DEVELOPMENT OF AN UNDERWATER FLAPPING FOIL TOW-TEST SYSTEM

Autonomous Underwater Vehicles (AUVs) using conventional methods of propulsion cannot operate in a dynamic environment such as the surf zone or around a reef. Bio-inspired propulsion mechanisms can provide the capabilities required to deal with the hazards of these locations. Animals such as birds, fish, and turtles use flapping foils as a means of propulsion and high maneuverability. Robotic flapping foils can be applied similarly to underwater vehicles like the AUV, Finnegan the RoboTurtle from MIT. Finnegan used four flapping foil "turtle fins" each with two degrees of freedom. One of those fins was outfitted to house force sensors which could provide instantaneous measurements of the forces acting on the foil. The purpose of this project, is to create a system capable of testing submerged underwater flapping foils in a tow-tank using force sensors to detect fluid flow phenomena. The force sensors had to be tested, installed, and calibrated so that they measured the forces about the desired axes. A method of attaching the fins to the towing carriage had to be devised and built out of aluminum. This attachment had to be sturdy due to dynamic loading and life expectancy for future use, streamlined where it would be submerged, and user friendly. The attachment also had to be able to adjust the depth of the fin. A data acquisition system was setup to record and transfer the force sensor measurements and foil positioning data from the towed flapping foil to the computer station. Tests were run to compare to past data in an effort to validate the testing system. The system can collect valid data currently, but improvements are necessary to achieve the desired functionality.


LIST OF TABLES
The purpose of this project is to provide a multi-functional testing platform to advance the capabilities of underwater vehicles using flapping foils.
Underwater flapping foil vehicles have potential to greatly expand the operating space of standard propeller-driven vehicle. This project is a component of the research done in the Robotics for Complex Underwater Environments Lab (R-CUE). Dynamic environments such as the surf zone, around reefs and wrecks, and also rivers are currently out of bounds to autonomous underwater vehicles (AUVs).
Operating near the seafloor is dangerous to vehicles unequipped to handle the different fluid flow effects that exist in that region. Force sensors installed in a flapping foil can give instantaneous estimates of the forces acting on the foil during motion. These forces can be used to quantify properties such as thrust, lift, and efficiency of the foil.
They can potentially be used to detect the presence of ground, walls, and the free surface.

Thesis Contents
Chapter 2 is a review of the literature pertaining to the flapping foils used in the project. It provides background information necessary to understand the foundations this project was built on. The Chapter 3 methodology explains the design, building, and testing of the major thesis components. These include the force sensors, carriage attachment, and data acquisition system. Chapter 4 includes the results of from force sensor calibration to the full system towed experiments. That section explores the trends and comparisons to past data. Chapter 5 provides details of modifications that will be made to the system, while also analyzing the existing sources of error. The future work is outlined in Chapter 6, which describes experiments that will be run with the current system, and testing of other projects currently in the research and design phase. Finally, the summary and conclusions brings all the basic findings together while providing a succinct statement of the project outcome.

REVIEW OF LITERATURE
The flapping foils used for this project come directly from Finnegan the RoboTurtle. Finnegan is an AUV developed at MIT which uses four flapping foils for all motion. The abilities of the flapping foils on this vehicle include propulsion, station keeping, high maneuverability, high turning speed, and low turning radii. Though the vehicle was large and weighed about 500lbs out of water, the flapping foils made it capable of advanced maneuvers compared to that of standard propeller driven vehicles.

Figure 1: Finnegan the RoboTurtle
The four flapping foils could achieve forward motion of 1.2 m/s, with each fin producing a maximum of 100N of lift with a mean thrust of 22N [Polidoro, 2003]. The solid model of one of the flapping foil systems is shown in Figure 4.
Design, construction, and testing of these self-contained flapping foil systems is detailed in [Polidoro, 2003]. In that thesis, the system was towed in a tank with similar dimensions to the URI tow-tank. The MIT tank is 30m long, 2.5m wide, and 1.1m deep, compared to the URI tank which is 30m long, 3.5m wide, and 2m deep.
The fin used was a standard NACA 0012 airfoil with a rectangular planform. There were four span lengths used of 0.3m, 0.4m, 0.5m, and 0.6m. Data provided includes thrust coefficient contours and time sequence lift and thrust data.
Data was collected in the MIT Tow-Tank using a six-axis strain gauge dynamometer which was mounted between the towing rig and the flapping foil. Force measurements were made externally to the system, at a sampling rate of 100 Hz. The experiments were towed at 0.5 m/s with just the fin submerged. The Reynolds number during testing was about 50,000. Sensor calibration, testing procedure, and post processing from that thesis created a basis for the methods used in this project. [Licht, 2008] described the creation of Finnegan the RoboTurtle from researching the swimming of an actual sea turtle to full-scale tests of the AUV in a pool. Finnegan used fins with a planform shape similar to a turtle and those same fins were used in this project. Licht gives insight on the wake formations from these flapping foils when they are fully submerged. Finnegan provides a direct link between testing the foils and their applied capabilities on a underwater vehicle. The maneuverability and other motion capabilities exhibited by Finnegan are highly dynamic compared to propeller driven vehicles. Finnegan was equipped with a sixaxis accelerometer, four narrow beam altimeters, a pressure sensor for depth measurement, and a downward facing doppler-velocity logger. These sensors provide adequate knowledge of the vehicle's location in the water, however there are no methods to instantaneously sense the fluid flow around the vehicle or the flapping foils. [Techet, 2008] investigated thrust coefficient contours over a range of Strouhal numbers and maximum angles of attack using a similar testing system. The contour plot from that paper with a heave to chord ratio ( ) of 1.5 provides the data that this project seeks to replicate to validate the new test system (Figure 2). Each red dot in that plot represents one test run with the corresponding Strouhal number and maximum angle of attack ( α max ). Thus, there were at least 28 tests performed. The shape of the fin used resembles a combination of a turtle fin like Finnegan used, and the rectangular planform from the Polidoro testing. A important difference is that the span and chord lengths were respectively 61.5% and 55% of the lengths used in this project.

