PERFORMANCE ANALYSIS OF AN UNDERWATER WIRE FLYING PROFILING VEHICLE

Profiling the water column from a moving vessel places significant limitations on the achievable spatial resolution of the data collected. Current solutions such as underway CTDs and undulating towed bodies are limited by vessel speed and operational water depths. With an increased focus on submesoscale processes (e.g., stirring, mixing, regulation of thermocline structure, chemical distributions along isopycnals and resolution of hydrothermal plumes), there is a growing need to collect profiles with high vertical and horizontal resolution. A new autonomous profiling vehicle has been developed to address this need. The autonomous wire flying profiling vehicle slides up and down a towed wire in controlled manner using the lift generated from actuated wing foils. Using a depth and velocity controller, the vehicle is capable of maintaining prescribed flight paths over a range of ship speeds and tow depths. The vehicle has achieved glide slopes (ratio of vertical to horizontal speeds) up to 2.5 for tow cables angles between vertical and 45 degrees during field testing. This results in a sample path with more horizontal resolution than otherwise achievable with other underway or undulating systems. This paper presents a summary of the vehicle’s capabilities along with field test data from a test cruise to the New England Shelf Break.


Introduction
Ocean processes on smaller horizontal scales are of increasing environmental importance. Scientific research continues to work toward identifying and understanding smaller scale processes in the ocean. As scientific research advances, the technology through which that research is conducted must follow suit. Existing technologies have limited ability to profile the water column with the horizontal and vertical resolution needed to capture many of the relevant spatial scales.
On horizontal scales between 10's of meters to 10's of kilometers, critical ocean processes occur such as stirring and mixing to regulate ocean thermocline structure, chemical distributions, and meridional overturning throughout the water column (Crawford 1986, Schiermeier 2007. As an example, in the deep ocean the enhancement of vertical mixing over regions of rough topography has been documented via measurements of turbulent dissipation rates (Polzin et al. 1997). This has significant implications for the abyssal stratification and circulation (Jayne et al. 2004). Using traditional microstructure instruments to measure turbulent dissipation rates in the deep ocean is very difficult and time consuming due to the deployment time required for such instruments. Water mass and current formations are also greatly influenced by topography. Evidence of topographic effects extends all the way to surface currents via the curvature of the current flow in response to bottom topography (Knauss 1997). Understanding the linkage between these processes is crucial to efforts aimed at predicting the consequences of the warming and freshening of highlatitude surface waters to the climate system (Lozier 2012 Figure 2). Flow simulations were used to hydro-dynamically optimize the WiFly body and wing foil shapes (Amaral 2013).
A model representation of the WiFly Profiler is shown in Figure 3 and the full scale prototype vehicle is shown in Figure 4. The vehicle has a total dry weight of 90 kilograms and an in-water weight of -2 kilograms. The internal structure of the vehicle is aluminum. The exterior shell of the vehicle is comprised of a combination of syntactic foam, machined plastic and 3-D printed plastic pieces. Due to the autonomous nature of the WiFly, an on-board control system is necessary to control the vehicle and perform pre-set or acoustically updated profiling tasks. The control system is a linear Proportional Integral Derivative (PID) control system to control depth or velocity. The system uses measurements of the vehicle's depth, tilt and speed through the water. Wing angles are adjusted continuously, within limits, to achieve the desired depth or vertical velocity. The wings can also be set to fixed angles if preferred.
The desired profiling behavior is setup in a mission file prior to each deployment.
The missions are broken into small legs, each with prescribed up and down profiling rates and depth limits. While underway, the acoustic modem can be used to modify the mission by adding legs, stopping legs, or issuing an abort. All science data collected during each mission along with a number of vehicle control parameters are logged and stored on board the vehicle and then downloaded post-dive. (1)

Vehicle Dynamics
Additionally, using a total speed through the water ( ) can be calculated independently of the vehicle's speed sensor. This total speed should be equal to the apparent velocity. Comparing the two velocities provides a validation of the speed sensor.

