly reflecting particles at a given particle density; • illuminating as uniformly as possible the flow field with a light sheet; • acquiring images of the particles located in this sheet with an imaging rate fast enough to maintain good resolution. In contrast to standard PIV (Particle Image Velocimetry) where the mean displacement of a small group of particles is sought, with Particle Tracking algorithms pathlines of individual particles are obtained. PTV (Particle Tracking Velocimetry) Main steps of Particle Tracking Velocimetry (PTV): - pre-processing the images to eliminate background noise; - determining the particle centroid coordinates in each frame with sub-pixel accuracy: • grey level “binarization”; • pixel labeling; • centroid determination; - tracking particle centroids frame by frame. PTV: image pre-processing – background removal ‘Raw’ images Background removal effect n=17 rows Pixel under investigation 1 I (i, k ) 2 n n=17 columns n n I (i, k ) i 1 k 1 !!! n has to be odd I filtered (i, k ) I (i, k ) I (i, k ) PTV: centroid determination - binarization Count foreground Threshold level for background pixel brightness: i.e. 100 Grey levels 0 Ideal threshold 255 PTV: centroid determination Threshold level for pixel brightness: i.e. 100 Options for computing the centroid position: - Arithmetic average of pixel coordinates - Weighted average of pixel coordinates and grey levels (weights) - Average over a gaussian distribution PTV: trajectories reconstruction t3 t2 t1 toll t0 Dmax Tracer density??? Problem What if tracer density is high? Feature tracking The main steps in FT are: • seeding the fluid with small highly reflecting particles at any particle density; • illuminating as uniformly as possible the flow field with a light sheet; • acquiring images of the particles located in this sheet with an imaging rate fast enough to maintain good resolution. Feature Tracking Continuity equation for the “optical flow” (“image brightness constancy constrain” (BCC)) DI I I I I u v I T U I t uI x vI y 0 Dt t x y t If the equation is computed at a single point, it only provides one equation for two unknowns, the velocity components. It is only when the equation is evaluated at each point in a region W surrounding the one under investigation (feature), that it provides sufficient information on U The cost function SSD (Sum of Squared Differences) over a window W surrounding the feature under investigation representing the dissimilarity between the image I at time tA (IA) and at successive time tB (IB), can be written 2 1 1 1 I DI 2 T I B I A dS dS I U dS SSD 2 W W W W t Wt W Dt t A 2 Feature Tracking To obtain a least squares estimation of U(x), the derivative of the cost function with respect to U is evaluated: 2 SSD 2 I I x T I I U dS 2 U WW t W I y I x Setting the equation to zero: I x2 dS W I I dS y x W I x I t dS I y dS W W U 0 2 I I dS W I y dS y t W I x or more simply: G U b 0 IxIy I x I t U I I dS 2 I y y t Feature Tracking The square matrix G is invertible if both eigenvalues 1 and 2 are non zero. Three different cases are possible: Ix=0 and Iy=0, i.e. uniform intensity distribution in both directions (case a): both eigenvalues are zero Ix=0 and Iy≠0, i.e. uniform intensity distribution in the x direction (case b) or Ix≠0 and Iy=0, i.e. uniform intensity distribution in the y direction (case c): one eigenvalue is null (1>0; 2=0 ) Ix≠0 and Iy≠0, i.e. not uniform intensity distribution in both directions (case d): both eigenvalues are positive (1>0; 2>0) y a) y x b) y y x x c) x d) FT vs. PTV Highlights on PTV –Best buffer identification and choice of the threshold value; –Centroid identification with sub-pixel accuracy; –Trajectory reconstruction: the nearest principle is employed. The algorithm applies the criterium of “minimum acceleration” when ambiguities arise. Highlights on FT in a pure traslational model – Image features selection – Features tracking on from frame No constraints to frame. image particle The matching measure density introduced to follow a feature (and its interrogation window) and “most similar” region at Noitssubjective