Figure 2: Thrust Coefficient Contours (Techet)
Techet performed experiments in a flow tank where water was forced past the flapping foil instead of the foil being towed through stationary water. Forces were again measured with a six-axis strain gauge sensor, a torque sensor was used to measure roll shaft input power, and potentiometers mounted to each shaft gave additional position data. The author notes that the location of the center of pressure on the foil varies while flapping, but to keep consistent with propeller notations a position of 70% of the span length from foil root to tip is used for nondimensional calculations.
The Reynolds number was varied between 27,000 and 55,000. [Licht & Dahl, 2013] used a heaving and pitching NACA-0012 airfoil in a small tow-tank to examine ground-effect on flapping foils. The foil aspect ratio (AR) or the ratio of chord to span length was ~6.4 compared to the AR of 4 used in this project. The airfoil used in their study is an extruded aluminum foil with a rectangular planform. Force sensing was done with a six-axis dynamometer. The foil heaves and pitches to flap near a solid wall of the tank. The wall was adjustable such that they could test different distances and vary the ratio. Their findings show the mean generated lift while the fin moves towards the wall changes significantly with relation to distance from the wall. Also, they found an 18% difference in peak magnitude of instantaneous lift in downstroke vs. upstroke. That variation is instantaneously visible and therefore does not need to be compared to past data. This would allow a vehicle to potentially use ground-effect to benefit operation in real time.

1 Foil Kinematics
Flapping foils create propulsion by generating a wake pattern of vortices that act like a jet to provide thrust. This pattern is called a reverse Von-Kármán street ( Figure 3). The foils can also be used to create lift and drag and be used as control surfaces to maneuver the vehicle. Each of the four foil systems has two degrees of freedom referred to as pitch θ and roll ϕ. They each consist of the same two cylinder design with the roll motor, control card, amplifier, and power input in the large main cylinder, and the pitch motor components in the pitch cylinder. The roll motor rotates the entire pitch cylinder about the Y-axis, and the pitch motor rotates the turtle-fin foil about the X-axis as shown in Figure 4. The kinematics and equations for each parameter used are the same as those in the theses of Victor Polidoro and Stephen Licht [1,2]. The most important of these are explained below, paraphrased from those theses.
The equation for the roll position of the fin is given by: The equation for pitch position of the fin is: Where is the roll amplitude in radians, is the pitch amplitude in radians, is the frequency of foil motion in radians per second, and t is time.
This rolling and pitching motion causes a varying angle of attack with two components as opposed to a static foil like those on an airplane. The first component is the instantaneous pitch angle, and the second is the angle of incoming fluid based on the ratio of the pitch and roll velocities. The total angle of attack is not constant over the span of the foil, but a location was chosen which corresponds to an assumed effective center of hydrodynamic force. This location is 70% of the span length from the root of the foil . The root of the foil is the end of the foil that is closest to the roll-axis.
Where is the distance from the center of the roll axis to the root of the foil, and s is the span length.
Using that distance, the angle of attack is given by: [Techet 2008] showed the maximum angle of attack , affects lift, thrust, and wake structure. The data comparison done later maps mean thrust coefficients ( ) The last nondimensional ratio of significance relates amplitude of the heave motion to the chord of the foil: Where the heave motion is defined by:

Force Sensing
This project makes use of two piezoelectric force sensors mounted inside the pitch cylinder of a flapping foil mechanism. Installing the sensors in the device should produce direct measurements of the forces acting on the fin. With the sensors in the pitch cylinder, the whole system can be submerged which eliminates some inertial effects and body drag does not need to be subtracted. One of the four foil systems from Finnegan the RoboTurtle [Licht, 2008] was already outfitted with special mounting plates needed to mount the sensors. The plate design was modified as described in the Installation Design section.

Sensor Details
Two Type 9602 sensors from Kistler were installed between aluminum plates inside the pitch cylinder. These sensors have integrated amplifiers, which makes the body larger, but removes the necessity for expensive external amplifiers and also reduces the system's complexity. Each sensor measures forces in three directions, Fx, Fy, and Fz. As shown in Figure 5, they have a donut shape that houses the piezoelectric ceramics that detect the forces. For this to work, each sensor must be preloaded by a bolt through that donut hole which means a constant force must be applied as a reference. The bolt must also have an elastically deformable plastic sleeve to make sure the forces delivered are not distorted by space between the bolt and sensing surfaces. The full specifications can be found in Appendix 1, but the most important ones are noted here. Each sensor outputs analog voltages between +/-5V, nominally proportional to forces in the principle directions, which can be converted to forces after calibration. The measuring range for the axial direction is -5kN to 5kN, and for the other two directions is -2.5kN to 2.5kN. The forces expected from testing are a safe factor of 100 less than those limits. With the integrated amplifier, each sensor is still small enough to fit in the cramped space of the pitch cylinder. Communication comes by way of Cat 5 cable using 7 of the 8 conductors.

