1. Introduction

Autonomous underwater vehicles (AUVs) are unmanned, untethered submersibles that can follow a preprogrammed path using sensory input without direct human control. They are being developed as platforms for oceanographic instrumentation as an alternative to traditional research vessels (Allen et al. 1997; Levine et al. 1997; Yoerger et al. 1997; Smith et al. 1995; Bellingham et al. 1994). Without a tether an AUV is able to navigate freely and, once below the region influenced by surface waves, is isolated from surface motion. This improves oceanographic data quality by removing motions associated with ship heave and allows for continuous data collection along an arbitrary subsurface path.

An appropriately chosen vehicle path can provide spatial information in one, two, or three dimensions. When sampling in one dimension, resolution is determined by the vehicle speed and the sampling period. When sampling in two or three dimensions, the overall resolution is determined by the spacing of vehicle track segments. The overall resolution and extent of sampling is constrained by a combination of the AUV’s speed and range, as well as by the largest length scale and the shortest timescale of the phenomenon to be resolved (Bellingham and Willcox 1996). When sampling an area or volume, greater vehicle speed allows finer overall spatial and temporal resolution, while greater vehicle range allows larger phenomena to be characterized.

However, achieving greater vehicle speed and range requires more stored energy. Onboard energy storage capacity can be increased with either larger or higher energy density batteries. Both options lead to increased cost of operation. Larger batteries lead to larger, more expensive AUVs requiring larger, more expensive support vessels and crew. High energy density secondary batteries are more expensive and cannot be recharged as often as cheaper, low energy density secondary batteries. Primary batteries have a higher energy density than secondary batteries; however, their use is made expensive by the fact that they cannot be recharged.

Reducing an AUV’s size and cost limits the amount of energy it can store in the form of batteries, which restricts it to slower cruising speeds over shorter ranges. Slow cruising speeds and short ranges inevitably diminish the usefulness of small AUVs in the ocean environment where typical currents are 10 cm s⁻¹ or greater, and length scales are of the order of one to thousands of kilometers. However, small lakes provide an environment in which a slow AUV with limited range and speed can have many applications. Length scales of phenomena in lakes are of the same order as the basin scale or less, limiting the required AUV range. Lacustrine currents are typically weak; hence, an AUV can navigate effectively at slow speeds using only dead reckoning. Risk of loss is reduced since small lakes present a limited search region (in depth and horizontal extent) should the vehicle disappear. A pinger–locator system is still required to locate a disabled vehicle in any sizable lake.

This article describes a small AUV, PURLII, and data collected with this vehicle in a small lake. Section 2 describes the construction of PURLII. Section 3 details the integration of a CTD unit on board PURLII, including considerations for maximizing CTD data quality. Section 4 describes the field site and missions conducted. Section 5 characterizes PURLII’s performance, and section 6 presents the temperature data collected. A summary is given in section 7.

2. PURLII

PURLII was designed to be an easily deployed, economic platform for carrying a wide range of oceanographic instrumentation. PURLII has a weight in air of ∼70 kg, and is composed of a flooded fiberglass faring that houses an aluminum pressure vessel, a CTD and pump, a depth sensor, an acoustic altimeter, three thrusters for propulsion, and syntactic foam for flotation (Fig. 1). PURLII and all of its support gear can be transported to a remote location in a light truck or two subcompact cars. A small row boat or canoe, two to three people, and a laptop computer for communication are required. PURLII is transported over rough terrain on a basket stretcher equipped with a single wheel. The stretcher removes the need for a boat launch and requires only that the vehicle transporting PURLII come within walking distance of the lake, thus greatly increasing the number of lakes that it can be deployed in. PURLII is deployed by sliding the vehicle and stretcher into the water and then removing the stretcher.

PURLII has no minimum functional depth limit and can operate at the water surface; it has a maximum functional depth limit of 70 m, which is set by the maximum range of its pressure sensor. PURLII has successfully operated at all depths between the surface and 70 m; its failure depth rating is the minimum rating of the various pressure vessels within the flooded faring (Helland 1997). The rating of the thruster pressure housing sets PURLII’s failure depth at 100 m.