parameters thehave successive time is the to be provided “Sum of Squared Differences” (SSD) among intensity values: the displacement is defined as the one minimizing the SSD. Problem What if the flow is 3D? The state of art on 3D techniques There exist a number of imaging-based measurement techniques for determining 3D velocity fields in an observation volume. Among these are: • scanning techniques (Guezennec et al. 1994, Moroni and Cushman, 2001a, b); • holographic techniques (Hinsch and Hinrichs 1996, Katz 1999); • defocusing techniques (Willert and Gharib 1992); • photogrammetric techniques (Maas 1992, Kasagi and Nishino 1990). Main features of scanning and photogrammetric 3D-PTV “Scanning” methods share the following basic steps (Moroni & Cushman, 2001): 1. 2. 3. 4. set-up calibration determination of centroids in 2-D; trajectories reconstruction in 2-D; trajectories matching 3-D. “Photogrammetric” methods share the following basic steps (Papantoniou and Dracos, 1990): 1. stereoscopic calibrated imaging and recording of a suitably illuminated particle flow; 2. photogrammetric analysis of the resulting images to derive the instantaneous 3-D particle positions; 3. tracking of the 3-D coordinate sets in time to derive the tracer trajectories. Scanning 3D-PTV optical rays are supposed to be parallel Three dimensional trajectory construction from two 2-D projected trajectories Photogrammetric 3D reconstruction Y' Pi Z Assumptions: Zi – model of imaging: pinhole camera modeled by its Y P' optical center C and the image plane R C – the theorem of photogrammetry holds Xi i y c x Yi Z0 Introduce the following reference frames: X0 z Y • world reference frame X,Y,Z; X' • 0image reference frame ; X •Z' camera standard reference frame X',Y',Z' 0 3D trajectory reconstruction In formulas: 0 c r11 ( X X 0 ) r21 (Y Y0 ) r31 ( Z Z 0 ) r13 ( X X 0 ) r23 (Y Y0 ) r33 ( Z Z 0 ) r12 ( X X 0 ) r22 (Y Y0 ) r32 ( Z Z 0 ) 0 c r13 ( X X 0 ) r23 (Y Y0 ) r33 ( Z Z 0 ) (X,Y,Z) (X0,Y0,Z0) R= rij (0,0) c R : object point coordinates : camera projective center coordinates : elements of 33 rotation matrix with angles , , : image principle point : image principle distance cos cos cos sen sensen cos sensen cos sen cos cos sen cos cos sensensen sen cos cos sensen sen sen cos cos cos 3D trajectory reconstruction Methods for detecting 3D trajectories: Geometric method Radiometric method Acquisition system calibration 2D centroid locations Tracking in 2D Correspondence Structure from stereo 3D location coordinates Tracking in 3D Calibration Aim: 9 parameters to be determined: 0, 0, c, , , , X0, Y0 and Z0. The camera constructor usually insures 0 and 0 to be equal to zero. Method: given r observations (r>u) li ai1 x1 ai 2 x2 ... aiu xu where xk are unknowns (equal to u); li is the observation; aik are coefficients. Since r>u s Axˆ l linear equation Corrections ˆx ( A T A) 1 A T l Calibration (2) non-linear equation li f i ( x1 , x2 , ... , xu ) A Taylor expansion is required, with starting values. Each point Pi of unknown or known coordinates (Xi,Yi,Zi) and known image coordinates ( i , i ), provides an equation as: 0 0 0 dX 0 dY0 dZ 0 dc si d c X 0 Y0 Z 0 0 d d ( ij ij0 ) si ... (analogous) 0 0 0 Calibration (3) R Y Z G X B Calibration (4) N points st st st Red camera 342 90 300 0 Green camera 349 90 0 0 Blu camera 360 90 60 0 Iterative procedure Calibration (5) Camera X0 (cm) Y0 (cm) Z0 (cm) c (cm) mean (pixel) st dev (pixel) Red -55.39 -17.29 14.55 2.72 89.47 299.35 -0.31 7.25 3.19 Green 12.09 -54.58 14.42 2.54 89.89 0.00 0.24 7.06 3.21 Blu 79.00 -18.19 12.64 2.62 92.86 59.49 -2.65 17.91 7.82 Need for a calibration procedure with an easier implementation Structure from stereo Correspondences using the epipolar lines Pf Pe Pd I3 E ( 23 ) i P Pa Pb Pc E12 I1 I2 Test on synthetic data The algorithm was tested on the basis of a synthetically generated data set simulating curling trajectories with a starting location randomly distributed in the world reference system. The effect of an increasing number of particle seeding the measurement volume, i.e. of trajectories, was tested. The effect