Installation Design
The pitch cylinder had a deeper cavity milled out of the Delrin to make space for the two aluminum plates that sandwich the force sensors. The plates provide the necessary preloading when bolted together. The bottom plate is bolted to the floor of the pitch cylinder cavity, while the motor components are mounted to the top plate.
This means that the bolts through the forces sensors that hold both plates together are the only way forces are translated through the system. The pitch shaft leads out of the cylinder to the fin. The locations of the two sensors (shown above) were determined based on the desired forces to measure, and available space around the other components. Two holes were drilled through both aluminum plates in-line with the pitch shaft. These holes were for the bolts going through the sensors. Two larger holes were drilled through the top plate for the communication cables to extend through.
The communication cables that came with the sensors had short Cat 5 tails which needed to be extended about 2m and terminated and the surface with RJ-45 connectors. Since the previous outputs of the pitch cylinder were wet mateable connectors with no unused pins, one of the connectors was removed and a new waterproof outlet was designed to take the two sensor Cat 5 cables along with the power and ground cables that go to the roll cylinder. The first choice was to just make a large enough hole and epoxy it closed, but this proved very difficult to keep waterproof since moving the cables deformed the epoxy. The hole was then threaded and a pipe nipple was screwed in with an o-ring to create a solid capsule to shield the epoxy from outside influences. This method proved effective, though it was still not waterproof as the epoxy used was very rigid. After a flexible epoxy replaced the rigid one, the connection held a vacuum, that is to say it was waterproof.

Calibration
The calibration of the force sensors involved designing and fabricating a method to hang test-weights applied to all three axes, collecting data with multiple weights at different distances from the sensors, and processing the data to output calibration matrices. A simple pulley system was created to hang weights on a line attached to different notches (indicated with red arrows) on one of the extra fins as shown in Figure 9. The pulley system proved unable to produce accurate results which can likely be credited to friction of the line and pulley. The next method chosen was to hang the weights directly from the notches with the fin angled perpendicular to the ground. The body of the flapping foil was rotated so that the forces were applied in the correct axes. This method allows independent variation of forces and moments. Therefore, each moment measured was a direct result of a force applied in the correct direction at the distance each notch was from the sensor in question.

Figure 9: Weights Hanging from Notches At Red Arrows
The weights used either had rope tied around them for hanging or were placed in small bags. The weights were 0.726kg, 1.725kg, and 2.269kg. Those values account for the weight of the bags and other hardware used in the system. The fin used has 6 notches spaced 0.04m apart, and 3 weights were hung from each notch. The data was collected using the full data acquisition system used for experiments. Each weight was given 10 seconds to hang and 10 seconds before and after hanging to let the sensors settle. The weights were labeled with letters and hung in the same order every time for testing consistency and repeatability. Once all data was collected, it was processed using Matlab to find the force sensor sensitivities (see 4.1 Calibration).

Carriage Attachment
The tank and mobile tow-carriage in the Sheets building on the Bay Campus of URI have been used for towing models and instruments in the past. However, no structure has previously existed that would provide a sturdy, and modular connection of a flapping foil to the carriage. Size and weight of the towing package contributed greatly to the design constraints. The importance of this structure is not solely for the flapping foil, but it is also expected to serve the needs of future towed mechanisms.

Design Constraints
The Carriage Attachment (CA) was designed originally for Ground-Effect experiments. Those experiments required the ability to tow a flapping foil and accessories at varied distances from the bottom. This implies that either the verticallyadjustable bottom sections of the tank (referred to as the beach) would have to be used, or the CA height must be modular. It was much more practical to make the CA modular than to go through the extensive beach adjustment process. The CA also had to be streamlined under the waterline so as not to add increased drag that might bend the attachment or hinder the movement of the carriage, or generate significant vortex induced vibrations.
The bulk of the towing package would itself require quite a rugged towing apparatus, but flapping also produces lift and thrust forces that translate through the structure. The three direction linear force constraint of the CA was to be strong enough to withstand up to 250N in each direction. This number comes from the conservative combination of the weight of what was towed (~170N) and a maximum of about 60N of fin-produced forces like lift and thrust [Polidoro, 2003]. Due to the static and dynamic loads, the strength of the structure was dependent not only on configuration, but also on the materials chosen. The materials had to be corrosion resistant over long periods of submersion, able to endure linear and torque forces, and still be light enough not to restrict the carriage movement or the ability to install and uninstall the structure.

Design Procedure
The first step taken was to create a solid model of the carriage and CA so it would fit and make use of the strength offered by the carriage beams. The use of 80/20 aluminum extrusions and connectors was an early choice for the CA. 80/20 is a highly adaptive building material coined "The Industrial Erector Set"(www.8020.net). The solid models of most 80/20 pieces are available to download for free and allowed easy construction of a system model with the carriage.

Build Parameters
Stock 80/20 extrusions were cut to the desired lengths and angles. Once all cuts had been made, the proper connecting plates and bolts ( Figure 15) were applied to put the upper truss structure together matching the model. All of the load-bearing 90 o connections use an "anchor" which was counter sunk into the 80/20 on two sides to create a high-strength locked connection (see Figure 15). All pieces are aligned within 1 o of their intended angle and the horizontal pieces were all leveled. The truss can either connect to the cube or the mast section.   Fastening each section together using bolts is not a very quick method which means modifying the CA configuration is time consuming. However, the structure is very sturdy compared to a quick-release design, and the time is on the order of minutes which is acceptable given the total time of average experimental use. There are no moving parts to fail, all components are inexpensive, and they come from the same manufacturer which makes them easily replaceable. The mast section was cut and welded at a local machine shop, Fabri-Tec Engineering (Wakefield, RI) based on the generated design drawings. To remedy the bolt issue, three of the bolts were removed and replaced with socket-head bolts that were screwed in from above the carriage. The one remaining original bolt was left as a guide to slide the CA down. This resulted in total bolting and unbolting times of less than 5 minutes with use of a hand drill and the proper bit.

