UNIVERSITY OF TARTU FACULTY OF SCIENCE AND TECHNOLOGY Institute of Physics

Gert Toming

DEVELOPMENT AND CHARACTERIZATION OF A HYDRODYNAMIC TEST BED AND A DIGITAL PARTICLE IMAGE VELOCIMETRY SYSTEM
Masters thesis (30 EAP)

Supervisors:

Prof. Maarja Kruusmaa, TU senior researcher Andres Hunt, TTU Ph.D student Ph.D Madis Listak, TTU researcher

Tartu 2010

TABLE OF CONTENTS
INTRODUCTION ...................................................................................................................... 4 FILOSE ...................................................................................................................................... 5 TASK SPECIFICATION ........................................................................................................... 6 1. PARTICLE IMAGE VELOCIMETRY (PIV) ................................................................... 7 1.1. 1.2. Introduction to PIV ...................................................................................................... 7 Tracer particles ............................................................................................................ 9 Particle properties and their motion in fluid......................................................... 9 Seeding the particles ........................................................................................... 11

1.2.1. 1.2.2. 1.3.

PIV hardware ............................................................................................................. 12 Light source ........................................................................................................ 12 Light sheet optics ............................................................................................... 13 Imaging and camera ........................................................................................... 14

1.3.1. 1.3.2. 1.3.3. 1.4. 2.

Image processing ....................................................................................................... 15

AVAILABLE DPIV SYSTEMS AND SOFTWARE ..................................................... 18 2.1. DPIV system in the University of Bath ..................................................................... 18 Laser ................................................................................................................... 19 Camera ............................................................................................................... 20 Software ............................................................................................................. 21

2.1.1. 2.1.2. 2.1.3. 2.2. 3.

Open source PIV software ......................................................................................... 22

CREATING THE FLOW TUNNEL AND THE DPIV SYSTEM .................................. 24 3.1. Development of the flow tunnel ................................................................................ 24 The initial version of the flow tunnel ................................................................. 24 The first rebuilding step of the tunnel ................................................................ 25 The second rebuilding phase of the tunnel ......................................................... 26

3.1.1. 3.1.2. 3.1.3. 3.2.

The DPIV system development ................................................................................. 29 Camera ............................................................................................................... 29 Laser ................................................................................................................... 30 Software ............................................................................................................. 32 2

3.2.1. 3.2.2. 3.2.3.

4.

RESULTS......................................................................................................................... 34 4.1. 4.2. 4.3. 4.4. Measuring the flow speed .......................................................................................... 34 Flow profile ............................................................................................................... 38 Studying fish swimming patterns .............................................................................. 40 Krmn vortex street ................................................................................................. 42

5. 6.

DISCUSSION .................................................................................................................. 44 FUTURE WORK ............................................................................................................. 45

SUMMARY ............................................................................................................................. 46 REFERENCES ......................................................................................................................... 47 KOKKUVTE ......................................................................................................................... 49 APPENDIX .............................................................................................................................. 51 Appendix A .......................................................................................................................... 51 Appendix B .......................................................................................................................... 53 Appendix C .......................................................................................................................... 54

INTRODUCTION
Particle image velocimetry (PIV) is an optical fluid visualization method for capturing velocity information in the region of interest just within a fraction of a second. The fluid, gaseous or liquid, is seeded with tracer particles and their motion is then recorded to calculate the velocity field. PIV has evolved together with digital computers and cameras introducing the new term, digital particle image velocimetry (DPIV). This work describes the development process of a DPIV system and a controlled hydrodynamic test environment in the Centre for Biorobotics. This equipment is mostly designed for the experimental work within an EU 7th Framework project FILOSE (FIsh LOcomotion and SEnsing). The urgent need to build a custom system for fluid field velocimetry on site arose as there were no flow tunnels commercially available and the cost of the commercial DPIV systems was unacceptably high. The rest of this work is organized as follows: in the first two chapters we give an overview of the basic principles of PIV and present a detailed description of one commercial DPIV solution as an example. In the third chapter, the design and building steps of a custom made flow tunnel and DPIV system is given. This includes the description and analysis of the two main stages of the flow tunnel development, description of the final version, and the description of the process of component selection. The developed system is experimentally verified by measuring the flow speed and studying the flow profile in Chapter 4. In this chapter experiments made for FILOSE is described as an example of the system is use. Finally the conclusion of the work and summary are given.

FILOSE
The work presented in the thesis is part of the project FILOSE (robotic FIsh LOcomotion and Sensing) that is an European Seventh Framework Programme research project in the field of Information and Communication Technologies. The project tasks are divided between five partners: Tallinn University of Technology (the project coordinator), University of Verona, Italian Institute of Technology, University of Bath, and Riga Technical University. The aim of this project is to develop technologies to achieve a breakthrough in the underwater robotics that is a growing field in science and industry. Current commercial underwater robots are robust, clumsy and inefficient in terms of energy consumption, especially when placed in unstable environments (such as turbulences, Krmn street, etc). They mostly use propellers for propulsion, which create lots of turbulence and beat up mud. This project aims to create a bio-inspired robot that could sense the flow and objects surrounding it as if a real fish would do, and adjust its behavior according to it. A quietly moving robot could be used in sea-life monitoring in shallow water and for many other purposes both in research and industry. In the work package of the Centre for Biorobotics (Tallinn University of Technology) it is necessary to set up a controlled hydrodynamic environment and DPIV system to: analyze hydrodynamic events; identify flow patterns associated with specific locomotion patterns; benchmark the efficiency, swimming speed and maneuverability of robotic fish prototypes. The result of the current work is intended to satisfy the initial technical needs to proceed with the work required in the Centre for Biorobotics.

TASK SPECIFICATION
A DPIV system and a controlled hydrodynamic test environment had to meet the following initial requirements: the test environment is a flow tunnel with user selectable laminar flow speed (up to 1 m/s); the cross section of the tunnel must be at least 500500 mm and 1500 mm in length; as a significant constraint to dimensions, it has to fit into the existing 41,51,45 m water tank; the DPIV system must be of low cost and thus has to be mostly custom made; with the DPIV system it must be possible to measure the flow speed and quantify the flow field in the near vicinity of the test object.

1. PARTICLE IMAGE VELOCIMETRY (PIV)

1.1. Introduction to PIV

Most common fluids, such as air and water, are homogeneous and optically transparent. Thus their motion is difficult or impossible to characterize as it is not readily visible. A great step forward in the investigation of fluid flows was made just after it was possible to visualize them. It allowed to conduct experiments to extract information about it. A great portion of the seminal research in the field was done by Ludwig Prandtl a German scientist who was first to design and utilize flow visualization techniques in a water tunnel in order to study unsteady separated flows behind various objects, such as wings [1]. Since then, many techniques have been implemented to measure flows including laserDoppler anemometry (LDA) and hot-wire anemometry [2]. However, most of them are single-point methods and provide accurate quantitative measurement of the flow only at some given point. Also, it is possible to mark the flow with smoke, dye or small particles to visualize it. Flow visualization techniques like PIV provide two-dimensional velocity field, however early flow visualization methods were prevalently qualitative. In more detail, PIV is an optical method of fluid visualization using small seeding particles [1, 2]. The principle of the PIV is based on the measurement of the displacement of small tracer particles. The particles are sufficiently small to accurately follow the fluid motion. They are illuminated with a thin light sheet and the light scattered from them is stored in one or two subsequent image frames with a known interval. Nowadays, typically a CCD camera is used for that purpose. With single camera systems it is possible to get only two velocity components on the measurement plane. In order to determine all three components of the flow velocity, the stereoscopic imaging must be used [2, 3]. The schematic of a typical stereo PIV measurement system can be seen in Figure 1.

Figure 1. Principle schematic of a typical stereo PIV measurement system. [2]

The recorded images are processed on a computer off line. This consists essentially of cross correlation analysis of the particle image patterns in small sub-domains, or interrogation regions, between the subsequent image frames. The particle-image pattern displacement divided by the image magnification and the time delay between the two image frames yields the local fluid velocity. This process is repeated for the entire image domain yielding in an instantaneous velocity field in a planar section of the observed flow. The PIV analysis typically yields 103 105 velocity measurements per image (depending on the image format and interrogation resolution) with a typical measurement accuracy of 1% [2]. All in all, PIV has been a major breakthrough in experimental methods in fluid mechanics and has made it possible to experimentally investigate aspects of flows that had previously been impossible.

1.2.

Tracer particles

1.2.1. Particle properties and their motion in fluid It is clear from the principle of PIV that this technique is based on the determination of the two fundamental quantities used to calculate the velocity: differentials of time and length. The measurement technique is indirect as it determines the particle velocity instead of the fluid velocity. Therefore, mechanical properties of the tracer particles have to be studied in order to avoid significant discrepancies between fluid and DPIV tracer particle motion. It has been assured that choosing neutrally buoyant tracer particles leads to that these particles accurately follow the flow [1, 2]. Also the light scattering characteristics of the particles must be considered. This aspect along with the fact that the tracers must be as small as possible, imposes conflicting demands: small particles cause smaller flow distortion than large particles while large particles scatter light more efficiently [3, 4]. Since the obtained images particle intensity and therefore the contrast of the PIV recordings is directly proportional to the scattered light intensity, it is often more effective and economical to increase the image intensity by properly choosing the scattering particle size instead of by increasing the laser power. The light scattered by particles is a function of the ratio of the refractive index of the particles to that of the surrounding medium, the particle size, their shape and orientation. The scattering also depends on light polarization and observation angle. Since the refractive index of water is considerably larger than that of the air, the light scattering of particles in air is at least one order of magnitude larger compared to particles of same size in water [1]. Therefore, much larger particles have to be used in water, which is acceptable since matching the densities of the particles with fluid is usually easy. In most experiments it is desired that the seeding particles were non-toxic, non-corrosive, non-abrasive, non-volatile and chemically inert. A wide variety of seeding materials matching these requirements are available. Examples of seeding particles and their properties, for liquids and gases, are shown in Tables 1 and 2 respectively. The reflectivity of the particles can be improved by means of metal coating for example, metal coated hollow glass spheres are often used in practice. Silver coated hollow glass spheres of approximately 10m diameter are shown in the Figure 2. [2, 3, 4] 9

Table 1. Seeding materials for liquid flow visualization [1]. Type Solid Material Polystyrene Aluminum flakes Hollow glass spheres Granules for synthetic coatings Liquid Gaseous Different oils Oxygen bubbles Mean diameter in m 10 100 27 10 100 10 500 50 500 50 - 1000

Table 2. Seeding materials for gas flow visualization [1]. Type Solid Material Polystyrene Alumina Al2O3 Titania Ti02 Glass micro-spheres Glass micro-balloons Granules for synthetic coating Smoke Different oils Liquid Di-ethyl-hexyl-sebacate (DEHS) Helium filled soap bubbles Mean diameter in m 0,5 10 0,2 5 0,1 5 0,2 3 30 100 10 50 <1 0,5 10 0,5 1,5 1000 - 3000

Figure 2. Micrographs of silver coated hollow glass spheres as seen at the magnification of 500 (left) and 5000 (right). [1] 10

Tracer particles can also be labeled with a fluorescent dye (e.g. Rhodamine). The dye absorbs the incident light which is then re-emitted at longer wavelengths. For example, when green light (532 nm) of a laser illuminates Rhodamine-covered particles, the particles not only scatter the laser light, but also emit orange-red (550-680 nm) light. Optical filters make it possible to separate tracer particles from other scattering objects, such as walls, bubbles, droplets and other particles by blocking the light with the wavelength of the incident light and passing the fluorescent [2].