of a wrong positioning of camera’s projective centers and rotation angles was tested as well. Test on synthetic data (2) Up to 1200 spots per frame, 100% of particles were matched With a 10% error in calibration parameters and 1200 spots per frame, 95% of particles were matched Applications • • • • • Porous media Convective boundary layer Subduction Multi-dune channel Fully developed turbulent channel Mapping flow during retreating subduction Purpose • To model the large-scale mantle circulation induced by subduction of a laterally migrating slab in three-dimensional dynamically consistent laboratory models • To compare experimental results and numerical simulation outcomes • To predict the path of melted material, the distribution of geochemical anomalies, the formation of back-arc basins Introductions • The Earth's interior consists of rock and metal. It is made up of four main layers: – the inner core: a solid metal core made up of nickel and iron (1200 km diameter) – the outer core: a liquid molten core of nickel and iron – the mantle: dense and mostly solid silicate rock – the crust: thin silicate rock material Image of Earth and the interior layers Introductions (2) Plate tectonics is a theory of geology developed to explain the phenomenon of continental drift and is currently the theory accepted by the vast majority of scientists working in this area. In the theory of plate tectonics the outermost part of the Earth's interior is made up of two layers: the lithosphere comprising the crust and the solidified uppermost part of the mantle. Below the lithosphere lies the asthenosphere which comprises the inner viscous part of the mantle. Introductions (3) The lithosphere is broken into giant plates that fit around the globe like puzzle pieces. These puzzle pieces move a little bit each year as they slide on the asthenosphere. The asthenosphere is ductile and can be pushed and deformed like putty in response to the warmth of the Earth. Subduction is the process in which one plate is pushed downward beneath another plate into the underlying mantle when plates move towards each other. The plate that is denser will slide under the thicker, less dense plate. Experimental set-up: assumptions • Viscous rheology – the Earth system is simulated using viscous rheologies • Self-Consistent Subduction – Slab pull is the only active force within the system. This ensures that the experimental subduction process is a self-consistent response to the dynamic interaction between slab and mantle. • Convectively Neutral Mantle – Flow is generated only by subduction. Thermal convection and global or local background flow that is not generated by the plate/slab system are neglected. • Isothermal Experiments – Thermal effects during the subduction process are neglected. • No Overriding Plate – The overriding plate is not modeled. We assume the plate is completely surrounded by fault zones whose viscosity is the same as the upper mantle one. Experimental set-up: materials • Silicone putty (Rhodrosil Gomme, PBDMS+ iron fillers) and glucose syrup are used as analogue of the lithosphere and upper mantle, respectively. • Silicone putty is a viscoelastic material with purely viscous behavior at experimental strain rate (= 1480 kg/m3, = 3.6 105 Pa s) • Glucose syrup is a transparent Newtonian low-viscosity and highdensity fluid (fluid #1: = 1415 kg/m3, = 30 Pa s; fluid #2: = 1382 kg/m3, = 3 Pa s) • The bottom of the test section plays the rule of the 660 km discontinuity (interface between upper and lower mantle). Scaling factors: 1 cm in the laboratory model corresponds to 60 km in Nature 1 Myr in Nature corresponds to 1 min The experimental apparatus Camera (top view) Camera (lateral view) x y Plate width - 30 cm - 20 cm - 10 cm z Experimental set-up: procedure • Glucose syrup is seeded with neutrally buoyant, highly reflecting air microbubbles acting as passive tracers. Air bubbles negligibly influence density and viscosity of the mantle. • The subduction process is manually started in all the experiments by forcing downward the leading edge of the silicone plate into the