Installation/Removal Method
Similarly, the time and method to get the CA to the bolting location was drastically altered. All intruding steel components were removed from the top of the carriage which allowed the CA to be raised up via winch, and lowered through the top of the carriage. This method not only reduced the total installation time from hours to around 20 minutes, but it also kept all participating parties dry the entire time. With some extra time and physical labor the installation process is doable by a single operator, though it is highly preferable to have two.

Performance Analysis
With the most updated installation technique, no problems have arisen in relation to the bolting throughout testing. There are also no visible vibrations or deformations under testing loads. All connections have held up and the structure has remained flush in the three axes which is indicative of positive integrity. Fluid flow around the mast section in both directions seems concurrent with the models implying similar forces applied due to the symmetrical and streamlined profile. Significant corrosion has not been detected. The CA has been stored hanging as well as braced against steel beams and concrete putting stress on many of the different members.
Even with the weight, and sometimes chaotic jostling that comes with a single user, the structure remains solid.  Figure 20 shows the current configuration.

Figure 20: Chassis Front Panel Connections
The chassis communicates to the land-based computer via the fiber-optic cable.
There is 100m of this cable available, though much of it is coiled up in its current installation. The cable sends the digital input signals to the computer where it is received and interpreted using the following programs. National Instruments provides proprietary software which allows the user to see all connections from multiple chasses, test the connections, and view the inputs being received. Galil Tools Lite is a free program used to control the flapping foil. A custom LabView2012 VI was used to monitor and record the force sensor outputs as well as poll the control card for instantaneous velocity, position, and torque estimates.

Configuration
The components housed on the carriage platform are mounted in ways to avoid outside objects during towing, and to resist movement caused by carriage direction changes. Both the chassis and power source sit between pieces of wood drilled into the platform that are large enough to stop any slipping movement, but small enough and out of the way such that all buttons and inputs are free.  The path of the fiber-optic cable is traced around the rim of the tank, and terminates at the computer. Figure 22 below shows a profile view of the carriage section of this model.

Calibration
The force sensors were calibrated in-situ by hanging 3 known weights from 6 notches in one of the fins. Each notch is 4cm apart starting from 0.175m from the roll axis. The weights were hung with the apparatus oriented to apply a force in the Zdirection, and then reoriented for the Y-direction. To find the resultant force acting on each sensor with each weight, the following equations were solved.
These equations correspond to the diagram below. , , and are the forces acting at the 2 force sensors and on the foil respectively. " " is the constant length between the force sensors, and " " is the varying length between the force on the foil and Sensor 2.
These calibration tests create plots of the voltage measurements from weights hung in one direction at different distances. The example below shows weights hung in the Z-direction.  The figures 25-28 show linear fit quality and coefficients between measured voltages at both sensors and the calculated Y and Z forces. Nominally pure Y and Z forces were applied, allowing a linear fit to be found directly between the voltages and forces of the Y and Z axis for each sensor. Each plot shows evidence that a separate slope exists for the data points from each of the three weights hung. These slopes may be a result of the testing apparatus or sensor installation. They could also likely be an effect of the moment applied by system components, but they will be further explored in future work. The possible presence of a small amount of cross-sensitivity between each axis on both sensors can be seen, but it is not possible to separate this effect from error caused by the angle of the fin while weights were hung. Therefore, the crosssensitivities were neglected. The slopes of the linear fit lines shown on each plot are the sensitivities (s) of the sensors in the indicated axes. Taking the inverse of these sensitivities gives the gain that is used with voltage measurements to find forces in the following equations.
From those equations, denotes force, denotes voltage measurement, the subscript letters correspond to sensor axes, and the superscript numbers state the sensor numbers. The sensitivities and gains are listed in the table below.

Calibration Error and Modifications
During calibration, human error could distort the readings based on hanging the weights, measuring the distances between them, and accurately aligning the fin angle. Furthermore, with more weights and more distances the total grid of measurements can be expanded which should increase the accuracy of calibration. The weights used could also be more precise too since the ones used are not from a standard set. Modifying the calibration method in these ways will increase the time it takes to calibrate, but if that becomes an issue the time could be reduced with an additional person helping during the procedure. There should not be any shortcuts taken during the calibration procedure as all data is essentially invalid without accurate calibration matrices.

Testing
The data from this project was collected to validate the test system by comparing to the results from past underwater flapping foil studies. The two sets of data for comparison come from the thesis discussing the design of the fin mechanics [Polidoro, 2003], and a paper looking at the performance of said fins [Techet, 2008].
Both ran experiments with a varying Strouhal number (St) and maximum angle of attack to find the mean thrust coefficient ( ) contours over the operational range of the fin at a specific heave to chord ratio ( ). values were compared to Techet, and instantaneous thrust and lift values were compared to Polidoro.
The tests run were almost exactly the same as those of Techet seen in Figure 2 with each red dot indicating a separate test run. To get enough data, the first four tests were run twice since their designated flapping frequency did not allow for ten or more flaps per test. There was a minimum set of ten full strokes required for each combination of St and . Therefore, there was a total of at least 32 tests run to create these results. The test matrix of those combinations is shown in Table 3. The full matrix with dimensional parameters is included in Appendix 4.

Analysis Process
There are four files created from each test run: (1)  The forces acting at each sensor are used to calculate the moment about the roll and pitch axis. After a 5 th order Butterworth filter at 10 Hz is applied, the time sequence moment data is phase averaged using the peaks of the pitch motor position data as zero-phase. Throughout this process, the range of data collected, the carriage velocity, and thrust coefficients are all saved in external files. All of the programs used for analysis can be found in Appendix 2.