1.2.2. Seeding the particles Sometimes, seeding the particles into fluid can be done very easily or does not have to be done at all. Natural seeding is the situation when the fluid has enough visible particles naturally present and further seeding is not needed. In almost all cases, it is desirable to add tracers in order to achieve sufficient image contrast. However, achieving the optimum flow seeding is generally acknowledged as the most difficult part of the PIV experiments, but once the flow is uniformly seeded with the correct concentration, the experiment has a high chance of success. For high quality PIV records, the scattering particle concentration of about 15 particles per interrogation volume is necessary [2, 4]. Since PIV experiments in liquid flows are almost always conducted in closed-circuit test loops, seeding can be done easily by suspending solid particles into the fluid and mixing them in order to ensure homogeneous distribution. The best practice is to increase the seeding concentration gradually from a low level until the desired one is reached. Dilutions up to 1:50000 are feasible [2]. In gas flows, the increased difference in density between the fluid and the particles can result in a significant velocity lag. Often the particles must be injected into the flow shortly before the gaseous medium enters the test section. The injection has to be done in the way that the flow is not significantly disturbed, but at the location and in a way that ensures homogeneous distribution of the tracer particles. Since the existing turbulence may not be strong enough to mix the fluid and particles sufficiently, the particles have to be supplied from a large number of openings. Health considerations are also important since the experimentalists may inhale seeded air, for example in wind tunnels with an open test section. [1, 2, 4]

11

1.3.

PIV hardware

1.3.1. Light source The measurement principle of PIV relies on the finite observation time used to detect the particle motion. The particles need to be illuminated and observed twice with a time separation between the observations. There are three main requirements for the light source suitable for PIV experiments: the short duration of illumination the practical criterion for the choice of the maximum duration of illumination is that the particle images do not appear as streaks but rather as circular dots. As an alternative, it is also possible to use a continuous light source when a high-speed camera is used and its image exposure time is short enough to avoid particle images appearing as streaks; the illuminated particles must be distributed within a thin light sheet such that they can be imaged in focus. Their position in depth is dictated by the light sheet thickness; the intensity of the light must allow the light scattered from the seeding particles to be detected by the digital imaging devices [2]. Lasers are widely used as illumination sources since they can produce pulsed, collimated and monochromatic light beam that can be easily shaped into a thin light sheet. The required pulse energy E is proportional to the linear dimensions of the area of interest. Typically, 100 mJ impulse is sufficient to illuminate a square plane with a side length of 10 cm in air, while about twice as big area is possible to be illuminated in water flows. The most common device used for PIV experiments is the solid-state frequency-doubled Neodym-YAG lasers ( = 1064 nm and = 532 nm). These usually can produce energies ranging from 10 mJ up to 1 J with very short pulse duration (515) ns [1, 2]. A white light source might also be used, although it cannot be collimated as well as monochromatic light. The spectral outputs of sources like Xenon lamps are well suited for use with CCD cameras because of their similar spectrum and spectral sensitivity. Systems are commercially available that can easily be triggered and offer a repetition rate that matches the video rate. The main advantage of the white light source is that they are significantly cheaper than lasers and their application is not hampered by laser safety rules [1].

12

1.3.2. Light sheet optics The circular-cross-section beam delivered by the light source is shaped into a thin sheet using cylindrical and spherical lenses. When using lasers with a sufficiently small beam diameter and divergence like argon-ion lasers, one cylindrical lens may be sufficient to generate a light sheet of appropriate shape. For other light sources like Neodym:YAG lasers, a combination of different lenses is usually required in order to generate thin light sheets of high intensity [1, 2]. Two common lens arrangements used to shape laser beam into appropriate shape are shown in a Figure 3. In the first case (a), a cylindrical lens expands the beam along the vertical direction and in the horizontal direction it has no effect, the spherical lens will focus it at a distance equal to the focal length to make it thin enough in the measurement region.

Figure 3. Light sheet formation optics: (a) constant sheet width. (b) linearly expanding sheet width. The Rayleigh length Ra is the length over which the light sheet has a minimum thickness [2].

However, method (a) is not practical when large regions of space have to be illuminated, unless a large spherical lens is available. The second scheme (b) is used for the illumination of large areas. In this case, the desired sheet width can be obtained at an appropriate distance L as a function of the focal length of the cylindrical lens f1 [2]. Care should be taken to avoid reflections that can cause damage to other optical components. In most cases, this can easily be avoided by the right orientation of the lenses so that the light could not reflect back to the source, as demonstrated in Figure 4 [1].

13

Figure 4. Orientation of the lenses. Not recommended: light can reflect directly back to the source.

Recommended: light could not reflect directly back to the source. [1]

1.3.3. Imaging and camera An image of the tracer particles in the light sheet is formed using a lens in front of the image sensor. Nowadays, usually a digital camera with CCD or CMOS sensor is used. Comparing CCD and CMOS sensors, the image quality of the state-of-the-art CCD cameras is better than in case of the CMOS cameras [6]. Thus, for high precision measurements a CCD camera should be used. Intensified CMOS cameras further reduce the image quality and spatial resolution. However, their high sensitivity allows measurements at low illuminations with moderate signal-to-noise ratios [6]. As with every other electronic device, the digital image sensors are subject to electronic noise. In case of PIV, the light scattered from small particles should ideally be captured on the otherwise black background. Due to the limited light scattering efficiency of tiny tracer particles, the recorded signal sometimes only barely exceeds the background noise level of the sensor as the observation area or the observation distance is increased [1]. The major source of the noise is the thermal effect, which also generates electron-hole pairs in image sensors that cannot be distinguished from those generated by the laser light scattering caused photoelectric effects. As a result, weak particle images can no longer be distinguished from noise. Since the production rate of the electron-hole pairs is constant at a given temperature and exposure, this dark current or dark count can be accounted with by subtracting a constant bias voltage at the output of the charge-to-voltage converter. The rate of generation doubles for every (67) C increase in temperature, which is the primary motivation to use the cooled sensors in scientific imaging [1]. Another source of noise is the read noise or shot noise which is a direct consequence of the charge-to-voltage conversion 14

during the readout sequence the readout noise increases with the readout frequency. A careful optimization of the conversion electronics, the reduced readout frequency as well as cooling the sensor, may limit the read noise to just a few electrons RMS per pixel [1]. Up to now, the prohibitive cost of these specialized cameras has made their use in PIV recording unfeasible.

1.4.

Image processing

In order to extract the displacement information from a PIV recording, some sort of interrogation scheme is required. Initially, this interrogation was performed manually on selected images with relatively sparse seeding. This allowed to track individual particles. With computers and image processing becoming more common in the laboratory environment, it became possible to automate this process. However, tracking an individual tracer particle in the images from exposure to exposure is only practicable in the low particle density case [1, 2]. In principle, a high data density is required on the PIV vector maps, especially in the comparison of experimental data with the results of numerical calculations. This requires medium concentration of the tracer particles in the PIV recording images. Medium concentration is characterized by the fact that the matching pairs of particle images in subsequent images cannot be detected by visual inspection of the PIV recording. Hence, statistical approaches had to be developed [1, 7]. The main objective of the statistical evaluation of the PIV recordings at medium particle density in images is to determine the displacement between two patterns of particle images, stored as 2D distributions of gray levels. In practice, this is accomplished by extracting small samples or interrogation windows and analyzing them statistically [1]. From a signal (image) processing point of view, the first image may be considered as the input to the system whose output produces the second image of the pair as shown in the diagram in Figure 5. The systems transfer function, H, converts the input image I to the output image I and is comprised of the displacement function d and additive noise process, N. The function of interest is shift by the vector d as it is responsible for displacing the particle images from one image to another. This function can be described, for instance, by a convolution with . The additive noise process N models the effects due to recording noise and threedimensional flow. If both d and N were known, the every next image could be calculated from 15

the previous. In the current case, with both images I and I known, the aim is to estimate the displacement field d while excluding the effects of the noise process N. The fact that the signals are not continuous the dark background cannot provide any displacement information makes it necessary to estimate the displacement function d using a statistical approach based on localized interrogation windows [1].

Figure 5. Idealized linear digital signal processing model describing the functional relationship between the successively recorded image frames. [1]

The best match between the images is found using the discrete cross-correlation function (Formula 1): , (1)

where variables I and I are the samples (e.g. intensity value matrices) extracted from the images where I is larger than the template I. The template I is linearly shifted around in the sample I without extending over edges. For each choice of sample shift (x, y), the sum of the sum of products of all overlapping pixel intensities produces one cross-correlation value . By applying this operation for the entire range of shifts, a correlation plane is

formed, as shown in Figure 6. The highest value in the correlation plane can then be used as a direct estimate of the particle image displacement. Using the known time between the two PIV recordings and particle image displacement, the velocity can be calculated [1, 7].

16

Figure 6. Formation of the correlation plane by direct cross correlation. Here, a 44 pixel template is correlated with a 88 sample. [1]

Figure 7. Histogram analysis for finding the most probable

displacement. [2]

17

2. AVAILABLE DPIV SYSTEMS AND SOFTWARE

There are some standard PIV systems commercially available, and several companies also offer to build customized systems. The price range for these systems, starts approximately from 1 000 000 Estonian kroons. Here, a short overview of the three leading companies in the field is given: TSI Incorporated is an American company with branch offices also in some Asian countries and in larger European countries, including Germany, France and United Kingdom. They offer a variety of measurement, diagnostics, monitoring and control systems, including PIV, micro PIV, stereo PIV and others [8]. The PIV systems are put together from different cameras and lasers according to the clients needs. LaVision is a German company that offers imaging software, high speed cameras and standard or custom-designed laser imaging systems for reactive and non-reactive flow field analysis [9]. They also offer smart optical sensor systems and intelligent imaging systems for material testing. Dantec Dynamics is a Danish company producing integrated measurement systems for diagnostics and research in the fields of fluid mechanics, solid mechanics, microfluids, spray analysis and combustion technology [10]. They also offer a variety of complete PIV systems.

2.1.

DPIV system in the University of Bath

The Ocean Technologies Laboratory in the University of Bath is a FILOSE project partner of the Centre for Biorobotics and they have acquired a commercial two-dimensional DPIV system. This is made by TSI Incorporated, according to the needs specified by the laboratory staff. The author had a possibility to participate in the DPIV training carried out by the manufacturing company representatives in the University of Bath. The training included the usage of the Insight 3G software, the high speed camera pco. 1200 hs and the laser (Excel) as well as the laser safety issues.

18

2.1.1. Laser The laser included in the DPIV measurement kit is manufactured by the company Laser Quantum. Laser model is Excel and it is a high power class IV laser with a green beam [11]. Detailed specifications of the laser are given in Table 3. Excel has its own separate power supply smd6000, which is controllable through an RS232 connection. This makes it possible to control the laser from a computer or from a separate remote control [12]. The system in Bath has a separate control unit with functions of switching the laser on and off using a special key, and controlling the laser power. The laser optics creates a linearly expanding light sheet as shown in Figure 3 (b). Table 3. Laser specifications [11]. Parameter Power Wavelength Beam size Bandwidth Divergence Power stability Polarization ratio Polarization direction Value 2W 532 nm 1,8 mm 30 GHz 0,5 mrad < 3,0 % RMS 100:1 Horizontal

As Excel is a high power class IV laser, it can damage skin and cause permanent eye damage as a result of looking into direct or diffused beam. All of the staff of the Ocean Technologies Laboratory was instructed regarding the laser safety [13]. The laser working area is fully isolated by the blackout curtains and the staff working in the area must wear the safety goggles. All of the laboratory doors have clearly visible laser warning signs and lights. When the laser is turned on, all the doors must be closed and locked, and warning lights must be switched on.

19

2.1.2. Camera The camera used for DPIV is pco. 1200 hs (Figure 8). This is a high speed 10 bit CMOS camera with an integrated image memory. It enables fast image recording at the speed of 1 GB/s and with the maximum frame rate of 1357 FPS [14]. The image data is transferred to a computer after the experiments via IEEE 1394a (Firewire) connection. Specifications of the camera are in more detail shown in Table 4.