glucose to a depth of 3 cm (180 km in nature). • Each experiment is monitored over its entire duration by two black and white progressive scan cameras imaging the lateral and top views. • Two neon lights produce a planar uniform radiation focused onto two normal cross-sections through the system. FT: Trajectories Fluid #2, plate width= 20 cm, lateral view FT: Trajectories Fluid #2, plate width= 20 cm, top view FT: Trajectories Fluid #2, plate width= 10 cm, top view FT: Velocity field Time-averaged velocity field - fluid #2, plate width= 20 cm - top view Results: evolution of subduction We observe that the subduction process evolves in three main stages: 1. initial transient sinking into the upper mantle 2. interaction with the 660 km discontinuity 3. steady state subduction regime Evolution of subduction: first stage At the beginning of the experiment, the trench retreats progressively accelerating and the slab dip increases. Both poloidal and toroidal advection cells can be Poloidal recognized in the cell velocity field since this initial transient stage. Images show mass exchange of mantle from the ocean to the wedge side of the plate. Evolution of subduction: first stage Top view Toroidal cell Evolution of subduction: second stage After the bottom of the test section is reached, the trench velocity significantly diminished while the tip of the slab folds and deforms in correspondence of the 660 km discontinuity. Mantle circulation slows down as well. Evolution of subduction: third stage Once the leading edge of the subducting plate has reached a stable arrangement at the bottom of the box, mantle circulation in a steady state regime establishes. The steady subduction velocity is a direct consequence of the constant slab pull force applied to the constant portion of subducted lithosphere. Less vigorous poloidal cell Results: influence of plate width •The subducting plate width (w) was changed preserving laterally unconstrained boundary conditions. The plate width strongly influences the subduction process. Increasing w from 10 cm to 30 cm, the trench velocity and consequently also mantle velocity decreases. The increase in w also strongly affects the vigor of mantle circulation. In particular, under the same density/viscosity mantle properties, a wider plate moves a larger amount of mantle material Conclusions • Our experiments confirm previous results in terms of subduction kinematics, identifying the presence of a typical sequence of stages in the kinematic evolution of a retreat subduction process: • the sinking of the slab into the upper mantle; • the interaction with the 660 discontinuity; • the steady state stage with the slab lying at the upper/lower mantle transition zone. • The dependence of the plate width and the subduction kinematics is confirmed. • Feature Tracking is a suitable technique to map and to quantitatively estimate the pattern of flow triggered in the mantle by subduction. • Rollback subduction generates a complex 3-D time dependent circulation pattern characterized by the presence of poloidal and toroidal components, both active since the beginning of the subduction process and evolving according to kinematic stages. Applications • • • • • Porous media Convective boundary layer Subduction Multi-dune channel Fully developed turbulent channel Turbulent mixing layer growth and internal waves formation: laboratory simulations Motivations and purposes The flux through the interface between the mixing layer and the stable layer plays a fundamental role in characterizing and forecasting the quality of water in stratified lakes and in the upper oceans, and the quality of air in the atmosphere. General aims of the investigation: • predicting mixing layer growth as a function of initial and boundary conditions • understanding the interaction between the mixing layer and the stable layer -> internal waves • describing the fate of a contaminant dissolved within the fluid phase Novelty of this contribution Enhanced equipment for measuring the fluid temperature providing a larger time resolution Large database of experiments run under