Sensor Forces to Fin Forces
The current distance between the force sensors do not allow for direct force measurement due to high noise production. Therefore, the forces must be calculated.
The force values that come from each sensor are in opposite directions creating a moment about the perpendicular axis at the origin between them. Refer to the reference frame in Figure 4. Forces along the Z-axis produce a moment about the rollaxis while forces along the Y-axis produce a moment about the pitch-axis. The equations are as follows: Where is the total moment applied, and are the forces from sensors 1 and 2 respectively, and and are the distances from sensor 1 and sensor 2 to the axis of rotation which serve as the moment arms. The diagram in Figure 29 shows these forces and distances.

Figure 29: Forces Applied to Sensors
The forces applied to each sensor are typically a factor of ten larger than the force acting on the fin at the center of pressure location. The constant moment arms are always 1.5cm. The sum of these opposing applied moments largely cancels each other out. The resulting difference is the magnitude of the pure moment acting on that axis, which can then be converted to a force on the foil.
The calculated moments are divided by the distance to the center of pressure ( ) to produce the forces at that point. Lift force (F L ) is determined from the Zdirection moment (M Z ) and thrust (F T ) is determined from the Y-direction moment (M Y ).

Thrust Coefficient Contours and Assumption
For preliminary validation of the test system, the results were examined using an assumed in the above formulas. To prove this method of assuming a is credible, the following explains and shows how that assumption affects the results.
The is a linear factor which changes contour values when modified, but the contour shapes remain the same. The must be positioned between the root and the tip of the foil, which is a 0.4m range and extends from 0.155m to 0.555m from the roll axis. Therefore, there is a range of values that we can set as the . To visualize the way contours differ over this range, plots with the center of pressure located at the minimum, 70% of the span, and maximum distances are shown in figures 30-32.
was calculated via the formula above where is the mean thrust from phase-averaged thrust data. is the water density, U is the fluid velocity, c is the chord length, and S is the span length. The thrust coefficient contour plots all show similar patterns which helps to validate this test system when compared to the Techet data in Figure 2. It is evident from those results that thrust coefficients do not linearly increase with increasingly aggressive flapping. That is to say flapping foils do not generate peak thrust by flapping at the largest maximum angle of attack for a given Strouhal number. This pattern is evident in the curve of contours which create peaks for each Strouhal number. Matching the general pattern of the prior contour plot gives a preliminary indication that this testing system will be able to recreate past data once the suggested modifications are implemented.
The assumed d CoP values from root to tip of the foil show plots with maximum that range from about 0.9 to 2.5. The past data in Figure 2 has a maximum of about 2.0, therefore a specific d CoP can be linearly interpolated to produce contours with matching magnitudes. This distance is at 17% of the span length or 0.223 m from the roll-axis.
Figures 30-32, show the dependency of thrust generation on Strouhal number. Figure 33 is a plot generated with the same data as Figure 31, but with contours of lift amplitude coefficients ( ). Lift amplitude is found by taking half of the summed maximum and minimum lift values. Therefore the formula for these coefficients is: Where F Z is the phase averaged lift values, and the denominator is the same as in the formula. With data calculated using the interpolated 17% , the instantaneous lift and thrust was plotted to compare to data taken in [Polidoro, 2003]. Significant differences exist between the test apparatus from that thesis and this present work.
Testing was done with the foil piercing the surface of the water instead of being part of a fully submerged structure. The roll angle was varied in the past and kept constant for present testing. The ratio was 1.6 in Polidoro, while this project matched the ratio of Techet, at 1.5. The towing speed, span length, chord length, Strouhal number, and maximum angle of attack have all been matched. Below is the time sequence lift data from the two projects as Polidoro does not provide non-dimensional time data.
The frequency difference is a function of the difference in roll angle used to achieve the same Strouhal number.

Figure 34: Polidoro Lift Comparison
The range of generated lift between the two theses is very close and the waveforms have similar anomalies. The average peak-to-peak magnitude from Polidoro, is 10N, as compared to 9N for this experiment with the 17% . Each plot shows a pronounced "bump" on the slopes of each wave with the larger being on the side of the wave with a positive slope. Also, each plot shows larger negative peak magnitudes than positive peak magnitudes. That discrepancy should not exist and is likely caused by homing misalignment.

Figure 35: Polidoro Thrust Comparison
The figure above shows the same comparison as before but this time for thrust generation. The same discrepancies are evident in peak magnitude as well as frequency. For thrust, peak-to-peak magnitude from Polidoro is 8N as compared to 5N from this experiment. Thrust is expected to be about twice the frequency of lift while being almost only positive. This is because thrust in a single direction is being produced on both upstroke and downstroke whereas lift is being produced in opposite directions on opposite strokes. The total lift-to-thrust ratio from Polidoro is 2.08, the total lift-to-thrust ratio from this thesis is 2.11. This comparison of ratios is compelling because it is unaffected by the . Both plots adhere to theory, and it the approximate similarity between peak-to-peak magnitudes again show this test-system has the potential to recreate past data.
To further illustrate the relationship between generated lift and thrust, the  The second trend of note is the thrust in relation to both St and α max which is also clear in the contour plots. Thrust increases and then tapers off through the α max range. At α max = 40 o the thrust peak magnitudes are much lower than those of lift even though this is relatively aggressive flapping. Conversely, thrust peak magnitudes are very close to those of lift when α max = 30 o for both the 2 nd and 3 rd rows of plots. When both magnitudes are similar, the α max is likely one of the values on the peak of the contour curves. That is, the α max correlates to a max for a given Strouhal number.