Figure 8. pco. 1200 hs camera [14].

The camera uses Nikon standard f-mount or c-mount optical input, which makes it possible to use a large variety of lenses. In the University of Bath, a AF-S Micro Nikkor 60 mm lens with the aperture of f/2,8 f/32 and the closest focusing distance of 18,5 cm, is mounted to the camera [15]. Table 4. Specifications of the pco. 1200 hs camera [14]. Parameter Resolution Pixel size Sensor format/diagonal Frame rate Spectral range Exposure time Interframing time (PIV mode) Data interface ns Unit Pixel m2 mm2/mm Frames per second Nm Value 12801024 1212 15,3612,29/19,67 636 at full frame; 1357 at ROI VGA 2901100 1 s1 s 70 IEEE 1394a, camera link 20

2.1.3. Software Insight 3G is a Microsoft Windows XP based data acquisition, analysis and display software platform from the TSI Incorporated. It is developed for: measuring microflows (using a microPIV system); time-resolved measurements of flow fields (using high frame rate cameras and high pulse rate lasers); global measurement of scalar properties (using Planar Laser Induced Fluorescence (PLIF) and special cameras); simultaneous velocity and scalar property (temperature, concentration, species)

measurements (using same type of cameras or using different types of cameras to capture particle images and global fluorescence fields); global spray diagnostics measurements.

Insight 3G allows the setup, control and operating of the entire diagnostics system and provides an on-line or batch-mode analysis of the image fields with detailed display of the results. Softwares graphical user interface is shown in Figure 9. For more sophisticated data analysis, it is possible to involve also MATLAB, which is included with the Insight 3G. For better data representation, there is a Tecplot add-on, however this software must be installed separately.

Figure 9. The user interface of Insight 3G [15]. 21

2.2.

Open source PIV software

Due to the high cost of the commercial DPIV systems it was decided to use freeware and an already purchased software (the latter include LabVIEW 8.6 and Matlab R2009a). Fortunately, several open source implementations of PIV software for different platforms are available. Here, a short overview of some of them is given. URAPIV Matlab URAPIV is an open source Matlab toolbox for particle image velocimetry analysis for fluid mechanics [17]. It is platform-independent and needs only preinstalled Matlab software to run. However, the software crashed continuously and it was impossible to test it in details. URAPIV Python (PyPIV) PyPIV is a Python version of URAPIV and it needs Python and Enthought Tool Suite to be preinstalled and running [17]. PyPIV calculates a particle image displacement field in pixels from two image files and writes it to a comma separated value (*.csv) data file. URAPIV-C++ URAPIV-C++ is a 32 bit Windows executable software [17]. It is possible to import two separate images or two image lists (for frame A and B) and the software accepts many image formats. The output data is written into a 5 column, tab-delimited text (*.dat) file. The columns are organized in the file as [x y u v Q], where: x - is the horizontal distance in pixels from the lower-left corner; y - is the vertical distance in pixels from the lower-left corner; u - is the estimate of the horizontal pixel displacement at (x,y); v - is the estimate of the vertical pixel displacement at (x,y); Q - is the quality of the signal: 0 if pass, 1 if failed.

As a serious drawback, the program is rather slow and it fails to work on platforms newer than Windows XP.

22

JPIV JPIV is a software package for PIV written in Java. The program is open source, platform independent, and available under the terms of GNU general public license [18]. Like URAPIV, also the JPIV gives a particle image displacement in pixels rather than the actual speed. All the source code is well documented and it is relatively easy to modify it when needed. mpiv mpiv is a PIV toolbox written in Matlab script [19]. It requires Matlab version 6.0 or newer, Matlab Image Processing Toolbox, Design and Analysis of Computer Experiments Toolbox and Matlab C/C++ Compiler version 2 or later. mpiv consists of two programs one for image processing and the other for data post-processing. The main purpose of this toolbox is educational. It is portable, easy to use and easily modifiable. The developers intention is to keep the codes short, simple and robust. MatPIV MatPIV, with a current version 1.6.1, is a PIV toolbox running in Matlab environment [20]. Its source code is available under the terms of GNU general public license, which means it is allowed to extend and modify it. In MatPIV it is possible to define the coordinate system unit length in order to know how far the particles move in the physical coordinates (centimeters), instead of commonly used camera coordinates (pixels). It is also possible to define the time delay value between the two subsequent frames. As a result of these improvements, the output of the program is velocity (cm/s), not just particle image displacement field in pixels, as it is in other tested open source software. It is also easy to change other important configuration parameters, such as interrogation window size, overlapping area etc, or mask out regions. Due to its best functionality, stable and relatively fast processing, this software package was chosen as the best suitable for data processing in our current and further work.

23

3. CREATING THE FLOW TUNNEL AND THE DPIV SYSTEM

The work plan of the FILOSE project required conducting DPIV experiments also in Centre for Biorobotics. This introduced the need for a specific controlled hydrodynamic testing environment. As it was mentioned before, there are no commercial solutions for flow tunnels available suitable for our needs. The building process of the test bed for the Centre for Biorobotics started from a custom-made existing piece of equipment which, however, had never been used before, and in testing, appeared to function much below the stated minimum requirements. The flow tunnel development process consisted of two redesigning and rebuilding steps. In parallel to flow tunnel development, also a DPIV system was constructed and used to evaluate the flow tunnel design. In the following, an overview of the building process of the flow tunnel and the DPIV system is given. For the exact requirements and constraints, please refer to the Section Task Specification above.

3.1.

Development of the flow tunnel

3.1.1. The initial version of the flow tunnel

In Centre for Biorobotics, there was a large water tank with dimensions of 41,51,45 m for testing underwater robots. As there were no other under-water robotics projects beside FILOSE at that time, the tank was free to take advantage of for the flow tunnel design. Building the flow tunnel inside the water tank is technically much easier as the minor leakage from the tunnel is not an issue. This saved a lot of resources in the initial development phase. Prior to the systems redesigning, there was already a 1,5 m long tunnel with a cross section of 0,50,5 m (Figure 10. A) present in the tank. The tunnel was made of plexiglas with a stainless steel frame. The flow was created using an external pump. During initial testing it appeared that this system does not match the requirements due to its low maximal flow speed (20 cm/s), high turbulence and poor visibility through plexiglas. Thus, the flow tunnel had to be re-designed and rebuilt. The process took place step-by-step and a short overview of the improvement steps is given in the following subchapters.

24

A .

Figure 10. Building steps of the flow tunnel. A the first version of flow tunnel. B the second version with a wider pipe and a new motor. C the third version a closed loop flow system with high speed laminar flow.

3.1.2. The first rebuilding step of the tunnel The first goal of the redesign was to achieve an increase in flow speed. For this, the tunnel input pipe was replaced with a plastic drain pipe with a diameter of 320 mm. To create a better water circulation within the limits set by the surrounding water tank, the input pipe was built in C-shape as shown in Figure 10. B. The low power external pump was replaced with a Neptun 23 outboard motor used to power boats. It was placed into the input pipe. Its gasoline powerhead was replaced with a 3-phase 4 kW AC electric motor and the motor speed was controlled using a Mitsubishi F700 AC frequency converter. Along with the replacement of 25

the propulsion unit, also the flow direction was changed so that water could enter the tunnel from the open area where it was rather steady during the operation. This improvement helped to reduce the turbulence significantly. With this new configuration, the flow speed up to 60 cm/s was reached. This however, still did not match the initial requirements.

3.1.3. The second rebuilding phase of the tunnel As the third and currently final version of the system, a fully closed flow tunnel was built with a waterproof box (for camera and other equipment) fixed directly on top of it (Figure 10. C and Figure 11. A). The closed flow circle has two main advantages comparing to the open one: first, all the flow generated by the propeller remains inside the tunnel. This means that to achieve the same flow speed both in an open and a closed system, much less power is needed in a closed one; second, the pump does not create turbidity outside the flow tunnel. To get an equal flow speed across the entire tunnel cross section, dividing walls (flow guides) had to be inserted into the sharp curve before the working section of the tunnel. Also, after the circular pipe, a custom made aluminum flow spreader was installed. Finally, a laminator was built and inserted into the tunnel directly before the working section (Figure 11. B). The laminator is made of 10 mm Lexan Thermoclear Plus 2UV sheet (polycarbonate) roof material and is 20 cm thick. After passing this 20 cm long stack of pipes, the flow is laminar without any vortexes. A . B .

Figure 11. Flow tunnel version 3. A the currently final version of the flow tunnel. B a part of the flow tunnel with the 200 mm thick laminator.

26

To get the correct parameters (Figure 12 A and B) for flow spreader and a suitable position for the dividing walls (Figure 12. C), some simulations were conducted with Solid Works 2010 Flow Simulation software. In order to reduce the computing time to reasonable, the model of the laminator was defined with 2020 mm apertures and 5 mm thick walls between them. A . B .

C .

Figure 12. Flow tunnel third version construction. A 3D model of the flow spreader. B a photo of the flow spreader. C the part of the flow tunnel with dividing walls (flow guides).

27

In Figure 13, the simulation results of the entire flow tunnel are shown. Both, turbulent flow in a pipe and laminar flow in the flow fume can be seen. Also, a velocity distribution over the cross section of the flow tunnel is seen in the image. Since our test equipment is always placed in the middle of the tunnel, the results shown in Figure 12 rather well meet the requirements for the flow tunnel design.

Figure 12. Simulations of the flow tunnel.

After all these improvements the requirements were finally met and the maximal speed of 115 cm/s was reached. The Neptun 23 outboard motor however turned out to be unreliable and had to be repaired several times.

28

3.2.

The DPIV system development

In this chapter a description of the imaging and data-processing system is given. It consists of the camera, the laser and software. 3.2.1. Camera The camera used for DPIV experiments is Dragonfly Express an IEEE1394b camera from Point Grey Research Inc. The maximum resolution of the camera is 648484 pixels and the maximum supported frame rate is 450 frames per second (fps). However, for the maximum frame-rate, the resolution has to be reduced significantly. For DPIV experiments a 6 mm F1,6 CCTV CS-mount attached to the camera. The camera is shown in Figure 13. During the experiments, different computers were used and therefore an extra IEEE1394b (Firewire) adapter was bought. The original drivers shipped with the camera could not be used with LabVIEW 8.6 environment and only the drivers from Microsoft Windows XP SP1 were functional in Windows XP operating system. It was time-consuming and complicated to roll back the SP1 drivers without computer automatically changing them again with new ones. With the PC running Microsoft Windows 7 it was relatively easy to get the camera to work since Windows 7 supports 800 Mbit/s Firewire standard by default.

Figure 13. Dragonfly Express camera on a tripod.

In order to save videos, software was developed in LabVIEW 8.6 (user interface can be seen in Figure 14.). Software modules made earlier by a student Andres Hunt in the lab were used 29

for adjusting the camera and manipulating the main parameters of the video stream [21]. It is possible to choose different video modes, change resolution, frame rate, brightness, contrast and many other parameters. For experiments, 320240 pixels resolution and the frame rate of 50 fps were used as it gave the best image quality and a relatively low noise level compared to higher resolution and higher frame rate.

Figure 14. The camera control and video recording software user interface.

However, low resolution and noisy images set remarkable constraints to the PIV measurements, the need to view a relatively large area introduce the need for tracking particles with large diameter. For PIV analysis, the optimal particle image diameter is between 1-4 pixels [2]. To analyze a 2025 cm area with the resolution of 320240 pixels, particles with diameter of more than 1,2 mm are needed. This big particles however might not follow the flow accurately.