several initial and boundary conditions Image analysis performed with the classical Particle Tracking (PTV) algorithm and the Feature Tracking (FT) algorithm Penetrative convection in the ABL z T T0 0 z y x T, Penetrative convection in lakes Z H EPILIMNIO T TERMOCLINO Entrainment (water flow, nutrients and contaminants) IPOLIMNIO TEMPERATURA Mixing layer visualization through LIF Experimental set-up Measuring techniques: velocity field • • • • Seeding the flow (100 m pollen particles); Illuminating the test section (500 W lamps); Acquiring images (2-CCD camera, 25 fps); Processing images through Classical Particle Tracking Velocimetry (PTV) and Feature Tracking (FT) Mixing layer: PTV (low seeding density) Mixing layer: FT (large seeding density) Internal waves: FT (large seeding density) Measuring techniques: Temperature Thermocouples are placed within the test section: - along a vertical line (to detect temperature profiles) - on the lower boundary (to test horizontal homogeneity) Features of a subset of experiments Experiment # 1 2 3 4 5 6 7 8 9 10 11 Tb0 (K) 288.15 287.05 287.15 288.39 285.15 286.66 285.15 293.76 294.32 294.00 287.54 TbC (K) 292.65 298.15 298.65 305.15 293.85 292.72 294.48 300.00 300.20 299.05 296.71 (T / z ) (K/m) 24.5 51.5 53.0 40.5 82.0 47.4 68.0 41.1 55.1 29.5 70.3 PTV OF X X X X X X X X X X X X X X X X X X Temperature profile before heating starts 0.20 0.18 0.16 Height (m) 0.14 0.12 Exp #1 Exp #2 Exp #3 Exp #4 Exp #5 Exp #6 Exp #7 Exp #8 Exp #9 Exp #10 Exp #11 0.10 0.08 0.06 0.04 0.02 0.00 282 284 286 288 290 292 294 Temperature (K) 296 298 300 302 304 Three methods for detecting the mixing layer height - Mean temperature profiles - Theoretical height according to the mean temperature - Velocity standard deviation profiles All methods assume horizontal homogeneity. The first two employ temperature data detected through the thermocouples. The last one employs velocity data reconstructed through PTV and OF. Mean temperature profiles for exp #3 z (T / z ) h(t) T0 0.12 2 min 0.1 6 min 10 min T(t) T 14 min 18 min Theoretical height according to the mean temperature in the mixing layer h(t ) 0 (T (t ) T0 ) Height (m) 0.08 22 min 0.06 26 min 30 min 0.04 34 min 0.02 0 282.0 283.0 284.0 285.0 286.0 287.0 288.0 Temperature (K) 289.0 290.0 291.0 292.0 Vertical velocity standard deviation for exp #3 0.16 0.25 min 2.25 min 4.25 min 6.25 min 8.25 min 10.25 min 14.25 min 18.25 min 24.25 min 30.25 min 0.14 Height (m) 0.12 0.1 0.08 0.06 0.04 0.02 0 0 0.0005 0.001 0.0015 sw (m/s) 0.002 0.0025 Height of the mixing layer for exp #3 Transilient matrix Dispersion phenomena occurring within the mixing layer are intrinsically non-local. An approach different than the ADE is needed Although the system is deterministic, the transilient matrix can be regarded as describing the set of probabilities that a tracer particle originates in one subvolume and ends up in another The transilient matrix is referred to as the conditional probability P(r,t|r’,t’) that a particle at position r’ at t’ ends up at position r at time t. Transilient matrix 6 i 5 4 3 2 1 j Cij Transilient matrix: physics destination FAST MIXING UPWARD Fast upward mixing SLOW MIXING UPWARD Slow upward mixing NO MIXING SLOW MIXING DOWNWARD FAST MIXING DOWNWARD source Scaling parameters w* g qs zi 3 Height of the mixing layer zi t* Convective velocity (g: acceleration of gravity, : thermal expansion coefficient, qs: surface kinematic heat flux) zi w* Convective time The phenomena occurring in the mixing layer are not steady: its height and mean temperature increase with time. Thus the scaling parameters are functions of time. In order to compare data not acquired simultaneously quasi-steadiness has to be assumed. Hence, after normalization, it can be considered time independent. Transilient matrix Transilient matrix Transilient matrix Transilient matrix Transilient matrix Transilient matrix Cross section through the transilient matrix Concentration begins as a function at the source depth, marked with a continuous line, and progressively more disperse curves correspond to later times. The