An aluminum pressure vessel holds the battery package and electronics. The battery package consists of four sealed lead acid batteries providing 24 A h at 24 V. Lead acid batteries were chosen because they are inexpensive, easily maintained, and can be cycled many times. The batteries must be removed in order to be recharged. Three battery packs are used: two on charge and one in the vehicle. The battery packs are changed by removing PURLII from the water, opening the main pressure vessel, sliding the electronics rack out of the main pressure canister, and lifting the battery pack from the rack. The turnaround time for battery replacement is approximately 30 min.

PURLII’s power distribution is as follows: hotel load, 36.7 W; vertical thrusters, 50 W on average; horizontal thrusters, 100 W at 50 cm s⁻¹. The CTD has its own power provided by six D-cell alkaline batteries, which are sufficient to power it for 48 h of continuous logging. Thus, PURLII’s total power requirement is 186.7 W. A battery pack is rated at 576 W h and therefore provides power for 3 h.

The navigation sensor suite consists of a depth sensor (based on pressure), acoustic altimeter, and a flux-gate compass (Table 1). Using time, compass direction, and depth information PURLII is able to navigate in a three-dimensional environment along simple paths. The acoustic altimeter allows PURLII to avoid the bottom by a preprogrammed margin. PURLII propels itself forward and backward using two horizontal thrusters located amidships (Fig. 1). Course correction is affected by thrusting differentially. A single vertical thruster located near the bow on the port side is employed to move PURLII vertically in the water column. PURLII is limited to a cruising speed of 50 cm s⁻¹ in the horizontal and 10 cm s⁻¹ in the vertical. Shallow water trials showed PURLII’s motion to be steady; as such, its cruise speed is assumed to be fixed. The horizontal cruising speed was estimated by measuring the time taken for PURLII to travel a known distance in a swimming pool. During the missions described in later sections, PURLII’s horizontal cruising speed was determined from the distance traveled during the prescribed time to the midmission turnaround point. Vertical speed is derived from data from the depth sensor.

The embedded controller for PURLII is a PC-104 stack, which employs a 80486-DX4 100 as the CPU. The stack consists of six cards that control the I/O, data storage, and processing required to control PURLII. The control loop for heading and propulsion control is run by a proportional–integral–derivative (PID) controller. This PID controller accepts the heading set point from the heading set-point selection logic, which arbitrates between the various heading set-point sources. PID controller feedback from the compass together with the heading set point generates the heading control signal. After the heading control signal is converted to a thruster rpm, it is either added or subtracted from the velocity set-point rpm to generate the left and right horizontal thrusts, respectively.

The vertical control feedbacks are depth and altitude. The depth and altitude control signals enter arbitration logic, which determines if PURLII should be controlled by altitude or depth. Similar to the heading control, the vertical control signal is converted to rpm, passed to the thruster interface where adaptive control loops ensure that the vertical thrusters are turning at the desired rate.

PURLII’s control software (PROTEUS) is a real-time scheduler developed by International Submarine Engineering for controlling remotely operated and autonomous underwater vehicles. PROTEUS employs an object-oriented architecture implemented in C++. From the user’s point of view, the most important property of PROTEUS is that the control systems, telemetry, and mission scripting can be modified without working on the underlying C++ source code. The implementation of PROTEUS on PURLII was unique because it was the first time it was installed on a PC-104 system. An AUV’s control system and mission scripts are entirely described by a set of text configuration files that are parsed by the mission executor at run time and executed by the PROTEUS kernel. Because the text files are parsed at run time, configuration changes can be implemented without recompiling PROTEUS. The ability to reconfigure the system is particularly useful while testing and during field work where parameters must be changed to suit new situations. Unlike other “artificially intelligent” controllers, PROTEUS does not pretend to “reason” about the world and change its actions in a heuristic fashion. All control loops and signal arbitration are fixed throughout a mission, and changes to the configuration files must be made offline while PROTEUS is not running.