3.2.2. Laser Initially, different white light sources (a regular light bulb, halogen light bulb, a car front light, commercially available laser diodes, etc) were tested. The laser diodes were not powerful enough and with the white light sources it was not possible to create a thin, light sheet with a constant width. The easiest and cheapest way to get a suitable powerful laser appeared to be from a DVD-writer, in particular, and LG with 22x write speed. The light sources wavelength is thus accurately not known, but is about 650 nm (red). It was a continuous class 3B laser with the maximum power of less than 250 mW.

30

The laser diode together with mounting constructions, lenses and mirrors for focusing the laser beam, was removed from the DVD-writer and placed into a new metal housing (see Figure 15). To ensure sufficient cooling, a 55 cm copper plate was soldered on the diode case and on its rear side and a cooling fan was installed. To limit the supply current and to protect the diode, a 3,3 resistor was added in series to the laser diode. A cylindrical lens was used to create a light sheet from the beam. The laser was placed 0,75 m away from the studied object in the flow tunnel. The length of the FILOSE robotic fish is going to be approximately 0,5 m. Calculating the inverse tangent, the suitable lens angle is 40. As the flow pattern in front of and behind the fish are also important to be observed, a 60 lens was chosen allowing at most 86 cm measurement length along the objects longitudinal axis. To meet class 3B laser safety requirements, the housing of the laser is fully closed (shown in Figure 16), the lens is mounted directly to the case. Also, when working with the laser, the whole tank and laser itself are covered with blackout curtains. Cooling Laser diode Resistor

Optics

Figure 15. The internal view of the laser housing. Cooling fan

Cylindrical lens

Power cables Figure 16. The laser. 31

3.2.3. Software As mentioned in Chapter 2.2, MatPIV 1.6.1 software was chosen to process the experimental results. MatPIV has many options and parameters. An overview of most important of them is given in Table 5. Table 5. Parameters of MatPIV 1.6.1 adjustable by users. Parameter name Example value T met 0,02 multin Frame period Method for PIV calculations: single PIV calculations with single iteration; multi PIV with three iterations through the images; multin PIV with n iterations through the images, uses manually given window sizes. wins woco [64 64; 32 32] worldco.mat Window sizes to use in calculations. Name of a file containing the mapping between pixels to centimeters. maskfile sntrld mask.mat 1,2 Name of a file containing predefined mask. Threshold for use with SNR-filtering to sort out false data. globtrld 1,3 Threshold for use with global filtering to sort out false data. loctrld 1,7 Threshold for using with local filtering to sort out false data. Description of a parameter

This software is originally designed for analyzing two images, nevertheless it was possible to modify it to work also with avi video files. During the work, a script (refer to Appendix A) was written with some improvements. The improvements included working with video files; saving data not only in Matlab specific format (mat) but also in comma separated value (.csv) spreader sheet files; saving vector images in jpeg format and also in avi video format; calculating the mean velocity. Software was adopted for LabVIEW 8.6 programming environment, adding a graphical user interface (Figure 17) that simplified using it.

32

The following functions were added: choosing the video file; exploring it frame by frame; choosing a frame from where to start DPIV analysis; specifying a number of frames to analyze; displaying the resultant vector field.

Figure 17. Graphical user interface for MatPIV.

33

4. RESULTS
4.1. Measuring the flow speed

The measurement results in this section are presented for the open tunnel configuration described in Chapter 3.1.1 (see Figure 10. B) with some improvements from the third version of the tunnel. These include the flow laminator that was placed in front of the flow tunnel and a waterproof box for optics on top of the tunnel. As characterizing the tunnel configuration is rather time-consuming and initially no faster flow speed than 50 cm/s was required, this configuration has been the most used one and its characterization is described here. The Dragonfly Express camera was mounted above the waterproof box on top of the tunnel at the height of 0,7 m from the ceiling of the tunnel. The distance to the measurement plane, located at the center of the tunnel, was 0,95 m as shown in Figure 18.

Figure 18. The measurement setup.

Before starting the measurements, EPS-5 raw material granules were inserted to the tank as tracer particles. Since the tank is watertight, the particles can be inserted once and used for all experiments. In order to find the relation between the frequency set on the Mitsubishi F700 frequency converter and the corresponding flow speed inside the tunnel measurement area, calibration measurements were carried out. The frequency range applied to the AC motor was (535) Hz with 2,5 Hz step. Although the maximal frequency supported by the converter is 50 Hz, the calibration was possible to conduct just up to 35 Hz due to the extreme turbulences in the pool that started to misplace the equipment. At every frequency, 100 frames of video were 34

captured and analyzed using MatPIV software. As an example of two subsequent frames captured at the frequency of 15 Hz applied to the motor is shown in Figure 19 (A and B). A . B .

Figure 19. Two sequential frames recorded at the motor frequency of 15 Hz.

In order to get the velocities from PIV calculations, a calibration image had to be created to convert the image pixels to centimeters. For this, a 55 cm grid as shown in Figure 20 was placed into the flow tunnel on the measurement plane. MatPIV function definewoco was used to carry out this grid calibration. This allowed to calculate the actual velocity vector fields afterwards as shown in Figure 21.

Figure 20. A calibration image for MatPIV. Also some air bubbles that interfere with the view to the tunnel are seen in the tunnel ceiling.

35

Figure 21. Vector field corresponding to the flow at the applied frequency of 15 Hz.

The water flow speeds in the tunnel at different AC motor frequencies were calculated in a steady flow using the DPIV system. For this, water speed over all the (calculated) vectors in 100 frames was averaged, resulting in total of 100frames 29lines 39rows = 113 100 samples for each frequency (for the Matlab script refer to Appendix B). At the frame rate used, this corresponds to 2 seconds of measurement time (50 fps). The noisy vectors with unrealistic values were not accounted, making the overall number slightly smaller. To validate the repeatability of experiments, all the measurements were carried out three times (three cycles from 5 Hz up to 35 Hz with the step of 2,5 Hz). Also, from the standard deviations the measurement uncertainties at 95% confidence level were calculated. The calibration results for flow speed measurements using DPIV system are shown in Figure 22 and in Table 6. In Figure 22, both the longitudinal and perpendicular flow speeds at different frequencies are presented along with the first calibrations measurement uncertainties (95% confidence level). Table 6 gives the measured longitudinal flow speed values in detail. To validate and compare the results, another series of measurements at the same frequencies was conducted using a 3-axes Doppler velocimeter ADVOcean-Hydra from 36

SonTek Ltd (provided by Institute of Marine Systems, Tallinn University of Technology). Its readings however, were very noisy and could not be used. Communication with SonTek Ltd support engineers to troubleshoot this problem is still ongoing.

Figure 22. Flow tunnel calibration charts. A flow tunnels longitudinal flow speed with respect to the applied AC motor frequency. B flow tunnels perpendicular flow speed with respect to the applied AC motor frequency.

As it can be seen in Figure 22 A, the flow speeds measured with DPIV in all three calibration runs stay well within the first measurements uncertainty. From Figure 22 B, it can be seen that the average values of the perpendicular speed tend to be slightly positive, but remain close to zero (i.e. are zero within the uncertainty range limits). The slight offset can be caused also by the misalignment of the tunnels flow symmetry axes with respect to the cameras axes.

37

Table 6. Applied AC motor frequencies and corresponding flow speeds in the open flow tunnel. Frequency (Hz) DPIV Velocity (cm/s) 5 7,5 10 12,5 15 17,5 20 22,5 25 27,5 30 32,5 35 7,2 9,6 13,8 17,4 21,1 26,8 28,8 32,9 38,8 40,0 44,8 50,7 51,5 Standard deviation (cm/s) 1,3 2,8 3,3 2,5 5,7 2,4 8,2 4,4 6,7 9,1 9,0 5,0 10,4

4.2.

Flow profile

To evaluate the laminarity of the resultant flow, also the flow profile was extracted from the measurement results, and analyzed. The velocities in the test area in velocity field matrices were plotted row by row (for Matlab script, refer to Appendix C). These rows should rather precisely correspond to actual streamlines on the researched plane of the tunnel. As in the previous section, the DPIV velocity values in every row are averaged over 100 frames. Video resolution of 320240 pixels was used. DPIV interrogation window size was 1616 pixels, and the window was shifted for each new measurement so that it overlapped 50% with the previous frame. Thus, 29 lines in the vector field grid were gained. PIV results are shown in Figure 23 and they clearly indicate different flow patterns in the tunnel at different average flow speeds. It is clearly seen that at higher speeds (above 25 cm/s), the flow is slower on one side of the measurement zone. At very low speeds, the flow variations in the measurement plane are much smaller. However, these flow conditions were still good enough to satisfy the

38

requirements set for the FILOSE experiments as they were carried out in a narrow region in the center of the flow tunnels working zone.

Figure 23. Flow profile. Velocities were calculated along the streamlines at different AC motor frequencies.

This unequal speed distribution over the streamlines in the flow tunnel is most probably caused by the flow tunnel design. Also, the flow speed is smaller at the boundaries due to viscous friction and the camera capture area might not be aligned exactly along the symmetry axis of the flow tunnel. Further improvements in the tunnel design are a topic of the future work.

39

4.3.

Studying fish swimming patterns

The flow tunnel described in this work has been used in the FILOSE project to investigate fish swimming patterns in a controlled hydrodynamic environment. The aim was to study the relevance of factors such as shape and elasticity distribution in the ability of a compliant device to imitate the kinematic behavior of a fish. For comparative experiments six prototypes (shown in Figure 25) with three different geometries cylinder, truncated elliptical cone and elliptical cone with tail (fish-shaped prototypes) and one freshly killed rainbow trout were tested. The geometry and material properties were estimated on the basis of the elasticity distribution of the real fish, derived from the myometry analysis of a real fish.

Figure 24. Compliant prototypes used in the experiments.

Test objects were oriented with their main axis along the longitudinal direction of the flow tunnel. Objects were supported using a vertical rod and a rotational joint. A waterproof DC motor was used to actuate the body in the transverse direction of the tunnel with fixed amplitude sinusoidal signal. Each object was tested in three flow conditions: static water and laminar flows with speeds 25 cm/s and 50 cm/s with five different frequencies (15 Hz). This corresponds to 105 experiments in total. Movement was captured by an overhead camera and images were postprocessed to obtain point-wise lateral displacement, bending curvature, phase velocity and bending moment distribution for each object. In Figure 26, two sets of captured images are shown.

40

Figure 25. Two sets of images captured during experiments in static water at 1 Hz. First row rainbow trout. Second row silicone tail prototype (P6).

The test results showed that the principle that as geometry gets closer to the geometry of fish, performance does too, does not necessarily hold - in all test scenarios, prototypes with fishlike geometry (P4, P5 and P6) perform significantly better than the rest. However, contrary to our expectation, cone-shaped prototypes did not perform better than the cylinder. For further research details refer to Appendix D. The results were published in O. Akanyeti, A. Ernits, C. Fiazza, G. Toming, G. Kulikovskis, M. Listak, R. Raag, T. Salume, P. Fiorini, M. Kruusmaa, Myometry-Driven Compliant-Body Design for Underwater Propulsion, In Proc. of the the IEEE International Conference on Robotics and Automation (IEEE ICRA 2010), Anchorage, Alaska, May 3-8, 2010.

41

4.4.

Krmn vortex street

The next series of experiments planned in FILOSE require creating an organized turbulence, a Krmn vortex street, in the flow tunnel. This work is already out of the initially required scope of the thesis and is explained to demonstrate the evolving requirements of FILOSE project. The Krmn vortex street is a pattern of regularly repeating swirling vortices caused by the unsteady separation of flow over bluff bodies [22]. The phenomenon is named after a Hungarian-American physicist Theodore von Krmn (1881 1963). For this experiment, a half-cylinder with diameter of 11 cm was placed in the middle of the flow tunnel as shown in Figure 26. The flow speed was varied from 7 to 42 cm/s and the flow was captured. The best results were achieved at the speed of 30 cm/s. Two captured image pairs are shown in Figure 27. Time separation between frames A and B is 0,02 s and time interval between capturing the two image pairs is 0,44 s. The flow direction is from right to left, and on the right side of the picture the edge of the semi cylinder is visible.