dotted line marks the boundary of convection zone. Δt*= 0.01 Δt*= 0.05 Δt*= 0.07 Δt*= 0.08 Δt*= 0.12 Δt*= 0.16 Δt*= 0.3 Δt*= 0.5 Δt*= 0.7 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 0 0.5 1 Destination depth 1.5 Internal waves 1 0.8 0.6 0.4 0.2 0 -0.2 0 20 40 60 80 100 Time (s) -0.4 Correlation coefficient for w for experiment #2 T N g z 1 Experiment # N (s-1) (s-1) 1 0.35 0.33 2 0.30 0.26 2 Conclusions and future works The experimental apparatus described allows: • to reproduce a stable stratification; • to study heat exchanges; • to investigate the penetrative convection phenomenon; • to visualize internal waves and to compute their features; • to describe the dispersion phenomenon within the mixing layer through the Transilient Matrix. Once the transilient matrix is known for a given flow, a number of quantitative measures of nonlocal transport can be obtained from it directly. For instance matrix moments with respect to the destination indices at each source level or relative to the source indices can be computed Applications • • • • • Porous media Convective boundary layer Subduction Multi-dune channel Fully developed turbulent channel Preliminary Tests On Fluid-Dynamic Features and Plastic Separation Feasibility of a Hydraulic Separator Purpose • To contribute to the issue of plastic materials recycling because of environmental awareness, need to conserve materials and energy, and growing demand to increase production economy. • To reconstruct a laboratory model of a hydraulic separator for low specific mass plastic particles (1 g/cm3). • To test the capability of image analysis techniques to reconstruct the velocity field within the facility. • To set up of the best operative conditions for the apparatus. Purpose (2) • Chaotic advection regards the complex behaviour characterising a passive scalar due to Lagrangian dynamics of flow. The main issue related to chaotic advection is the enhancement of transport it produces. • Laminar flow can give rise to chaotic behaviour of Lagrangian particle trajectories even though the Eulerian velocity at any given point in space is fixed or periodic in time • Chaotic mixing and complex distribution of material can be produced in which nearby fluid elements diverge strongly from each other The experimental apparatus 300 12 4 15 C1 60 30 30 20 C3 C4 C5 C6 C7 C8 8 8 30 R1 R2 R3 R4 R5 R6 R7 R8 Longitudinal section (dimensions in mm) Flow out Flow in A multi-dune channel is a device constructed from a sequence of closed parallel cylindrical tubes welded together in plane. The complex is sliced down its lateral mid-plane and the lower half is shifted laterally and then fixed relative to the upper half. The experimental apparatus I8 O8 I7 O7 I6 O6 I5 O5 C1 C2 C3 C4 C5 C6 C7 200 C8 O4 I4 O3 I3 I2 O2 30 20 8 O1 I1 R1 R2 R3 R4 R5 R6 R7 R8 Upper and lower view of the apparatus (dimensions in mm) Flow out Flow in 300 Flow rate The multi-dune is filled through 4 distinct tanks each linked to two input nozzles (I1+I2, I3+I4, I5+I6, I7+I8). The experiments were run for five elevations of the tanks: 2.10 m (test series A), 2.30 m (test series B), 2.50 m (test series C), 3.00 m (test series D) and 3.50 m (test series E). TEST CASE HYDRAULIC HEAD AT THE INPUT NOZZLES (m) FLOW RATE (l/min) A 2.1 9.61 B 2.3 9.92 C 2.5 10.56 D 3.0 11.47 E 3.5 12.54 The acquisition system Green plastic powder (Ø=0.25 mm) preconditioned with a solution of water and sodium hydroxide to neutralize the electrostatic charge was used as a tracer. A high-speed camera allowed acquisition of 250 frames per second (spatial resolution 480420 pixels). A high wattage lamp illuminated a light sheet for image acquisition. Negative of an acquired image. The camera imaged chamber #3 FT – Trajectories: lower flow rate FT - Trajectories: larger flow rate Results – Velocity field Case A – Lower flow rate Case E – Larger flow rate Horizontal velocity component Results – Velocity field Case A – Lower flow rate Case E – Larger flow rate Vertical velocity component Results – Velocity field Case A – Lower flow rate Case E – Larger flow rate Horizontal velocity