Data propagation is the fundamental operating mechanism for PROTEUS. Instead of real-time software modules communicating through a typical send–receive–reply mechanism or a global variable blackboard, PROTEUS specifies what actions are performed when a piece of data is updated. If a piece of data (a variable) changes value, an “event” occurs, and this event is propagated to each of the relevant software modules (components).

In addition to event propagation, PROTEUS also handles real-time scheduling of the components once they have been triggered. Each component has a different priority level, and action procedures with higher priority are executed before those with lower priority. When an event, or multiple events, triggers a set of components, the corresponding action procedures are queued on PROTEUS’s scheduler waiting list. Actions are always processed to completion before another action is started unless a hardware interrupt preempts the current action and places a higher priority action into the scheduler’s queue.

The missions described here were trials of PURLII, and no means of locating it should it disappear were implemented. To ensure PURLII’s recovery, a 40-m-long, 2-mm-thick nylon line was attached to its stern. A 10-cm-diameter spherical float was attached to the other end of the line. A segment of 2-mm line was attached to the float and paid out from a following support vessel such that the line between the float and the canoe was slack. This system had the effect of slowing PURLII’s cruising speed to about 35 cm s⁻¹ and, as will be discussed later, added a bias to PURLII’s heading.

3. CTD installation

Temperature, conductivity, and pressure data were collected with a Sea-Bird Electronics Inc. Seacat SBE19 CTD profiler, using a 3000-rpm SBE5 centrifugal pump and the recommended Sea-Bird ducting (Sea-Bird 1992). The thermistor has a calibrated accuracy of 0.01°C, a resolution of 0.001°C, and adjusts to 63% of a step change in temperature in 0.58 s. The pressure sensor has a maximum depth rating of 334 db, a calibrated accuracy of 0.85 db, and a resolution of 0.05 db. The conductivity cell can be used in two modes: normal range (0.0–6.5 S m⁻¹) for oceanographic applications and low range (0.0–0.6 S m⁻¹) for freshwater applications. The normal conductivity range has a resolution of 10⁻⁴ S m⁻¹, while the low conductivity range has a resolution of 1.5 × 10⁻⁵ S m⁻¹.

By using the complete CTD logger assembly, there is no need for PURLII to log CTD data. As well, the CTD can be removed and used conventionally when needed. The pump allows the CTD to be housed within the AUV’s fairing, reducing vehicle drag. The pump and the duct together help reduce salinity spiking due to temperature and conductivity sensor time response mismatch. The pump provides a constant flow of water, which ensures that the conductivity sensor’s response time is constant, and the duct forces the parcel of water sampled by the temperature sensor through the conductivity sensor. See Horne and Toole (1980), Topham and Perkins (1988), and Gregg and Hess (1985) for a discussion of the problem of salinity spiking due to temperature and conductivity sensor time response mismatch. Other in situ water property sensors designed to operate with plumbing can be easily integrated into PURLII with the CTD acting as datalogger.

Lake water is drawn through a 36.5-cm section of 0.79-cm internal diameter tubing, over the temperature sensor, along the ducting, through the conductivity cell, through a section of 1.1-cm internal diameter Tygon tubing, through the pump, and then exhausted (Fig. 1). In order to avoid the effects of the AUV’s hull on the water sample, the CTD intake was located 13 cm aft of the vehicle nose (15 cm when measured along the hull) and 1 cm away from the hull.

An upper bound for the extent of the region of influence of the hull is the boundary layer thickness estimated for a turbulent boundary layer along a flat plate with no longitudinal pressure gradients. An estimate for the thickness of a flat plate turbulent boundary layer is given by (White 1986)

View Expanded

where δ₉₉ is the vertical distance above the flat plate where the velocity is 99% of the free stream velocity; x is the distance along the plate; and Re_x = Ux/ν is the local Reynolds number, where U is the vehicle velocity, and ν is the kinematic viscosity of water. For PURLII’s cruising speed of 35 cm s⁻¹, x = 15 cm, and ν = 10⁻⁶ m² s⁻¹, Re_x is 5.3 × 10⁴, and δ₉₉ is 0.51 cm. As δ₉₉ is less than the 1 cm that the intake extends from PURLII, the vehicle’s boundary layer is not expected to affect the sample water under normal operating conditions.