Figure 26. Semi-cylinder in the flow tunnel.

The results of the PIV analysis with the interrogation window size of 1616 pixels are shown in Figure 27 (1C and 2C). The colors are indicating the vorticity: the red is indicating the counter-clockwise vorticity while the blue is indicating the clockwise vorticity. Green marks the area with no vorticity detected. In Figure 27 (1C and 2C), the movement of three clockwise and three counter-clockwise vortices is visible. However, the phenomenon was unstable in the experiments and occurred only for a few seconds at a time, only at the speed of 30 cm/s. The work to achieve the stable Krmn vortex street is still in progress as it is necessary for the future experiments. 42

1A

2A

1B

2B

1C

2C

1 3

2 3

3 1

Figure 27. DPIV raw data and result. 1A and 1B are the first captured image pair, 2A and 2B are image files captured 0,44 s later. 1C and 2C are the corresponding DPIV vorticity analysis results.

43

5. DISCUSSION
As a result, a working hydrodynamic test bed was developed in two redesign steps, and the final stage was characterized. The measurement results confirm that the system meets the requirements set by the FILOSE project. Although the maximal flow speed requirement was reduced during the work down to 50 cm/s, the speed of 1m/s and higher can be achieved. It is possible to control the flow speed in the working area of the flow tunnel and the flow is laminar enough to conduct the experiments. With the constructed DPIV system it is possible to measure flow speed and investigate the flow structure as shown in Figures 21 and 25. Average flow speeds were numerically determined at different AC motor applied frequencies and the speed distribution over the tunnel cross section on the measurement plane was characterized for each of these average speeds. The flow laminarity appeared sufficient for current FILOSE project requirements. Also, it is possible to create high speed laminar flow or organized turbulence the von Krmn vortex street, by inserting an object (for example a cylinder or a semi cylinder) into the flow. However, this phenomenon was unstable, being visible only for a few seconds. As a result of building the equipment described in the work, it was possible to conduct experiments for FILOSE project. The system design appeared to be a success, however for future experiments, improvements are still needed. In particular it is necessary to further improve the laminarity of the flow at higher speeds as it is most probably the step leading to successful creation of stable a Krmn vortex street.

44

6. FUTURE WORK
To meet the emerging experimental requirements, it is still necessary to improve the flow tunnel and DPIV system as it is planned to conduct experiments similar to those described in [23-25]. In the following, the planned improvements are described in more detail. Modifications that have to be made in the design of the flow tunnel: It has to be avoided that the air bubbles can gather under the upper glass of the tunnel (as in Figure 20). This can be achieved by lowering the glass plate about 5 mm so that the glass is in the same level with the top of the laminator box and water can flow under the glass undisturbed. More laminar flow has to be achieved since its inhomogeneity may be the main reason preventing the steady Krmn street to appear. Due to frequent failures, the Neptun 23 outboard motor needs to be replaced.

To achieve better signal-to-noise ratio in PIV, hardware improvements are needed: The current camera must be replaced with a new one capable of at least the resolution of 12801024 pixels at 200 frames per second. It must be equipped with a wide angle high aperture lens. Smaller and lighter scattering particles that would follow the flow more accurately need to be found. In case of using considerably smaller particles, more powerful light source needs to be constructed. After these improvements, more accurate experiments can be made.

45

SUMMARY
The aim of this work was to create a controlled hydrodynamic test environment and a DPIV (Digital Particle Image Velocimetry) system to conduct experiments for the EU Seventh Framework programme project FILOSE (FIsh LOcomotion and SEnsing) in Centre for Biorobotics, Tallinn University of Technology. An overview of technical requirements of a DPIV system is given along with an example of a commercial DPIV system. As the commercial DPIV imaging system costs are high and commercial flow tunnels are not readily available, the customized system had to be constructed on site. The developed DPIV system consisted of a high-speed camera and a laser. The laser consisted of a class 3B laser diode from a DVD-writer. The thin light sheet was created using a cylindrical lens. As seeding particles, the EPS-5 raw material granules were used. For data processing, Matlab toolkit MatPIV 1.6.1 was chosen from available open source software, and it improved. This system satisfied the flow visualization needs sufficiently. In the existing flow tunnel the flow was too turbulent and slow. The first system redesign phase resulted in improvement of both parameters. At the second redesign phase, laminar flow was achieved and met the initially set requirements for the maximum flow speed of 1m/s. A reduced version of the final setup was characterized in detail and used in the experiments to investigate fish swimming patterns, and the results were published. The flow tunnel configuration was calibrated to correlate the applied AC motor frequency and the corresponding flow speed in the tunnel. To evaluate the flow laminarity, the flow profile over the tunnel cross section on the measurement plane was extracted. At higher speeds (above 25 cm/s), the flow appeared slower on one side of the measurement zone resulting in the decrease in laminarity. Still, the setup satisfied the initial task requirements. The work was further extended to create organized turbulence the von Krmn vortex street. However, this phenomenon was unstable, being visible only for a few seconds. To solve this problem, it is necessary to improve the flow laminarity in future. The camera with higher resolution is also needed to improve the quality of flow characterization. All in all, the constructed system meets the task requirements and has been successfully used in scientific experiments.

46

REFERENCES
1) M. Raffael, C. Willert, S. Wereley, J. Kompenhans, Particle Image Velocimetry: A Practical Guide, Second Edition, (Springer-Verlag Berlin Heidelberg 2007). 2) C. Tropea, A. L. Yarin, J. F. Foss, Handbook of Experimental Fluid Dynamics (Springer-Verlag, Berlin Heidelberg 2007), pp. 287 296, 309 - 327 3) J. Westerweel, Digital Particle Image Velocimetry Theory and Application, (Delft University Press 1993). 4) A. Melling, Tracer particles and seeding for particle image velocimetry, Meas. Sci. Technol. 8 1406, doi: 10.1088/0957-0233/8/12/005 (1997). 5) J. Westerweel, Fundamentals of digital particle image velocimetry, Meas. Sci. Technol. 8 1379 doi: 10.1088/0957-0233/8/12/002 (1997). 6) R. Hain, C. J. Khler, C. Tropea, Comparison of CCD, CMOS and intensi fied cameras, Exp Fluids 42:403411(2007). 7) R. D. Keane, R. J. Adrian, Theory of cross-correlation analysis of PIV images, Applied Scientific Research 49: 191-215 (1992). 8) TSI Incorporated, http://www.tsi.com, 15th of April 2010. 9) LaVision, http://www.lavision.de/, 15th of April 2010. 10) Dantec Dynamics http://www.dantecdynamics.com/, 15th of April 2010. 11) Excel Technical Data Sheet, (Laser Quantum, Stockport, 2007). 12) smd6000 Technical Data Sheet, (Laser Quantum, Stockport, 2007). 13) DPIV Laser Safe Operating Procedures, (University of Bath, Bath, 2010). 14) pco. 1200 hs / pco 1200 s Technical Data Sheet, (PCO AG, Kelheim, 2007). 15) AF-S Micro NIKKOR 60mm f/2,8 G ED Product Manual, (Nikon Corporation, Tokyo). 16) INSIGHT 3GTM, Data Acquisition, Analysis, and Display Software Platform, Users Guide, (TSI Incoprorated, Shoreview, 2009). 17) URAPIV an open source Matlab and Python Software for PIV analysis , http://urapiv.wordpress.com/, 04th of April 2010. 18) JPIV, http://www.jpiv.vennemann-online.de/, 05th of April 2010.

47

19) http://www.oceanwave.jp/softwares/mpiv/ 05th of April 2010. 20) mpiv - MATLAB PIV Toolbox , http://folk.uio.no/jks/matpiv/, 06th of April 2010. 21) A. Hunt, Feedback control of a coupled IPMC (Ionic Polymer Metal Composite) sensor-actiator, (University of Tartu, Tartu, 2009). 22) Wikipedia, Krmn vortex street, http://en.wikipedia.org/wiki/Krmn_vortex_street, 22nd of May 2010. 23) J. L. Tangorra, G. V. Lauder, P. G. Madden, R. Mittal, M. Bozkurttas, I. W. Hunter, A Biorobotic Flapping Fin for Propulsion and Maneuvering, presented at the IEEE International Conference on Robotics and Automation, Pasadena, USA, May 19-23, 2008. 24) B. P. Epps, P. V. Alvarado, K. Youcef-Toumi, A. H. Techet, Swimming performance of a biomimetic compliant fish-like robot, Exp Fluids, DOI 10.1007/s00348-0090684-8, (2009). 25) B. Ahlborn, S. Chapman, R. Stattford, R.W. Blake, D.G. Harper, Experimental simulation of the thrust phases of fast-start swimming of fish, The Journal of Experimental Biology 200, 23012312, (1997).

48

VOOLU VISUALISEERIMISE JA HDRODNAAMILISE KATSESEADMESTIKU EHITAMINE JA KARAKTERISEERIMINE

KOKKUVTE Gert Toming Selle t eesmrk oli eksperimentaalse vedelikednaamika katseseadme (voolu tunneli) ning heljumi kiirusvlja digifotograafia (edaspidi DPIV) seadmestiku konstrueerimine. Seadmestik on vajalik eksperimentide lbi viimiseks Euroopa Komisjoni seitsmenda raamprogrammi projekti FILOSE (inglise k. FIsh LOcomotion and SEnsing) raames TT Biorobootika Keskuses. Kesolevas ts antakse levaade DPIV ssteemi phitdedest ja nudmistest ning nitena tutvustatakse terviklikku kommertsssteemi. Kuna kommertsiaalsete DPIV ssteemide hinnad on krged ja voolutunnelit ei ole vimalik valmislahendusena osta, tuli nimetatud ssteemid ise luua. Konstrueeritud DPIV ssteem koosneb kiirest digitaalkaamerast ja laserist, mis valmistati DVD kirjutaja 3B klassi laserdioodi baasil. Silindrilise ltse abil moodustati huke valgusleht ja valgustati vette lisatud EPS-5 tooraine graanuleid. Nhtavate osakeste liikumine salvestati kaameraga edasiseks analsiks Matlab keskkonnas loodud vabavaralise programmiga MatPIV 1.6.1. Programmi tiendati vastavalt vajadustele. Selline ssteem rahuldas DPIV mtmiste nuded. Algses voolutunnelis oli vool liiga aeglane ja turbulentne ning tunnel vajas modifitseerimist. Esimese mberehitamise tulemusena paranesid mlemad nitajad oluliselt ning peale teist mberehitust saavutati piisavalt laminaarne vool ja juti voolukiiruseni 1 m/s. Viimase voolutunneli krbitud versioon karakteriseeriti ja kalibreeriti. Leiti kasutatavale mootorile rakendatud sagedustele vastavad tunnelis saavutatud voolukiirused ning valideerimaks voolu laminaarsust uuriti voolu profiili eksperimendiala likes. Kuigi selgus, et suurtel kiirustel (le 25 cm/s) on vool eksperimendiala hes servas aeglasem, rahuldasid saadud tulemused esialgsed nudeid. Voolutunnelit kasutati eksperimentides kalade ujumise uurimiseks.