√ (5)
Another important parameter in defining performance guidelines is stall. Stall is defined as a reduction in the lift coefficient generated by a wing as the angle of attack increases (Crane 1997). There is typically a critical angle of attack for a wing foil, beyond which stall occurs. Most studies conducted to determine the critical angle of attack suggest an angle of near 15 degrees. The exact angle varies depending the specific wing shape and the Reynolds number.
The shape of the full scale body and wing were selected based on a study of drag coefficients over a range of scale model body sections (Amaral 2013). It was predicted that a lower drag vehicle will be able to obtain higher glide slopes across a range of tow cable angles. The current shape was selected to have low drag over a range of flow angles between 0 and 50 degrees.
While climbing and descending the tow cable, the WiFly vehicle will experience different forces generated by the wings. As the cable angle is increased, generating lift in the direction of motion becomes increasingly difficult. Figure 6a shows a schematic of the forces acting on the vehicle during a climbing scenario at a 30 degree cable angle.
The primary lifting force ( ) acts perpendicular to the apparent flow. This generates a resultant force ( ) that is acting to pull the vehicle and tow cable back and a force ( ) propelling the vehicle along the tow cable. Figure 6b shows a schematic of the forces acting on the vehicle during a descending scenario at a 30 degree cable angle. Again, the primary lifting force acts perpendicular to the apparent flow generating a resultant force that is directed more along the tow cable in this situation. There is very little force being applied perpendicular to the tow cable. As the cable angles are increased the lifting forces are diminished in the direction of motion and the vehicle's ability to climb or descend the tow cable decreases. At extreme high cable angles, will diminish to the point where it can no longer propel the vehicle along the cable. The behavior of such an event will mimic a stall in that the vehicle will stop climbing or descending and experience a drop in . However, in this scenario the wings are functioning properly and stall has not occurred. Additionally, the forces acting on the tow cable have the potential to alter the cable's shape.
To account for the shape of the cable it is necessary to calculate a layback distance to properly locate the vehicle behind the ship. The shape of the cable will change as a function of depth with the ship speed and horizontal speed of the cable. Repeated profiles yielded similar cable angle measurements for both upwards and downwards motion (Figure 7). When a sufficiently heavy clump weight is fixed to the cable, the forces from the vehicle acting on the cable are not sufficient enough to alter the shape of the tow cable between upwards and downwards profiles. Using the cable angle at a given depth and the difference in depth from the previous data point, an average cable profile can be calculated. From the top of the cable profile, a horizontal layback distance was then calculated for each depth. A sample cable layback shape can be seen in Figure 7b.
GPS coordinates were then corrected with this offset. Figure 8 shows the change in vehicle profile when correcting for layback.