To avoid the effects of PURLII’s vertical thruster on data quality, the CTD intake was placed on the side opposite the thruster. While the vehicle is being driven up and down it will be moving away from thruster-induced turbulence. Only when PURLII is changing direction from up to down or down to up will the thruster-induced turbulence potentially be in the vicinity of the CTD intake. The situation of PURLII moving vertically in one direction while the thruster is pushing the opposite direction occurs for less than 6 s (based on thruster status recording resolution) at the end of a profile. This 6 s represents the final ∼60 cm of a profile.

Sample water travels 36.5 cm between the intake and the thermistor. Ideally, this distance should have been shorter, but space limitations within the vehicle and the position of the vertical thruster dictated the location of the CTD. The mean flow rate of water within the sampling system was estimated by setting up a similar configuration on a bench and measuring the time taken to fill a known volume. Using this method, the flow rate was estimated at 25 mL s⁻¹. The volume of the plumbing up to and including the conductivity cell is 24.3 mL. Therefore, it takes about 1 s to flush the plumbing up to and including the sensors. In this time PURLII moves 35 cm horizontally (50 cm when PURLII is not towing a float) and 10 cm vertically, hence setting a lower bound on the spatial resolution of CTD data along the vehicle path.

In order that the flow of water through the plumbing be independent of vehicle velocity, Sea-Bird recommends that the intake and outlet should be positioned such that they experience similar dynamic pressures under most vehicle operation. The intake and exhaust are both oriented normal to PURLII’s hull and located along the port side of PURLII. Hence, the only dynamic pressure differences between the intake and the exhaust will be due to turbulent pressure fluctuations along the hull. Since PURLII travels at 35 cm s⁻¹, turbulent pressure fluctuations along the hull are not expected to significantly modify the flow rate in the plumbing.

Since the Sea-Bird pump is not powerful enough to self-prime, the plumbing must be flooded before the pump is turned on and the intake must remain submerged. The plumbing is configured with no loops or kinks so that priming is accomplished by tilting PURLII beneath the water until any trapped air is expelled. Since the intake is below the water surface when PURLII is properly ballasted and floating freely, automatic on/off control of the pump is not required.

4. Field tests

Experiments were conducted at Loon Lake, which is located 50 km east of Vancouver in the University of British Columbia Malcom Knapp Research Forest (Fig. 2). It is the largest lake in the research forest, with a length of 1.6 km, a width of 0.3 km, a mean depth of 26 m, and a maximum measured depth of 55 m. There is a small (500 m × 200 m, ∼15 m deep) arm extending to the southeast near the south end of the main body of the lake. The arm is connected to the main body of the lake by a shallow sill that has a maximum depth of 2 m.

Two thermistor chain moorings, labeled T2 and T4 in Fig. 2, were installed in Loon Lake. The thermistor chains, built with Richard Branckner Research Ltd. (RBR) TR1000-FR self-recording thermistors, were moored in Loon Lake for a 5-week period (15 November to 20 December 1996) and recorded temperature once every minute. The PURLII missions were conducted roughly in the middle of this 5-week period.

PURLII completed five missions on three separate days: 26 and 27 November, and 2 December 1996. Each mission consisted of a return trip along the long axis of Loon Lake. PURLII was programmed to move up and down between 10 and 20 m while following a particular compass heading for a specified period of time. When the specified period of time elapsed, PURLII surfaced and turned through 180°. PURLII then descended to 20-m depth and recommenced moving up and down between 10 and 20 m along a return heading, until a second specified period of time elapsed. At this time, PURLII resurfaced and shut down. If, while the vehicle descends, its acoustic altimeter detects the bottom within 5 m, PURLII rises to 10 m and then attempts once again to descend to 20 m.