49

Saadud tulemuste phjal valmis teadusartikkel O. Akanyeti, A. Ernits, C. Fiazza, G. Toming, G. Kulikovskis, M. Listak, R. Raag, T. Salume, P. Fiorini, M. Kruusmaa, Myometry-Driven Compliant-Body Design for Underwater Propulsion, mis publitseeriti ICRA 2010 konverentsil (3-8 Mail Anchorages, Alaskal). Lisaks seatud eesmrkidele ttati ka organiseeritud turbulentsi ehk Krmni tnava loomise kallal. Nhtus osutus ebastabiilseks ja oli ilmnemisel jlgitav vaid mne sekundi vltel . Probleemi lahendamiseks on vajalik edaspidises ts muuta vee voolu laminaarsemaks ning paremaks jlgitavuseks tuleks hankida krgema resolutsiooniga kaamera. Kokkuvtteks vib elda, et loodud ssteem osutus pstitatud nuetele vastavaks, on olnud kasutusel uurimusts ning on selle kasutamine ja tiustamine jtkub ka edaspidi.

50

APPENDIX
Appendix A
Improved Matlab script to use video files with MatPIV 1.6.1 software, as a result velocity data is saved in Matlab specific format .mat and also in comma separated value .csv files. Plotted velocity vector fields are saved as .jpeg image files.
clear all; %%%%%%%%%%%%%%%%%%%%% Settings % Time separation between base and cross images. T=0.02; %50FPS met='multin'; % Use interrogation window offset wins=[64 64;64 64; 32 32;32 32; 16 16;16 16]; %windowsizes to use in the iterations woco='worldco1.mat'; % World coordinate file fo calibration. globtrld=3; % threshold for use with globalfiltering loctrld=1.7; % threshold for use with local filtering med='median'; % Use median filtering in localfilt int='linear'; % interpolate outliers using linear interpolation maskfile='polymask.mat'; % name of file containing the pre-defined mask A=aviread('C:\Experiments\videod\14 Mai\Run1\15Hz.avi'); % load the movie %%%%%%%%%%%%%%%%%%%% vektorvideo = avifile('C:\Experiments\1_16_Mai\pics\Run1\vektor_video.avi'); h = []; % handle to the figure for i=1:length(A)-1 % PIV calculation [x,y,u,v,snr]=matpiv(A(i).cdata,A(i+1).cdata,wins,T,0.5,met,woco,maskfile); % SnR filter: [su,sv]=snrfilt(x,y,u,v,snr,snrtrld); % global filter: [gu,gv]=globfilt(x,y,su,sv,globtrld); % local median filter: %[lu,lv]=localfilt(gu,gv,loctrld,med,x,y); % interpolate outliers [fu,fv]=naninterp(gu,gv,int,maskfile,x,y); %Saving data files save(['C:\Experiments\1_16_Mai\data\Run1\vel_field',num2str(i),'.mat'],'u', 'v','snr','fu','fv'); csvwrite(['C:\Experiments\1_16_Mai\data\Run1\fu_',num2str(i),'.csv'],fu); csvwrite(['C:\Experiments\1_16_Mai\data\Run1\fv_',num2str(i),'.csv'],fv); if i==1

51

save coordinates.mat x y end %Plotting data Plot1 = quiver(x(1:4:end),y(1:4:end),fu(1:4:end),fv(1:4:end),2); axis tight hold on; %Add streamlines %h=mstreamline(x,y,fu,fv,2); %Calculate and add vorticity to the plot w=vorticity(x,y,fu,fv,'circulation'); pcolor(x(2:end-1,2:end-1),y(2:end-1,2:end-1),w), shading flat, colorbar %Calculate and add velocity magnitude to the plot %w=magnitude(x,y,fu,fv); %pcolor(x,y,w), shading flat, colorbar hold off; %Save calculated vectorfields as image files h = gcf; if i < 10 image_name = 'C:\Experiments\1_16_Mai\pics\Run1\im_000'; end if i < 100 && i >= 10 image_name = 'C:\Experiments\1_16_Mai\pics\Run1\im_00'; end if i < 1000 && i >= 100 image_name = 'C:\Experiments\1_16_Mai\pics\Run1\im_0'; end filename = sprintf('%s%d.jpeg',image_name , i ); saveas(h,filename,'jpg'); %Add add calculated vector field to the video file vektorvideo = addframe(vektorvideo,h);

end vektorvideo = close(vektorvideo); %close vide file

52

Appendix B
Averaging vectors in 100 frames.

clear;

%Defining a variable use as a counter count = 0; %Reading data from files for i = 1:100 asukoht = 'C:\Experiments\1_16_Mai\data\Run3\35Hz\vel_field'; number = num2str(i); laiend = '.mat'; failinimi = [asukoht, number, laiend]; fail = open(failinimi); failid(i) = fail; end %Averaging data [m,n] = size(fail(1).fu); fu_mean(1:m, 1:n) = 0; fv_mean(1:m, 1:n) = 0; counter = 0; for i = 1:m for j = 1:n for k = 1:100 if (isnan(failid(k).fu(i,j)) == 0) counter = counter + 1; fu_mean(i,j) = fu_mean(i,j) + failid(k).fu(i,j); fv_mean(i,j) = fv_mean(i,j) + failid(k).fv(i,j); end end end end fu_mean = fu_mean/100; fv_mean = fv_mean/100; %Save data save(['C:\Experiments\1_16_Mai\data\Run3_35Hz_vel_field_mean.mat'],'fu_mean ', 'fv_mean');

53

Appendix C
Matlab script used to study the velocity profile of the measurement area. For this the measurement area velocities in the calculated velocity field matrices were plotted row by row.
clear; %Opening data file data(1) = open('data/Run2_5Hz_vel_field_mean.mat'); data(2) = open('data/Run2_7,5Hz_vel_field_mean.mat'); data(3) = open('data/Run2_10Hz_vel_field_mean.mat'); data(4) = open('data/Run2_12,5Hz_vel_field_mean.mat'); data(5) = open('data/Run2_15Hz_vel_field_mean.mat'); data(6) = open('data/Run2_17,5Hz_vel_field_mean.mat'); data(7) = open('data/Run2_20Hz_vel_field_mean.mat'); data(8) = open('data/Run2_22,5Hz_vel_field_mean.mat'); data(9) = open('data/Run2_25Hz_vel_field_mean.mat'); data(10) = open('data/Run2_27,5Hz_vel_field_mean.mat'); data(11) = open('data/Run2_30Hz_vel_field_mean.mat'); data(12) = open('data/Run2_32,5Hz_vel_field_mean.mat'); data(13) = open('data/Run2_35Hz_vel_field_mean.mat');

%Finding the nuber of rows and columns [m,n] = size(data(1).fu_mean); for i = 1:13 %Calculating mean values and standard deviations count0 = 0; for j = 1:m for k = 1:n if (isnan(data(i).fu_mean(j,k)) == 0) count0 = count0 + 1; arvud_fu(count0) = data(i).fu_mean(j,k); arvud_fv(count0) = data(i).fv_mean(j,k); end end end u_std(i) = std(arvud_fu(1:count0)); v_std(i) = std(arvud_fv(1:count0)); u_mean(i) = mean(arvud_fu(1:count0)); v_mean(i) = mean(arvud_fv(1:count0)); end for i = 1:13 %Filtering the data row by row %used variables count1 = 0; count2 = 0; fu_data = 0; fv_data = 0; temp1 = 0; temp2 = 0;

54

for j = 1:m %Filtering count1 = 0; temp1 = 0; for k = 1:n %Flow tunnel's longitudinal speed if (isnan(data(i).fu_mean(j,k)) == 0) if (((data(i).fu_mean(j,k) < u_mean(i) + 2*u_std(i))) &&... ((data(i).fu_mean(j,k) > u_mean(i) - 2*u_std(i)))) count1 = count1 + 1; temp1(count1) = data(i).fu_mean(j,k); end end end elements_fu(j) = count1; fu_data(j,1:count1) = temp1(1:count1); end r_elements_fu(i) = count1; %r_elements_fv(i) = count2; ridu_fu(i,1:29,1:r_elements_fu(i)) = fu_data(1:29, 1:r_elements_fu(i)); end %Standardhlbe ja keskmise leidmine for i = 1:13 for j = 1:29 for k = 1:r_elements_fu(i); fu_rida_mean(i,j) = mean(ridu_fu(i,j)); end end end

%Ploting data x = -14:14 plot(x, fu_rida_mean(1,1:29), '-*r', x, fu_rida_mean(2,1:29), '-*g',... x, fu_rida_mean(3,1:29), '-*b', x, fu_rida_mean(4,1:29), '-*c',... x, fu_rida_mean(5,1:29), '-*m', x, fu_rida_mean(6,1:29), '-*y',... x, fu_rida_mean(7,1:29), '-*k', x, fu_rida_mean(8,1:29), '-*k',... x, fu_rida_mean(9,1:29), '-*r', x, fu_rida_mean(10,1:29), '-*g',... x, fu_rida_mean(11,1:29), '-*b', x, fu_rida_mean(12,1:29), '-*c',... x, fu_rida_mean(13,1:29), '-*m'), grid('on') legend('5Hz', '7,5Hz', '10Hz', '12,5Hz', '15Hz', '17,5Hz', '20Hz',... '22,5Hz', '25Hz', '27,5Hz', '30Hz', '32,5Hz', '35Hz')

55

Myometry-Driven Compliant-Body Design for Underwater Propulsion

O. Akanyeti1 , A. Ernits2 , C. Fiazza1 , G. Toming2 , G. Kulikovskis3 , M. Listak2 , R. Raag2 , T. Salum ae2 , P. Fiorini1 , M. Kruusmaa2

Abstract Within the broader scope of underwater biomimetics, in this paper we address the relevance of factors such as shape and elasticity distribution in the ability of a compliant device to imitate the kinematic behaviour of a sh. We assess the viability of myometry as a tool to determine candidate mechanical parameters without relying solely on analytical models; we show that we can obtain elasticity distributions that are both consistent with previous theoretical investigations and experimentally better adherent to the passive kinematics of a biological embodiment (rainbow trout).

I. INTRODUCTION The overall goal of our work is to design a biomimetic underwater propulsion system with higher power-thrust efciency and maneuverability than current underwater vehicles. The biological inspiration is steadily swimming sh. In order to reproduce the dynamics of a sh during its steady swimming, in particular sub-carangiform swimming in rainbow trouts, we rst study its structural morphology. Our starting point is assessing the properties of the devices structure, examining in particular the following three questions: 1) Which characteristics of sh morphology enable sh to achieve high swimming performances? 2) Which mechanical design approach is most suited to develop an underwater vehicle with similar characteristics? 3) In the chosen approach, which design parameters are most relevant? What methodology can we successfully employ to determine values for such parameters? The rst question is answered in literature. Previous research on sh swimming ([1], [2], [3], [4], [5]) stresses the importance of few key features in achieving high swimming efciency: i) undulating motion mechanics, in which a body wave travels downstream with phase speed greater than the shs swimming speed; ii) the ability to alter the body wave speed, by adjusting tail beat frequency and/or body wavelength; iii) the ability to tune the body wave in terms of amplitude and phase. In light of this set of competences, we address the following question What is the simplest mechanical design that can accomplish this overall functionality?. Traditionally, robots that mimic sh ([6], [7], [8]) are built with rigid components connected by joints. This design style leads to complex mechanisms with inevitable controllability difculties.
of Computer Science, University of Verona, Italy for Biorobotics, Tallinn University of Technology, Estonia 3 Dept. of Theoretical Mechanics, Riga Technical University, Latvia
2 Center 1 Dept.