Profiling Performance
To evaluate the flyer concept, the vehicle was tested aboard the R/V Endeavor during November 2013 at the Rhode Island Shelf Break (Figure 9). The profiling performance was tested by conducting multiple dives at depths ranging from 10 to 800 meters. The test results are summarized in the following sections.
To validate the glide slope predictions, initial missions were conducted with the vehicle stepping through a series of wing angles at various ship speeds and wire angles.
A sample section of data from such a test is shown in Figure 10. Different ascent and decent rates were achieved as the wing angle was varied. Vertical speeds reached 3.7 meters per second moving upwards and 2.3 meters per second in the downward direction for a mean ship speed of 1.5 meters per second.
Data from a small section of a velocity profile plotted against the along track distance is shown in Figure 11. During this scenario, the vertical velocity was being regulated via the control system by continuously adjusting the wing angle. Figure 11b shows the cable angles varying between 0 and 45 degrees during profiles between 200 and 360 meters. is due to the layback of the cable behind the ship. Apparent velocities during downwards movement are less than during upwards movement ( Figure 12). This is due to the shape of the cable. As the vehicle moves downwards, the cable is both moving forward through the water while extending further away from the ship. Additionally, it is more difficult to generate sufficient lift during downwards movement.
Data from a hold depth scenario at 10 meters is shown in Figure 13. During this time the clump weight was being lowered and the tow cable was passing through the vehicle as it held depth. As a result, the cable angle began to increase. The ship speed increased and the vehicle began to experience an increase in apparent velocities. The control system was able to adjust the wing position accordingly and successfully hold the depth.
To assess the overall performance envelope of the vehicle, all steady state sections of the dives were extracted and separated into upwards and downwards segments. The achieved glide slopes were calculated from these data. Segments with incomplete data or significant variations in ascent or decent rates were not considered. Additionally, data near the transition from an upward motion to a downward motion and vice versa were removed. Under these steady state conditions, the vehicle achieved glides slopes ranging from 0.1 to 3.0 while climbing over a range of tow cable angles between 0 and 55 degrees. Glide slopes ranging from -0.1 to -3.0 were achieved while diving over the same range of tow cable angles.
The results for steady state, upward motion glide slopes are shown in Figure 14. For cable angles less than 20 degrees, the achieved glide slopes are higher than predicted in Figure 2 and there is no clear dependence on along cable velocity. Glide slopes of greater than 1.5 were commonly achieved at along cable speeds less than 2.0 meters per second. As the cable angle was increased, some dependence on along cable velocity becomes apparent. Faster along cable velocities resulted in higher glide slopes with the fastest velocities coinciding with the boundary of glide slope performance. The glide slope decrease at higher cable angles was as predicted and overlaps with the curves in Figure 2 at high cable angles. As expected, there is a direct relationship between wing angle and glide slope (Figure 14b). Larger wing angles resulted in higher glide slopes across all cable angles.
For downward motion of the vehicle, the glide slopes were calculated separately and are shown in Figure 15. Results are more uniform than expected with along cable velocities of 0.5 to 2.0 meters per second frequently resulting in glide slopes higher than -1.0. Performance at cable angles greater than 40 degrees exceeded the upwards motion glide slopes at the same cable angles. At low cable angles (less than 10 degrees), the WiFly performance was again better than expected. There appears to be no dependency on along cable velocity for achieving specific downward motion glide slopes. Higher wing angles again resulted in higher achieved glide slopes (Figure 15b). It should be noted that these tests cannot be considered exhaustive, but do represent the most common operating scenarios over the entire test cruise. Also, the point density in glide slope plots is due to the fact that some particular profiling missions were run for longer periods of time compared to others.

Stall and Angle of Attack
In order to evaluate the limits of the WiFly vehicle, stall was intentionally induced multiple times. A stall like event at high wing angles is shown in Figure 16 and Figure  17. During this stall event, the vehicle was profiling between 350 and 600 meters depth and the cable angle exceeded 50 degrees. The target velocity was reached early in the ascent and appears to stabilize with a wing angle of 40 degrees (Figure 16c). At 300 meters, a decrease in the vertical and apparent velocity occurs and the angle of attack begins to increase past the point where stall is likely for this wing shape (Figure 17b). The linear control system responds to the reduced velocity by increasing the wing angle to generate more lift, resulting in a further increase in the angle of attack. Although full flow separation from the wing is not discernable, the reduced lift during stall reduces the upwards velocity ( ) as the upwards along cable force ( ) is overcome by increased drag. At the upper depth limit, the vehicle begins a downward segment and recovers from the stall.
Angle of attack was calculated for all dives as in equation (4). By calculating the angle of attack, performance guidelines could be defined for the control system to include angle of attack limitations to enhance vehicle efficiency. A high cable angle vehicle profile is shown in Figure 18. The angle of attack is shown in Figure 18f. Positive angles of attack correspond to upwards motion and negative angles of attack correspond to downwards motion. As the vehicle begins the upwards motion, the wing angle and angle of attack dramatically increase. As the desired velocity is reached, the angle of attack decreases and the vehicle enters steady state flight. In general, this is the more common trend in comparison to Figure 17, with the angle of attack typically less than 15 degrees.
In some instances however, as seen in Figure 17b, the angle of attack is calculated as negative during upwards motion. From the free body diagrams in Figure 6, if upwards lift is being generated then a negative angle of attack is physically impossible and can be explained if the apparent velocity measurement ( ) is being under predicted by the speed sensor. An under predicted would result in a smaller calculated horizontal flyer velocity ( ) and consequently a larger calculated glide slope. In order to evaluate the accuracy of the speed sensor measurements, horizontal flyer velocities were determined independently from shipboard ADCP horizontal velocity measurements. Once an estimate of the horizontal velocity was determined from the ADCP, a new or apparent velocity was calculated using equation (5).  Figure 19b shows the percent error between the sensor and ADCP calculated apparent velocities. For this section of data, the mean error is 5.2 % ( Figure   19c). This suggests that on average, the speed sensor is under predicting the apparent velocity measurement by about 5% when compared to the ADCP data. Over the entire set of steady state data used in determining glide slopes, there are times where the speed either under predicts apparent velocity such as in Figure 19 or over predicts apparent velocity. Over prediction seems more prevalent at higher cable angles ( Figure 19b). As such, new apparent velocities were calculated using the ADCP data for all steady state data used in the previous glide slope plots. Figure 20 shows the achieved glide slopes calculated using ADCP horizontal velocities. Figure 20a shows data from steady state, upwards motion corresponding to the data in Figure 14. The ADCP horizontal velocities result in lower calculated glide slopes. The max glide slope during upwards motion was 2.4. However, the achieved glide slopes now are more represented of the prediction curves in Figure 2. At all three cable angles where the model was tested (15 O , 30 O , and 45 O ) the vehicle achieved higher glide slopes than predicted. Additionally, a stronger correlation with apparent velocity is noticeable, with the highest apparent velocities located along the edge of performance of the vehicle. Figure 20b shows data from steady state downwards motion corresponding to the data in Figure 15. Again, using ADCP horizontal velocities results in lower calculated glide slopes. The max glide slope during downwards motion was -2.0.
Results are uniform with along cable velocities of less than 1.5 meters per second resulting in glide slopes of -1.5 over cable angles from 0 degrees to 35 degrees. At high cable angles, performance was better than expected.