Introduction
The force sensing system has a design flaw that produces unacceptable error in linear force measurements. Significant design changes are required to the pitch cylinder components for the system to perform correctly. With these modifications, the current tow-testing apparatus and analysis programs can still be used as they are for future experiments, though some adjustments could be made to increase repeatability.

Modifications Required for Accurate Linear Force Measurements
The current method of producing lift and thrust data, mentioned in the Methodology Section, using moments and assumed centers of pressure should not be necessary. The reason it was used is that there is too much noise in the force signals to accurately measure forces applied directly to the fin. This is likely caused by too little lateral distance between the sensors. Currently, the sensors are 3cm apart. That distance was caused by limited available space to install the sensors without having to heavily modify the organization of the pitch cylinder components. The force sensors should now be moved as far apart as possible (~12cm), an increase in distance by a factor of ~4. This should significantly reduce the signal to noise ratio by the same factor.
Another possible source of force measurement error comes from the sensors being bolted between two rigid plates. The forces on the fin are being translated through the top plate to the sensors which could be applying a moment at the bolts as the plate is bent. The solution to this is to cut the plate in half and create a gap between the two halves such that applied forces to the new plate sections will not interact with each other. Similarly, moments could be applied during these tests by the two rigid pitch shaft bearings. To remove the ability for these bearings to apply moments, the rigid bearings will be replaced with spherical bearings.

Modifications for Better Repeatability
Another adjustment will be to change the location of the pitch shaft homing flag. The homing flag was originally set with the assumption that the fin could move through its full range of motion, and then the program could work from a known zeroposition. For testing on this system and especially during experiments near a wall or bottom, the range of motion for the fin is constrained. However, the zero-position should be the same for every test and setting the homing flag to that new position with high accuracy and repeatable precision could eliminate the discrepancies in peak lift and thrust magnitudes shown in the data plots. This involves rotating the copper flag and tightening the collar that holds it so that it says fixed during testing. While homed though, the plane running through the center of the foil span lengthwise must be able to consistently reset parallel to the tank bottom. A more accurate 0 o reset would reduce the discrepancy between downstroke and upstroke force magnitudes.

Error Analysis
This tow-test system employs many instruments which all have factory quantified tolerances. Those tolerances certainly account for an important portion of the total error, though most cannot be improved and are likely negligible on the scale used for acceptable results. Vibrations in the system likely cause significant noise in the force measurements. Much of this noise can be filtered out while still preserving the waveform, but there is no way to completely quantify or remove many of the vibration sources. The flapping foil is mounted to the bottom of the carriage attachment without any fairing for all testing in this thesis, but the profile is blunt enough that the pontoon setup could direct flow around the towing package with reduced noise. This will likely be tested during future experiments.
The carriage motor is very old a worn belt and gears which disrupts smooth carriage motion, this contributes to the inaccurate speed adjustment as described in the testing procedure section. The roll and pitch motor mechanisms can also add error like that of the carriage. The motors, encoders, gears, and belts have some tolerances and possible damage due to age. Specifically, the belts connecting the pitch motor to the pitch shaft are tightened by hand which means there could be some delay in what the motion controller tells the motor to do and what the actual motion of the foil is. These types of errors can be monitored and pieces can be replaced, but they can still account for anomalies that show up in the force output waveforms.

Introduction:
This future of this test system is focused on the ability to easily adapt and run new experiments. Due to the sources of error mentioned, and room for improvement, the testing setup can and will be modified. Ground-effect testing is set to be the first new experiments. Then, the effects changing the shape of fin in different ways will be examined. There is even research and designs for embedding pressure sensors in the molded fins. Each of these sets of experiments explores areas of research with little if any previous data.

Flapping Orientation
A change will be made to see if there is a noticeable difference between flapping oriented towards the wall (horizontal wall testing) compared to flapping oriented in the center of the tank (vertical wall testing). The latter, along with adding water to the tank is expected to further reduce the possibility of fluid boundary effects on the foil. To accomplish this, the flapping foil system is require d to be rotated 180 o which means the masts will rotate similarly. While there should be no noticeable differences between the masts towed forwards or backwards, it is better to be consistent with the design. Of course, with the rotation of the masts and flapping foil, the carriage will now be moved in the opposite direction starting from the previous back of the tank and moving towards the beach. No extraneous issues should arise from this change other than taking caution not to actually hit the beach with the submerged system during testing. Markings are already in place to indicate where the carriage can go, but future experiments may change the beach profile.