A. Modelling sh as visco-elastic bodies We have decided to use the visco-elastic compliant-body approach recently proposed in [9]. This approach is inspired by [10], in which the body of a real sh is modeled as a visco-elastic beam. In this approach, devices are compliant continuous exible bodies, in which material distribution allows a minimal set of input functions to exploit resultant modes of vibration for locomotion. Locomotion-inducing waves travelling along the elastic body are achieved through phase differences between subsequent points in the body; phase depends on viscosity and timing is critical. Hence, two important design parameters emerge in affecting device performance: spatial distribution of elasticity and of viscosity. These design parameters encapsulate i) geometry of the body (shape and size), ii) material distribution along the body, and iii) visco-elastic characteristics of each constituent material. 1) Analytical Approach: A promising way of computing these parameters is through analytical models. The standard model for sh comes from slender-body theory ([1]), which has been developed for small displacements but can be extended further. In recent studies ([10], [11]), the sh body has been modeled as a visco-elastic beam and analyzed in the domain of Euler-Bernoulli/Timoshenko beam theory. Lateral motion is described by the solution of the partial differential equation (PDE) associated to the visco-elastic beam. Given a geometry, the desired kinematics and the set of available actuation signals, a PDE model can relate prototype behavior to the target parameters. For simple geometries and uniform material distributions these equations can be solved to obtain reliable values ([12]). Moreover, PDE models can be used to determine favorable operating conditions such as optimal operational range of frequencies and position of the excitation point. However, PDE models, in general, suffer from an enormous increase in computational complexity with increasing dimensionality. As the number of parameters increases for geometries with tail, ns and non-uniform material distributions, the models get more complex and it is extremely difcult to solve their characteristic equations. This leads to over-simplied models which perform poorly when compared to biological sh. 2) A myometry-based approach: In this paper, we propose an alternative method and test its viability. We wish to obtain design parameters from direct investigation of real sh bodies. For direct analysis, we perform myometry on sh. Myometry is a method for non-invasive measurement of biomechanical properties of the muscles; myometry data

can enable us to better understand how the visco-elastic properties of sh muscles change along the body. If successful, this method allows bypassing analytical difculties. In this paper, we focus in particular on the elastic properties of sh; we attempt to identify the elasticity distribution and employ it as values for soft-bodied robot design. The main benets of the proposed design method are: i) It is a simple, fast and efcient method for computing the desired modelling parameters. ii) It requires few theoretical assumptions. This enables us to design more realistic models with complex shapes and non-uniform material distributions. iii) Moreover, the method can be used to crossvalidate the analytical modelling approach and complement it; for instance, ndings from myometry can be used as boundary conditions for analytical models or to reduce the dimensionality of unknown model parameters. B. Paper methodology Our methodology can be decomposed into four phases: 1) We perform myometry on biological specimens, measuring muscle properties along the body. We derive estimates for the elasticity distribution for real sh. 2) On the basis of geometry and estimated elasticity, we estimate the target properties of our biomimetic device: geometry, material properties, material distribution. 3) We manufacture prototypes for comparative experiments; prototypes vary in elasticity, elasticity distribution and geometry. 4) We conduct experiments to verify the kinematics of the prototypes against a real trout. We assess the kinematic similarity between prototype performance and real sh. We describe similarity by correlating body bending; we also evaluate the differences in point-wise displacement. II. FISH MYOMETRY To perform myometry we employ a Myoton, a device developed by M uomeetria AS. ([13]). The device operates by locally stimulating the tissue with a small impact and recording the resulting dampening oscillation through an accelerometer. A microprocessor analyzes the signals to output tone (frequency of the oscillation in Hz), logarithmic decrement (logarithm of the ratio of consecutive peaks) and stiffness of the tissue (in N/m) ([14]). Tone characterizes muscle tension, whereas decrement quanties a muscles ability to restore its initial shape after contraction. Stiffness refers to the muscles resistance to changes in shape due to external forces. Myometry is generally used to measure human skeletal muscles ([15]-[16]). In measuring relatively smaller-sized muscles (such as sh muscles), we expect that the Myoton readings do not depend solely on local muscle properties, but up to a certain distance (the effective depth) also on the neighboring materials along the direction of the applied force (Figure 1). However, so far no data is available in the literature to establish with precision what the effective depth is. When tested on a homogeneous material with varying

thickness, the myometers stiffness readings are found to be affected by the width at the point of measurement. We can consider the myometer readings k in terms of an equivalent EA stiffness: k = w (x) , where E is Youngs modulus, A is the area associated to the applied force and w(x) is the width at the point of measurement. We have also tested the myometer on a homogeneous cone with known elasticity, and compared myometer readings with the theoretical values for elasticity. We found that the myometer follows k = (x) EA , w(x) (1)

where (x) is a scaling coefcient with value approximately 0.5, and that adherence to theoretical values is poor when material thickness is very small (sensibly less than 5 cm). In terms of sh measurements, this means that sh is in the range of reliable thickness in the central section (see Figure 2, regions R1 R4).

Fig. 1.

Cross sectional view of sh body during myometry.

A. Experiments with Trout For myometry experiments we used two fresh-water trouts. One sh measured approximately 40cm and the other approximately 50cm in length. Myometry was performed shortly after death, to minimize intervening changes in muscle properties. During measurements, the specimen were placed on a soft pillow-like surface. For each sh, we chose 25 sample points, to cover the surface of the body in the regions of interest. Sampling points are grouped in 5 regions (from midsection (R1) to tail (R5)), each containing 5 measurement points; each measurement was repeated 10 times. Figure 2 shows location of the sampling points for the second specimen. We take measurements only on 5 points per region to contain the duration of the myometry experiments, to avoid changes in stiffness due to rigor mortis. B. Results and Analysis Table I summarizes myometry data and elasticity estimates. Figure 3 shows that in sh E increases towards the tail. The measurements in Region 5 should be considered less reliable because the width is very small and interference from the underlying material is not negligible. We believe the increase in E is related to the decrease in muscle-bone ratio of the body as we approach the tail, the relative effect of bones on stiffness measurements increases.

TABLE II G EOMETRY, LENGTH (L) AND ELASTICITY (E) VALUES FOR MANUFACTURED PROTOTYPES . E LASTICITY VALUES : E1 = 454.4KPa
AND CONE

E1 = 722.0KPa. S HAPES : CYLINDER (A), ELLIPTICAL TRUNCATED (B) AND FISH - LIKE (C ). G EOMETRY: r1 AND r2 ARE MAJOR AND
EXTREMITY.

MINOR RADII FOR THE ELLIPTICAL CROSS SECTION AT THE LARGEST

Fig. 2.

Measurement points in myometry experiments (sh 2)

TABLE I M YOMETRY READINGS AND ELASTICITY ESTIMATES . log. dec. Fish 1 R1 R2 R3 R4 R5 Fish 2 R1 R2 R3 R4 R5 2.1 0.1 2.2 0.1 2.4 0.1 2.7 0.1 2.8 0.1 1.4 0.2 1.2 0.2 1.2 0.2 1.3 0.2 2.2 0.1 freq. (1/s) 15.4 0.9 16.9 0.5 19.0 0.5 21.4 1.0 13.4 0.6 22.0 2.6 23.7 2.4 25.7 2.3 31.3 1.9 35.8 2.4 stiff. (N /m) 347.8 18.4 359.4 24.8 414.5 26.5 422.0 19.7 340.0 17.7 466.9 57.1 519.7 53.3 569.9 54.1 629.1 22.1 505.0 24.6 elas. (KPa) 222.6 224.8 275.7 478.3 618.5 247.9 285.0 370.9 610.3 703.2

Prot. # P1 P2 P3 P4 P5 P6

E (Pa) E1 E2 E1 E2 E1 E1 , E2

Shape A B B C C C

L (cm) 25 25 25 30 30 30

Geometry (cm) r1 = 4.5, r2 = 2.25 r1 = 6, r2 = 3 as P2 as P2 as P2 as P2

As bone is stiffer than muscles, E locally increases. Moreover, near the tail muscles are more tightly packed and there is more cartilage and tendon at the tail. This is also coherent with the theoretical models described in [10].

Fig. 3. Elasticity distributions of two sh. Elasticity values for two silicone materials are also plotted. Both materials are used during manufacturing prototypes (see Section III).

of 8, elasticity 450KPa) and Elite Double 22 (E22, Shore A scale: hardness of 22, elasticity 720KPa). Figure 3 shows the elasticity values of the two silicone materials along with the biological distributions obtained via myometry. E8 lies in the middle of the biological range (200-700 KPa) and can be considered representative of the average elasticity of sh, whereas E22 approximates the highest value measured in sh. The goal of our experimental procedure is assessing the passive kinematic similarity of biomimetic devices with respect to a sh body. For comparative experiments, we employ three different geometries (cylinder, truncated elliptical cone and elliptical cone with tail). Prototype P1 is a cylinder made of E8. This prototype is used as analytical reference, because a solved partial differential equation describing its dynamics is available in [9]. Prototypes P2 and P3 (made of E22 and E8, respectively) are designed as elliptical truncated cones. This geometry can be seen as intermediate in complexity between a cylinder and a realistic sh-shape. Prototypes P4 to P6 are elliptical cones terminating with a tail (sh-shaped prototypes). Prototypes 4 and 5 are made from E22 and E8, respectively. To approximate an increasing elasticity distribution and mimic the trend found from myometry, prototype P6 is hybrid: the rst part of the body is made of E8 and the posterior section, tail included, is fabricated out of E22. See Table II for physical properties of prototypes displayed in Figure 4. In total, we have 6 prototypes available for comparative experiments with varying elasticity, elasticity distribution and geometry. A. Fabrication of Prototypes

III. COMPLIANT-BODY DESIGN Using myometry data, we have thus identied a candidate distribution of elasticity for our biomimetic device (gure 3). Direct investigation of sh has provided information on both a desirable trend increase E towards the tail and a desirable value for average elasticity. The next step is to identify materials whose Youngs modulus lies in the biological range. We employ siliconebased materials Elite Double 8 (E8, Shore A scale: hardness

The manufacturing process comprises two steps: mould making and casting. For truncated elliptical cone-shaped bodies, a positive mould was cut out of expanded polystyrene (EPS) using electrically heated wire. After smoothing the surface with polyvinyl acetate (PVA) glue, the positive mould was covered with glass ber cloth and painted with epoxy resin to obtain the negative mould. More complex moulds were milled on a computer numerical controlled (CNC) machine. As the silicone we used is virtually impossible to

TABLE III AVERAGE S PEARMAN RANK CORRELATION AND AVERAGE DIFFERENCE

IN LATERAL DISPLACEMENT. L EFT: ACROSS ALL TEST SCENARIOS ; R IGHT: AVERAGED FOR FREQUENCIES IN THE 1-4 H Z RANGE . 5 H Z IS NOT IN THE BIOLOGICAL RANGE OF ACTUATION .

Fig. 4. Compliant prototypes used in the experiments. The prototypes are manufactured using two different silicone materials: i) Elite Double 22 (green) and Elite Double 8 (pink). The elasticity values for E8 and E22 are, respectively, around 450KPa and 720KPa.