5. Vehicle performance

Figure 2 shows the approximate vehicle ship track for the northbound leg of mission 3 conducted on 27 November 1996 beginning at 0908 PST. Figure 3 shows parameters logged by the vehicle sensors during the 50-min duration of the northbound leg of mission 3. The goal of the missions was to have PURLII follow a vertical sawtooth path that encompassed the thermocline. A recent vertical temperature profile obtained with a conventional CTD showed the thermocline to be between 10 and 20 m. The bottom line in Fig. 3a is the total lake depth in the vicinity of the AUV, as obtained by summing the acoustic altimeter and depth sensor readings. The vehicle depth as recorded by the depth sensor shows up as the sawtooth-shaped line between 10 and 20 m in Fig. 3a. The peaks of the sawtooth are 70–80 m apart. Even though the vehicle was slightly positively buoyant, it dove faster than it came up. This is because the vertical thruster is more efficient thrusting down than up. The modulation of the sawtooth pattern that occurs until minute 8 is due to PURLII’s bottom-avoidance behavior, as described in section 2. The spike near the end of the lake floor depth record (bottom line, Fig. 3a) is a single spurious reading from the acoustic altimeter. Spurious readings such as this did not affect PURLII’s behavior, and for navigation purposes the altimeter was found to perform reliably.

During this leg of the mission, PURLII was programmed to follow a heading of 340°. Figure 3b shows that PURLII maintained a heading between 340° and 348°. The eastward bias in heading is likely due to the slack line attaching the float to the canoe that was following on PURLII’s port side. At the end of 50 min the vehicle heading swings to the east and PURLII climbs to the surface (Figs. 3a,b), as specified by the mission program.

Before a mission PURLII is ballasted such that it is slightly buoyant and sits even in the water. It is clear from the pitch and roll records that while PURLII was ballasted even in pitch, there was a roll of 3.5° (Figs. 3c,d). Due to this bias in roll throughout the mission, PURLII’s compass heading shifted from 340°–344° while diving to 345°–348° while moving up. When PURLII was diving it rolled to 6° down on the starboard side and pitched 3°–4° down. Due to the rolled position of PURLII, the vertical thruster provided some horizontal thrust, which turned the vehicle westward. When PURLII was climbing it had a roll angle of 2° down on the starboard side and a pitch up of 3.5°–4.5°. Since PURLII was still rolled down to starboard, the vertical thruster had the effect of pulling the vehicle’s nose eastward. At the top of an ascent and the bottom of a descent there is an overshoot of 2° in pitch, and 0.5° in roll, before PURLII stabilizes.

Since the objective is to collect high-quality water property data, the vehicle configuration must optimize sensor effectiveness. In the case of PURLII, improvements in data quality can be made by reducing vehicle influences on sample water and with more accurate positioning. CTD data quality could be improved by moving the CTD forward, thus reducing the length of tubing between the intake and the sensors, which would reduce the smoothing effects of mixing within the tube. Also, improvements in CTD data quality can be effected by moving the intake forward, thereby reducing the potential for the vehicle hull to modify sample water. Positioning accuracy could be improved by correcting heading bias and fluctuations. Adding a vertical thruster on the starboard side would balance the vertical thrust and thereby remove the large shifts in roll that cause heading fluctuations. The heading bias could be removed by replacing the safety line and float with an acoustic pinger–locator system. Since the time of the missions described in this article, PURLII has been fitted with an acoustic pinger–locator system and a second vertical thruster.

6. Temperature data

A comparison of mean temperature profiles from the CTD on board PURLII and that measured by the thermistor chain at T4 is displayed in Fig. 4. Thermistor chain temperature data averaged over the duration of the northward leg of PURLII’s mission 3 are shown as symbols with error bars. The vertical error bars correspond with a estimated depth accuracies of the thermistors, and the horizontal error bars represent the thermistor accuracy of 0.05°C. The temperatures measured by the CTD on board PURLII during the northward leg of mission 3 were averaged in 0.5-m depth bins between 10 and 20 m. These values were assigned to the middepth of the bin and then plotted. The dashed–dotted lines represent one standard deviation to either side of the mean.