Ground-effect
The main purpose of this entire test-system is focused significantly on the ground-effect experiments. Ground-effect (GE) is a fluid flow phenomena created when airfoils are near a solid surface. The vortices formed under the airfoil compress instead of fully forming like those formed above the foil. This effect can produce increase lift, and decreased drag. GE occurs when is less than 0.5 with being the height from the fin to ground, and c being the chord length. This means that sensing GE could indicate when a vehicle is near a solid surface, or could even allow the vehicle to utilize the benefits of GE in real time.
GE has been extensively researched as it relates to airplanes and helicopters.
Russian Ekranoplanes are ground-effect airplanes that flew very close to lake surfaces while carrying much more weight with less necessary thrust power than a similar sized high altitude plane. The image below shows an Ekranoplane with rocket launchers on the top which would not be very aerodynamic, and very heavy. The cockpit in the picture can be used to scale the massive size of the plane further alluding to the extra available weight due to GE.
Due to the rolling motion of the flapping foils, the fin will be moving alternatively closer to and farther from the flat surface (bottom of tow tank) during tests. Therefore, the hypothesis is that GE should only be visible on flapping foils when the stroke moves the foil into the aforementioned 0.5 ratio. Also, the downstroke should have greater effects than the upstroke due to forcing vortices downwards towards the bottom.
Each set of ground-effect tests will be run with the flapping foil submerged at different distances from the bottom of the tank (h ground from ). The h ground to begin with will be with the beach fully lowered as it is currently and without the cube spacer on the carriage attachment. This will be the open water test. The easiest depth adjusting is done by simply adding the cube spacer so that the h ground is one span length lower than in the open water case. From there, the beach could be adjusted both with the cube on or off for finer yet more difficult h ground tuning.
The goal of this testing will be to be able to detect ground-effect instantaneously without the need to compare to past data. This goal is a direct reflection of a paper testing ground-effect on heaving and pitching foils [Licht & Dahl, 2013]. The ability to sense ground-effect on a test setup in a tow-tank with linear motion is significantly different from being able to do the same on a vehicle in a dynamic environment. Further research and testing would be necessary to differentiate ground-effect from other forces, and to develop a method to integrate this feedback with the vehicle controller.

Alternate Fins
There are multiple ways to change the fins used which can each produce thrust, lift, and efficiency results that could be used to enhance underwater flapping foil vehicles. The most significant changes would likely come from different planform shapes. From the introduction, the current fin design is shaped like a turtle fin. That planform has some advantages as it sheds tip vortices better than a rectangle would, but the chord changes so often which could result in lost thrust and lift. Therefore, tests can be done with shape from rectangles as a baseline up to more biological fin shapes for comparison. The current fins were molded 5+ years ago and used enough to show regular wear and tear. The molded material is flexible, but molding a foil around the internal frame could have created air pockets and flexibility that varies instead of remaining constant. Therefore, the effects of changing the internal frame design and properties of the molding material could be tested on this system. The three main parameters that control the size of the fin are the span, chord, and airfoil used. Since the R-CUE lab is currently working on molding and 3-D printing new fins, they could be made with different aspect ratios. A higher aspect ratio achieves lift and thrust while minimizing losses due to tip vortices, however they would need more power to operate and more space. Conversely a lower aspect ratio allows for more narrow and finer maneuverability while sacrificing the efficiency of the fin.
The current airfoil is a NACA 0012, but depending on the mission or vehicle operation environment other NACA foils could provide better results. Even adding some camber to the foil could be advantageous if the vehicle were to require very high lift in one direction. Having the ability to swap foils could give a single vehicle the ability to maneuver in a variety of environments or just enhance mission-specific operations.

Torque Sensor
A Kistler Type-9039 torque transducer similar in make to the force sensors will be installed on the pitch shaft to allow calculations of efficiency with those torque measurements. This sensor will be compressed between collars and spring washers to hold it in place and to be properly preloaded. The cable from this sensor will attach to a small amplifier mounted on the roof of the pitch cylinder. Since the sensor will be rotating with the pitch shaft, there needs to be a way to always know how far the shaft has rotated from starting position so the cable is not stretched tightly or unplugged.
The communication cable receiving power and sending voltage measurements to the surface must be installed which means a new wet-mateable connector will have to be connected to include the necessary conductors and make sure to keep the pitch cylinder waterproof.

Embedded Pressure Sensors
The force sensors in the pitch cylinder are very useful in measuring the fluid forces acting on the foil as it flaps. However, they do not give any information regarding the fluid flow field over the foil. This can be done by embedding pressure sensors in the foil either during or possibly after molding them. The idea is to create an array similar to the lateral line sensing of many fish. A method using the same analogy with just two pressure sensors was applied to a robotic fish [Ježov et. al. 2012]. In that case, they were measuring periodic vortices (pressure maxima) generated by fluid flow past a bluff body. They found increased swimming efficiency in synchronizing the flapping of the fin to the matrix generation period. Another study [Venturelli et. al. 2012], used 10-sensor arrays on two sides of a model fish and could distinguish between steady flow and Kármán vortex streets. Other students in the R-CUE lab already have extensively researched related literature, possible pressure sensors to use, and have designs on how to integrate them into the fins. Instantaneous pressure sensing like this could greatly improve obstacle avoidance capabilities.

SUMMARY AND CONCLUSIONS
The goals of this thesis were to design, construct, test, and validate a system to research underwater flapping foils in a tow-tank. The design had to withstand large static and dynamic forces without being too cumbersome to operate efficiently. The materials were selected for corrosion resistance and to be streamlined while submerged in the tow-tank. Force sensors were installed in the pitch cylinder of the flapping foil, calibrated in-situ, and connected to the data acquisition system on the carriage platform. This data acquisition system was setup to record the force measurements, poll the motion controller of the flapping foil, record laser distance measurement, and then send it all via fiber-optic cable to computers. The files created by that data were processed and analyzed to produce the Results chapter.

Introduction
This Appendix section will explain the processes necessary to run tests, collect data, and analyze the data to form results. There have been significant improvements made to the apparatuses used as well as the collection and analysis programs while testing has been going. This means there could have been discrepancies between testing methods. These differences will be noted, and the corrections will be explained to show integrity of the data collected.