P1 P2 P3 P4 P5 P6

All tests Spr. Lmd. 0.29 0.06 9.11 0.35 0.36 0.05 13.76 0.60 0.22 0.07 13.41 0.74 0.63 0.05 5.97 0.46 0.65 0.06 3.99 0.12 0.71 0.06 3.94 0.18

Reduced range (1-4 Hz) Spr. Lmd 0.33 0.06 9.15 0.38 0.36 0.06 13.83 0.66 0.29 0.06 13.62 0.84 0.68 0.04 6.10 0.56 0.73 0.04 3.98 0.15 0.77 0.05 3.95 0.22

glue or x with screws after curing, actuator mounts were inserted into the moulds during casting. The mounts are rigid plates with threaded holes for mounting and holes or pores to form a strong interface with silicone. IV. COMPARATIVE EXPERIMENTS A. Experimental Setup The experiments were conducted in the ow tank of Tallinn University of Technology. The size of the tank is (4x1.5x1.5)m3 , with working section volume (0.5x0.5x1.5)m3 and cross sectional area (0.5x0.5)m2 . The tank is aerated and powered with an electric motor to generate a laminar water ow. The tank is equipped with a Doppler sonar velocity log system to measure the laminar ow speed. We tested all 6 prototypes (Figure 4) and one rainbow trout (Oncorhynchus mykiss), immediately after death. Test objects were oriented with their main axis along x and up along z with gravity lying in z . Objects were supported using a vertical rod along z and a rotational joint. A waterproof DCmotor was used to actuate the body in the transverse direction (xy-plane). The applied torque was a sinusoidal signal with xed amplitude (1Nm). Each object was tested in three ow conditions: static water and two laminar ows, with speeds 0.25ms1 and 0.5ms1 . The static water regimen serves as a reference condition, as the corresponding equations have already been analyzed ([11]), whereas the highest laminar ow speed is close to the cruising speed of rainbow trouts. For each ow condition, each prototype was tested under applied torques of different frequencies (1-5 Hz). This corresponds to 105 experiments in total. In all cases, we tracked the kinematics of the models for approximately 30 seconds. Movement was captured by an overhead camera recording the test object against a lighted background. The camera images were logged with frame rate 60 Hz; images were postprocessed to obtain point-wise lateral displacement, bending curvature, phase velocity and bending moment distribution for each test object. Figure 5 shows two sets of sample images (a trout and the hybrid prototype P6), recorded in static water with operating frequency 1 Hz. In order to generate a good design for our biomimetic device, we investigate if tuning elasticity distribution and geometry can bring the prototype closer in behaviour to

our biological reference. Better adherence to the passive kinematics of the trout is considered better performance. Correlation between the prototypes and the reference specimen with respect to kinematic characteristics can verify if our myometry-based design method is a useful tool. We use Spearman rank correlation coefcients to relate the bending performance of prototypes to sh (curvature distribution); we also measure the difference in lateral displacement between prototypes and sh and evaluate the mean absolute lateral motion difference.

Fig. 5. Two sets of images captured by overhead camera (xy-plane) during experiments in static water at 1Hz. First row: rainbow trout; second row: hybrid prototype (P6). The time interval between consecutive snapshots in the set is 100ms.

B. Results a) Complex interplay between geometry and elasticity: Figure 6 presents Spearman rank correlation coefcients and mean position difference. In Table III we summarize performance results across the 15 test conditions. In all test scenarios, prototypes with sh-like geometry (P4, P5 and P6) perform signicantly better than the rest. However, contrary to our expectation, cone-shaped prototypes did not perform better than the cylinder; the cylinder outscores both P2 and P3 consistently on error and P3 also on correlation. Thus, geometry is an important factor in achieving shlike swimming, although by itself it cannot predict relative performance. The principle that, as geometry gets closer to the geometry of sh, performance does too, does not necessarily hold even if elasticity is not allowed to vary. We also observe that the hybrid prototype (P6) performed slightly better than P4 and P5. P6 has the highest number of

Fig. 6. Experimental results. Row 1: Spearman rank correlation (similarity in bending behaviour), all correlation coefcients presented in graphs are statistically signicant ( p < 5%). Row 2: lateral motion difference (dissimilarity in amplitude behaviour).

best performances for both similarity criteria (8 w.r.t curvature and 10 w.r.t amplitude). Non-negligible improvement (6-8% across all tests and 4-9% in the reduced 1-4 Hz range) arises even from a crude 2-value approximation to the experimentally determined elasticity distribution. This conrms that designing compliant prototypes with varying elasticity is a key technology for biomimesis of sh. Turning attention to the role of average elasticity, we now consider prototypes with the same shape. The soft sh (P5) performs better than the hard sh (P4), but the hard cone (P2) performs better than the soft cone (P3). This contradicts the expectation that a change in average elasticity affects all shapes in the same manner. Also, the principle that, as average elasticity gets closer to the elasticity of sh, performance does too, does not necessarily hold even if shape is not allowed to vary. All these observations suggest a complex relationship between geometry and elasticity. b) Fish torso is overdamped beyond 3Hz in static water: In static water, the similarity between prototypes and real sh decreases with increasing input frequency. The phenomenon is more evident after 3Hz. The kinematic behaviour of sh changes distinctly after 3Hz; the anterior part of the body hardly bends and bending amplitude increases very slowly towards the tail. Figure 7 illustrates the bending motions of the sh body at two different frequencies (1Hz and 4Hz). The results are coherent with previous research reporting that the sh body is overdamped at high frequencies ([10]). Although the natural frequency of the prototypes is in the range of the shs natural frequency, no such overdamping is observed for prototypes. At higher frequency, the overall type of motion does not change for prototypes aside from the

obvious reduction in amplitude. This observation suggests that sh are mechanically engineered to operate at lower frequencies and this feature does not rest solely on geometry but requires tuning elasticity and viscosity distribution along the body. Overdamping of this sort can be observed also for experiments in laminar regimen, although the trend is far less clear than in still water.
h(x,t)
0.15

Lateral Motion
1Hz 4Hz

0.00

0.15

Fig. 7. Lateral motion h(x,t) of a trout body at 1Hz (blue) and at 4Hz (red). Note that at 4Hz the body torso hardly bends.

c) Observations on Passive Bending Moments: We now examine the distribution of the elastic bending moment ME in sh and sh shaped prototypes (P4, P5 and P6). ME is 2 h(x,t ) and is shown in Figure 8 for obtained as ME = EI x2 one cycle in static water at 1Hz. ME decreases gradually and approaches zero at two thirds of the body length. This trend is consistent with the reduction in passive visco-elastic bending moments predicted in [10], except for position of the peak amplitude for ME . In free swimming sh, the peak amplitude is just before the mid-part of the body, whereas

in our case ME is highest at the origin. This is due to the actuation style: in our experiment the sh body is subject to sinusoidal torque input at one extremity. In other words the boundary conditions are not the same: M (0) = 0. This shows that favorable effects arise in the hydrodynamic moment even when actuation is localized. Moreover, in prototypes there is a lateral asymmetry in the distribution of elastic bending moments, which is completely absent in the biological reference. Asymmetry seems to stem from the actuated extremity, so it might be possible that the phenomenon is at least partly originated by the fact that prototypes lack a head.
Me
2.0

Fish

Me 1.5
0.0

Prototype 6

The hybrid prototype performed better than other shshaped prototypes. We believe that it is the increasing elasticity that grants P6 an edge. Future Work: To approximate the elasticity distribution of sh we manufactured the body of the hybrid prototype (P6) in two sections using two different materials. This is a crude 2-value approximation and the performance is affected by the discontinuity. We are, therefore, investigating ways of designing prototypes with continuously-varying elasticity. We are also examining the following question: how can myometry readings be used to improve and/or simplify the analytical models?. We are looking for ways of employing myometry with live sh and working on developing a simple method to estimate the viscosity distribution from myometry experiments. VI. ACKNOWLEDGEMENTS This work is carried out under the FILOSE project, supported by the European Union, seventh framework programme (FP7-ICT-2007-3). R EFERENCES
[1] M. J. Lighthill, Note on the swimming of slender sh, Journal of FluidMech, 1960, vol. 9, 305-317. [2] M. J. Lighthill, Large Amplitude Elongated Body Theory of Fish Locomotion, Proceedings of the Royal Society, Biological Sciences, vol. 179, 1971, pp. 125-138. [3] F. Hess, and J. J. Videler, Fast Continuous Swimming of Saithe (Pollachius Virens): A Dynamical Analysis of Bending Moments and Muscle Power, Journal of Experimental Biology, vol. 109, 1984, pp. 229-251. [4] M. S. Triantafyllou, and G. S. Triantafyllou, and D. K. P. Yue, Hydrodynamics of Fish-like Swimming, Journal of Fluid Mechanics, 2000, vol. 32, pp. 33-55. [5] George V. Lauder and Peter G. A. Madden, Learning from Fish: Kinematics and Experimental Hydrodynamics for Roboticists, International Journal of Automation and Computing, vol. 4, 2006, pp. 325-335. [6] Michael S. Triantafyllou, and George S. Triantafyllou, An Efcient Swimming Machine, Scientic American, vol. 272:3, 1995, pp. 272-3. [7] D. S. Barrett, M. S. Triantafyllou, D. K. P. Yue, M. A. Grosenbaugh, and M, J. Wolfgang, Drag Reduction in Fish-like Locomotion, Journal of Fluid Mechanics, vol. 392, 1999, pp. 183-212. [8] J. Liu and H. Hu. Novel Mechatronics Design for a Robotic Fish. IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), Edmonton, Canada, 2005, pp. 2077-2082. [9] Pablo Valvidia y Alvarado, Design of Biomimetic Compliant Devices for Locomotion in Liquid Environments, Ph.D. Thesis, MIT, 2007. [10] J. Y. Cheng, T. J. Pedley and J. D. Altringham, A Continuous Dynamic Beam Model for Swimming Fish, Philosophical Transactions of the Royal Society B: Biological Sciences, vol. 353, 1998, pp. 981-997. [11] Pablo Valdivia y Alvarado, and Kamal Youcef-Toumi, Design of Machines With Compliant Bodies for Biomimetic Locomotion in Liquid Environments, Journal of Dynamic Systems, Measurement and Control, vol. 128, 2006, pp. 3-13. [12] Brenden P. Epps, Pablo Valdivia y Alvarado, Kamal Youcef-Toumi, and Alexandra H. Techet, Swimming Performance of a Biomimetic Compliant Fish-like Robot, Experimental Fluids, 2009. [13] M uomeetria AS., Myoton:Inventions for Performance, available at www.myoton.ee. [14] Zenija Roja, Valdis Kalkis, Arved Vain, Henrijs Kalkis, and Maija Eglite, Assessment of skeletal muscle fatigue of road maintenance workers based on heart rate monitoring and myotonometry, Journal of Occupational Medicine and Toxicology, vol: 1:20, 2006, pp. 1-9. [15] Georg Gavronski, Alar Veraksit, Eero Vasar, and Jaak Maaroos, Evaluation of viscoelastic parameters of the skeletal muscles in junior triathletes, Physiological Measurement, vol. 28, 2007, pp. 625-637. [16] Mario Bizzini, and Anne F. Mannion, Reliability of a new, hand-held device for assessing skeletal muscle stiffness, Clinical Biomechanics, vol. 18, 2003, 459-461.

0.0

2.0 0

3.5

x
Prototype 5

Me
1.5

Me
2 0

x Prototype 4

0.0

3.5

Fig. 8. The elasticity bending moment distribution (ME (x)) along the body (x) of sh and sh-shaped prototypes (P4, P5 and P6). The experiments were conducted in static water at 1Hz.

V. C ONCLUSIONS In this paper, we conduct comparative experiments to investigate the relevance of factors such as geometry and elasticity, in a visco-elastic bodys ability to imitate the kinematics of real sh. We also test the viability of myometry in orienting the design of biomimetic devices. The outcomes of our experiments are listed below: Myometry is an interesting tool and can be used to obtain more information about the biomimetic reference that we are trying to mimic. In particular, we demonstrate this fact by estimating the elasticity distribution of biological sh. There is a complex relationship between geometry and stiffness. Tuning one of them independently from the other does not always guarantee an improvement in performance. This limits the application of heuristic design methods based on trial and error processes and suggests that formal design methodologies, such as using myometry, are needed to optimize our biomimetic devices. The torso in sh is overdampened after 3 Hz. This indicates that the viscosity distribution plays an important role in regulating bending moments along the sh body. It also suggests that the structure itself generates stability for swimming. Therefore, if the task is to replicate efcient sh swimming, a noticeable component of stabilization derives from the embodiment itself. We expect that the task of control shall not be entirely in charge of guaranteeing stability.