Temperature data collected while PURLII performed the northward leg of mission 3 are displayed in Fig. 5 as a function of depth. Each temperature profile is offset by 0.5°C for display purposes. The first four profiles are cut short at about 16 m because of PURLII’s bottom-avoidance maneuvers. During the 50-min duration of this segment of the mission, PURLII was able to record four temperature profiles between 10 and 16 m, and 23 profiles between 10 and 20 m. This vertical extent encompasses the thermocline of the lake at this time. The profiles show a deep, weak thermocline with the center of the gradient region at about 14 m, and a surface-layer temperature of 6.2°–6.3°C.

Calm weather on 26 and 27 November 1996 ceased when wind started in the evening of 30 November and persisted until the evening of 2 December 1996, with snow falling heavily on 2 December 1996. In order to see the effects of the storm on the thermocline PURLII was deployed on 2 December 1996 (mission 5). A ship track similar to that of 27 November (mission 3) was followed. The temperature data from the northward leg of mission 5 are displayed in Fig. 6 as temperature versus depth, with each up and down profile separated by 0.5°C. The thermocline depth has dropped to about 17 m and the surface layer temperature has cooled to 5.8°–5.9°C.

Apparent temperature inversions are evident in the temperature record of 2 December 1996 (Fig. 6). Since PURLII was moving through the water 10 cm vertically and 35 cm horizontally every second and there is no other information about the temperature field, all that can be said is that the local slope of an isotherm is greater than PURLII’s climb or descent angle of about 16°. Whether or not the apparent temperature inversions constitute overturning events is difficult to determine without more information (Thorpe 1977); however, they are indicative of the presence of high-frequency internal waves (Thorpe et al. 1996).

The general tilting, curving, and variable thickness of the thermocline on both 27 November and 2 December 1996 (Fig. 6) are indicative of a combination of first mode horizontal, and first and second mode vertical internal seiches (Münnich et al. 1992). The considerable variability between subsequent profiles on 2 December 1996 (Figs. 5 and 6) is suggestive of sporadic mixing events due to Kelvin–Helmholtz instability (Thorpe et al. 1977) and penetrative convection (Imberger 1985). The high mode internal seiching and sporadic mixing events are not likely to have been detected had only five or six conventional CTD casts been obtained in the same period of time.

7. Summary

PURLII, with its single vertical thruster and no active control surfaces, was able to follow a vertical sawtooth pattern while avoiding the lake floor. By doing so while recording water temperature, effective two-dimensional slices of the temperature field, 1 km long and 10 m high, were created. Each “snapshot” of the temperature field was obtained in under 50 min. Such a dataset provides information about the horizontal variability of the vertical structure of the temperature field with a profile spacing of about 75 m and a vertical resolution of about 10 cm. The data collected encompassed the thermocline region of a lake before and during a wind event. The middepth of the thermocline was found to have deepened by about 3 m, and the surface-layer temperature was found to have dropped by about 0.4°C between 27 November and 2 December 1996. Temperature features indicative of high-frequency internal waves were found on both 27 November and 2 December 1996. Temperature features indicative of sporadic mixing events and high-frequency internal waves were evident on 2 December 1996. These features would not necessarily be evident using conventional techniques.

The ability to collect many CTD profiles, between two preset depths and along a predetermined path, is an improvement over conventional CTD data collection. Conventional methods require setting up a number of stations along the long axis of the lake, then driving a boat to the first station, lowering instruments to a desired depth, retrieving the instruments, then driving to the next station, and so on. Each station could be completed, and the boat driven to the next station in no less than 7–10 min, depending on the distance between stations. In 50 min, five to seven stations could be completed, whereas PURLII is able to complete 27 profiles.