Testing Setup
Before putting the fin and carriage attachment (CA) in the water, it is important to check that all connections work, and that the pitch cylinder is watertight.
Testing the connections just involves turning the DAQ Chassis power on and the 2channel power source on, and then using the computer to run a simple motion while recording force sensor data. Once this is confirmed, a vacuum test is performed on the pitch cylinder. There is a bolt on the end of the cylinder that goes through the delrin into the interior. This bolt should be removed along with its o-ring, and the vacuum suction should be applied to the bolt hole. Once the vacuum reaches one atmosphere of pressure, and is turned off, the cylinder is watertight if the pressure does not change for an extended period of time. 5 to 10 minutes should be acceptable. Then, the bolt and o-ring can be replaced so the installation process can continue.
The installation of the CA with flapping foil has been outlined in the CA section, Chapter #. Once the structure has been bolted down, the wiring connections are fairly straight forward. The outputs from the force sensor Cat 5 cables connect to the labeled inputs for the I/O boards, and power and ground wires for the force sensors and the foil are connected to their respective channels on the power source. The laser turns on automatically when the chassis gets power. Before the power source and DAQ chassis are turned on, the fin must be rotated to the upper roll limit with 0 o pitch.
This positioning is the origin set as (0,0) by the control card which allows the user to set a desired starting position relative to the same origin for every test. With everything powered up, the computer on-land can then be turned on. If the computer is turned on prior to those steps, it may not recognize the carriage platform chassis. In Galil on the computer, for the testing in this project, the starting position of the fin from the origin mentioned is set using these commands: This means the absolute position of the fin (PA) relative to (0,0) is -5200 counts of the roll motor, and no change in the pitch. BG begins the motion. It is always good to constantly check the position of fin whenever attempting to return to this point.
Resetting after each test should be close to -5200,0 using the command TP for tell position. However, since there are 50,000 counts in the roll motor and 102,000 counts in the pitch motor, the difference in position can vary by up to 100 counts without even being off by a single degree so some leeway is granted.
Presumably, the setup so far has been done with the carriage in a position other than the correct starting position, so the next step is to move the carriage there. This is done using the carriage motor controller with switches for direction, starting and stopping the motor, and a potentiometer that regulates carriage speed. The starting position doesn't always have to be the same, and is based on the depth of the fin compared to the bottom, as well as the desired test travel distance. The travel distance can be calculated on a per-test basis in the equation below by using the flapping frequency, carriage velocity, and desired number of flaps.
There are taped markings along the rim of the tow-tank that indicate 1.5 meter sections to be used for positioning and timing. As mentioned in chapter #, the velocity is constant through each test and therefore can be found with the time it takes for the carriage to travel between two known points. Currently, this is done using the LDM 42, but previously it has been timed using a stopwatch and visually watching the carriage pass those markings. Tests were run multiple times to ensure consistent measurements with velocities varying by mere thousandths of a second per test.

Tow-Testing Process
When everything is ready to run a test, the program FSCollect is run in LabView to record force sensor output for 20 seconds. With no foil motion this provides a baseline (zero) as a reference to compare the same output from 20 seconds without motion at the end of the test. The force sensors have a linear drift which can be removed using these two zeroing files for each test. A benefit of running the zeroing program for 20 seconds before each test is it allows the water in the tank at least that amount of time to settle between each test and each carriage reset.
With the baseline zero data recorded, the foil motion is started using Galil Tools Lite (Galil), which is the proprietary program for the control card on the foil system. Control is done by setting a flapping frequency, a roll amplitude, and a pitch amplitude using the commands: Due to the direction of towing and the setup of this particular foil, the pitch amplitude is always set as a negative value in this project's testing. CHANGE = 1 is the command that applies the previous commands. With that done, the foil will start flapping, but Galil cannot stay open during data collection due to the collection program's need to take over connection to the foil system control card. So, with Galil closed, the program DataCollect is run in LabView. This program runs the same force sensor collecting and displaying as FSCollect, and it also polls the foil system control card for the motor torque (V), velocity (counts/s), and position (counts) at a rate of 1 polled measurement per 10 force sensor measurements. Testing has been done with the force sensors at 200hz and therefore the polling at 20hz. After DataCollect is started and connected to the foil, the program LDM is run simultaneously to for carriage velocity data. The carriage is then run for the experiment.
The carriage is controlled via the box shown in the image below. The direction switch is set first, the motor start button is depressed, and the velocity potentiometer is turned to the correct position at a calm pace. The carriage should accelerate to the desired constant velocity inside one 1.5 meter section. A hose clamp is connected to the potentiometer (as shown below) to set a maximum velocity limit at the desired test velocity. The excess length of the clamp hits the silver U-bolt to limit the turn of the potentiometer. This is currently the best option for velocity regulation, and the position of the clamp is tested 3 times minimum to make sure it drives the carriage at a consistent velocity within a hundredth of a second variation. The carriage stops in a safely short distance so the stop button can be depressed at the end of the testing travel distance.

Figure 41: Carriage Motion Controller
After the carriage is stopped, any running programs are stopped as well. Then, upon reopening Galil, the roll and pitch amplitudes are set to 0 degrees thus resetting the fin position using commands: It is at this point the FSCollect program is run again for the 20 seconds of end zeroing data. Then, the carriage can be returned to the starting position. At the starting position, all the testing steps can be repeated starting from running FSCollect for the 20 seconds of starting zeroing data.
There is one parameter that is very important to check before each test.
Resetting the fin to the home position is followed by a command "TP" that polls the motor controller for the position in counts from 0 that each motor is at. Clearly this value should be as close to if not exactly 0 every time, but they typically vary by up to ±50 counts on the pitch motor and up to ±20 counts on the roll motor. These variations are generally negligible since there are 102,000 total counts in pitch and 50,000 counts in roll. That means that a difference of hundreds, even a thousand counts in pitch would be required to alter homing by just 1 o . These values should always be monitored, but there should not be too much worry.