variance Results – Velocity field Case A – Lower flow rate Case E – Larger flow rate Vertical velocity variance Results - Streamlines 2: Sector Sector 3: 1: principal secondary vorticity transport flow. The vorticity zone. zone. IfIt a fluid thrust is moves proportional takes place where particle from to vertical the isflow the velocity principalvelocity component and in higher. A particle region to this region, conjunction with entering thethe flow gravity in it will have and buoyancy this region can back chance to come determines the destiny move one to the from previous of a particle. If the thrust camera to assuming another chamber, is larger thanflow the net the principal weight, thenot particle will thrust will prevent interact it to fall with out. the principal transport flow and, consequently, it will be displaced in the following chamber. Applications • • • • • Porous media Convective boundary layer Subduction Multi-dune channel Fully developed turbulent channel PTV for the characterization of turbulent channel flow: comparison of experimental and simulation approaches Objectives of this study • Image Analysis Techniques algorithms need to be tested with experiments of well-known flow properties. • Tests can be performed by analysing synthetically generated images or experimental images • At the present stage of development, are both PIV, PTV or FT useful for the study of near-wall turbulence? • A proper description of turbulent flows requires evaluating the mean velocity and the root-mean square (rms) of the fluctuating velocity together with the turbulence scales. The experimental set-up Turbulent channel flow (d= 2 cm, x/d = 80, z/d = 10) Tracers (p/f = 1.06, dP = 40 mm) FT: effects of parameter mindist FT: effects of parameter mindist Synthetic and real images Composition of 10 real images subtracting 24 pixels between consecutive ones Synthetic and real images (2) Composition of 10 synthetic images subtracting 24 pixels between consecutive ones DNS data 2 Mean velocity profile normalized by the wall-shear velocity y/h 1.5 1 0.5 u+ 0 0 5 10 15 20 2 y/h Turbulent intensities and Reynolds shear stresses normalized by the wall-shear velocity 1.5 u'+ v'+ 1 (u'v')+ 0.5 0 -1 0 1 2 u'+,v'+,(u'v')+ 3 PTV results Y/d 1 0.9 0.8 0.7 u+ - Re= 2900 v+ - Re= 2900 u+ Re= 5400 v+ Re= 5400 u - Kim 0.6 0.5 0.4 0.3 0.2 0.1 0 -5 0 5 10 15 u+ 20 Plot of the mean velocity as a function of depth PTV results 1 0.9 0.8 0.7 u'+ - Re=5400 v'+ - Re=5400 u'+ - Re=2900 v'+ - Re=2900 u'+ - Kim v'+ - Kim 0.6 Y/d 0.5 0.4 0.3 0.2 0.1 0 0 1 2 3 4 5 u'+, v'+ Plot of the turbulent intensities as a function of depth 6 PTV results 1 0.9 0.8 (u'v')+ - Re=2900 (u'v')+ - Re=5400 (u'v')+ - Kim 0.7 0.6 Y/d 0.5 0.4 0.3 0.2 0.1 0 -1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 (u'v')+ Plot of the Reynolds stress as a function of depth 1 FT results 2 1.8 1.6 1.4 u'+ CF v'+ CF (u'v')+ CF u'+ NCF v'+ NCF (u'v')+ NCF 1.2 y/h 1 0.8 0.6 0.4 0.2 0 -1 -0.5 0 0.5 1 1.5 2 2.5 3 u'+,v'+,(u'v')+ 3.5 4 Plot of turbulent intensities as a function of depth PIV results 2 1.8 1.6 1.4 1.2 u'+ v'+ (u'v')+ 1 y/h 0.8 0.6 0.4 0.2 0 -1 -0.5 0 0.5 1 1.5 2 2.5 u'+,v'+,(u'v')+ 3 Plot of turbulent intensities as a function of depth Results: Lagrangian framework Translient matrix at time t= 0.001 s Results: Lagrangian framework Translient matrix at time t= 0.017 s Results: Lagrangian framework Translient matrix at time t= 0.047 s Results: Lagrangian framework Translient matrix at time t= 0.058 s Conclusions • PTV is useful for studying near-wall turbulence both in the Eulerian and in the Lagrangian frameworks; • low sensitivity of the results on the image analysis “parameters”; • a larger acquisition field is required. Applications • • • • • Porous media Convective boundary layer Subduction Multi-dune channel Fully developed turbulent channel Application of 3D-PTV to Track Particles Moving Inside Heterogeneous Porous Media Motivations and outlines Flow within porous media is three-dimensional We are interested in describing the dispersion process in a Lagrangian framework 3-dimensional experimental