CTD Data Resolution
Subsections of CTD temperature measurements taken along an east/west transect near 39.8 degrees N latitude are shown in Figure 21.

Testing Limitations
Although field tests were successful in proving the WiFly concept, there existed several limitations. These tests were designed to test the WiFly's performance under a wide range of wing angles and velocities. In order to achieve this, the control system employed on the vehicle was designed to be linear in repsonse. However, this frequently resulted in a slow response to the desired velocity seen in Figures 11c, 16c and 18c.
Additionally, the lack of a stall limiter sometimes allowed for a stall to occur unintentionally during testing, resulting in decreased performance or an aborted profile.
When large amounts of cable are towed, the cable layback shape can be dramatic. This can result in a significant offset distance between the recorded position and the actual position of the vehicle behind the ship. The layback correction can account for this but it is very difficult to independently verify that the interpreted tilt measurements are predicting the actual cable shape at any given depth.

Conclusion
Oceanographic research is continuing to press toward understanding and identifying smaller scale processes. The need for an efficient method to sample the water column with both vertical and horizontal resolution in order to aid in this quest has become more apparent. This paper has outlined the results from field testing of a new autonomous wire-flying profiling vehicle designed to address this need.
Field testing has validated the wire flying concept as a novel and efficient means to sample the water column. Comparable existing technologies are limited by vessel speed and deployment depth resulting in achievable glide slopes typically less than 0.25. The WiFly vehicle was able to consistently achieve glide slopes from 0.1 to 2.0 over a range of cable angles and water depths. Additionally, the WiFly was able to traverse the tow cable at controlled vertical speeds between 1.0 and 3.0 meters per second from shallow water down to 800 meters depth. The vehilce was able to follow control commands and maintain steady state flight for extended periods of upwards and downwards motion. Furthermore, the WiFly was able to increase the profiling spacing by a factor of three or more when directly compared with the Oceanscience UnderwayCTD.
Field testing helped to determine the limitations in the WiFly system, which will aid in refining the technology. Better estimation of the angle of attack will greatly improve the vehicle's performance and better predict stall events. The control system employed during testing was not designed to fully optimize the vehicle's performance. A next iteration will certainly improve performance.      . Free body diagram for the wing forces generated during climbing and descending situation. The wing produces a net force F, with lift and drag components relative to the apparent flow velocity. The net force has projections along and perpendicular to the wire. (a) is able to pull the flyer up the cable and there is no significant force perpendicular to the cable. (b) is pulling down the cable and there is relatively little force perpendicular to the cable.