techniques (3-D PTV) suitable for studying conservative tracer movement in porous formations homogeneous or heterogeneous at the bench scale have to be employed Simplification The simplification inherent in the laboratory model • Use of Pyrex particles of various dimensions and sizes to form a porous matrix • Use of glycerol for the fluid so that the matched-index technique is applicable • Use of air bubbles for the tracer that move passively with the flowing glycerol • Create the flow field with a hydraulic pump of constant mean, but variable velocity • Image only interior of the matrix so that boundary effects are minimized Simplification (2) Why choose 3-DPTV • Tracer dispersion phenomena are naturally described in a Lagrangian framework as the tracer acts as a tag of the fluid particles • 3-dimensional techniques (3-DPTV) suitable for studying conservative tracer movement in porous formations homogeneous or heterogeneous at the bench scale have to be employed Why choose air bubbles as tracer • They are passive • They have a substantially different refractive index that glycerol • They are easy to use and to re-use Scanning 3D-PTV – Experimental apparatus “Scanning” 3D-PTV Experimental Set-up for porous media Carefully calibrating we obtained the following results: 75% of trajectories where matched with tolerance=10 pixels Photogrammetric 3D-PTV- experimental apparatus Photogrammetric 3D-PTV Experimental set-up for porous media Creating heterogeneous media Het1 Sector2: =1 cm Sector3: =0.7 cm Sector1: =1 cm Het2 Sector2: =0.4 cm Sector3: =1 cm Sector1: =0.7 cm (a) (b) Positioning of the heterogeneous medium inside the column through a PVC shaper with three sectors Trajectories reconstructed XZ Plane YZ Plane How can we describe particle dispersion? The effects of heterogeneity on the dispersive process are studied by examining the evolution of several scattering functions. For a time stationary velocity field, the self-part of the intermediate scattering function Gs is defined as 1 Gs ( x, t ) N N [x ( X (t ) X (0))] i 1 i i The relative scattering function, Gr, compares the growth of the separations of particles from an initial distribution. 1 Gr (x, t ) (x ( X i (t t0i ) X j (t t0 j ) N t ( N t 1) i j Results: (b) (a) (c) Gs (self part of the intermediate scattering function) for Hom2 at 3, 13, and 26 seconds, (a), (b), and (c) respectively Initially the spread about the mean for the short time scale is greater in the transverse direction, but subsequently the spread of particle displacement is greater in the longitudinal direction. The center of mass in the longitudinal direction at early times remains near wt, but by 13 s it shows a significant divergence from wt (w is the longitudinal mean velocity) Results: The relative scattering function is an indication of the internal mixing between particles. If the growth of Gr for one system is less than another with similar velocity variances, then there must be more internal mixing in the system with little growth in Gr. Het1 and Het2 are an example of this, indicating more internal mixing for particles in Het1. (a) (b) Gr (relative scattering function) for Het1 (a) and Het2 (b) at times 3, 6 and 9 seconds Conclusions: •It is possible to reconstruct 3D trajectories through matched refractive index 3D-PTV. •From these trajectories we can compute: mean square displacements, velocity covariances, classical dispersion tensor, intermediate scattering function and generalized dispersion tensor. •The mean displacement square root clearly highlights the different behavior of homogeneous and heterogeneous media, even if it can not be used to quantitatively measure the dispersion coefficient. •The classical dispersion coefficient formulation is valid only when the hypothesis of long-time and large-distance is satisfied. •The Lyapunov exponent theory allows the dispersion process to be described through the "doubling time" at a certain scale of couples of particles belonging to the cloud. Applications • • • • • Porous media Convective boundary layer Subduction Multi-dune channel Fully developed turbulent channel