Underwater robotic system for reservoir maintenance

3.1. Design

The tracked robot is equipped with two track drive units by Inuktun Minitrac, intended for underwater operation. Usage of tracks has the main advantage over wheel drive that it may provide much better performance in terms of traction on a variety of surfaces. The track units feature elastomeric belts with treads, driven by DC motors. The robot chassis is based on an adjustable stainless steel frame that has mounting interfaces for inspection or cleaning equipment and watertight enclosure for control and power electronics. In Fig. 2, it is presented, equipped with a 3D sonar.

3.2. Kinematics

Mathematical model of the robot was created in order to appropriately implement control algorithms and path planning on the bottom of a tank. An initial step was to create a simplified model of a track drive. The elastomeric track was modelled as an inextensible tape wound on a driving sprocket and idler, similarly to the approach described in [7]. Next, the model of the track drive was integrated with a kinematic model of the entire robot.

Fig. 2Tracked mobile robot – prototype

Fig. 3Velocities of the characteristic points during turning of the tracked robot

It was assumed that the robot will move on a flat and inclined surfaces. Kinematic equations were developed for a characteristic point of the robot (described as point C in Fig. 3), using similar approach to [7]. It was assumed that the robot moves on a planar surface, that may be inclined by angle of inclination $\gamma$ with respect to the $x$ axis. The modeling of motion on inclined surface is limited to the case when the robot moves straight uphill or downhill as depicted in Fig. 5:

where $r$ is radius of the track drive sprockets, $H$ is the distance between the tracks, $s_1$, $s_2$ are the slip ratios of track 1 and 2 respectively, ${\dot{\alpha }}_1$, ${\dot{\alpha }}_2$ are angular velocities of the sprockets, $\beta$-the heading angle of robot frame with respect to the $z$ axis and $\gamma$ is the slope inclination with respect to horizontal surface

3.3. Dynamics

The model of the robot dynamics was created on the basis of Lagrange energetic methods. Initially, kinetic and potential energies were calculated in a way that each moving part of the system was analysed individually. Afterwards, total energies of track drive modules and frame assembly were combined to obtain total energy, taking into consideration all forces acting on a robot moving underwater, shown in Fig. 4.

Further calculations involved usage of Lagrange equations of the second kind. The final form of the dynamics model equations that include nonholonomic kinematic constraints is denoted by Eq. (2):

where $m_R$ is mass of the frame, $m$ is mass of the track, $W_t$ represents rolling friction force, $P_u$ is the cable pull force assumed as constant, $F_w$ is the buoyant force, $F_D$ is hydrostatic resistance force, $M_{s1},M_{s2}$ are torques on the drive sprockets of tracks 1 and 2, $I_R$ is the moment of inertia of the robot frame, $I_x$, $I_z$ are the reduced moments of inertia of the track drive module, $M_p$ is the moment of transverse resistance, $G$ represents the gravity force, $\eta$ is the drive efficiency and ${\lambda }_1$, ${\lambda }_2$, ${\lambda }_3$ are the Lagrange multipliers (friction forces).

Fig. 4Forces acting on the robot during underwater locomotion

In the model it was assumed that the external forces such as rolling friction force, pull force, gravity force, buoyant force and hydrostatic resistance force are evenly distributed to calculate torques of both track drives. Validation of the mathematical model was conducted using a prototype of the robot. First of all, a mathematical model of the robot was implemented in Matlab/Simulink environment. A 3D path that involved avoiding an obstacle and climbing an inclined surface was set as an input for the simulation with maximum robot velocity. Further, the same path was followed by the prototype in the laboratory, on a test rig. The planned path is presented in Fig. 5. Resultant driving moments on each of the track drives were compared to the simulation results as presented in Fig. 6.

Fig. 5Trajectory of the robot for the simulation and experiment

Fig. 6Driving moments on track drive motors – simulation and experiment

We observe that the values of moments obtained during validation agree with the values obtained during simulation, however multiple peaks are present even during motion with constant speed. This is caused by alternating changes of load carried by consecutive treads of the elastomeric tracks.

Further analysis of the signal confirmed that the frequency of track treads coincides with the peaks. This may be filtered to obtain much smoother plot.

4.1. Design

The ROV designed for the water tank inspection has two different tasks to perform: scan the surface of the tank walls under the water line and inspect the construction of ceiling and walls above the water line. It is equipped with two cameras to satisfy these tasks – a monochrome, high resolution camera, for photographing the surface of walls underwater, and a colour, low resolution, pan-tilt-zoom camera, used to navigate and inspect the ceiling and wall structure above water. Because of the way these cameras are mounted, which is the result of using the most of their functionality, an additional camera is needed to allow safe and easy docking on the tracked unit and avoid obstacles when submerging – this camera is located in the bottom part of the vehicle, looking downward. The high definition inspection camera is placed at the front of the robot in a watertight housing, equipped with a dome port to minimise image distortion due to air-water interface. The pan-tilt-zoom camera is placed perpendicularly to the inspection camera, in a watertight housing with a spherical dome, to fully utilise the camera's field of view when it rotates. It is at the top of the whole robot and can emerge from water when the robot surfaces. At the front of the mini-sub there are two high power LED light-heads with a beam angle of 120 deg. Some additional low power LED lights are located at both sides of the vehicle and the docking camera has its own. The vision system is complemented with four laser modules installed in a rectangular pattern, parallel to the high resolution camera's optical axis. The laser modules are encased in watertight housings with adjustable mounts. The red laser spots generated by these modules are used to measure distance and orientation of the inspected surface, calculate the image scale and remove perspective distortion from the images. Docking of the mini submarine is possible thanks to the use of the docking camera which allows to orient the robot correctly, looking at the illuminated marks on the tracked unit. Other important role of the docking camera is monitoring space under the robot to avoid collisions and entangling in the tether. To firmly connect the two units together there is an electromagnetic lock mechanism. The electric brushless thrusters move the robot horizontally and vertically. These two directions are separated. Two thrusters move the robot in surge and yaw. The central thruster moves the robot in heave. It is also used to keep the robot at a desired depth generating a small amount of thrust as the vehicle is designed to have slightly positive buoyancy to surface automatically in case of failure. Mini submarine has ballast tanks and electronics under the glass-fiber body. The ROV design is presented in Fig. 7 accompanied by a prototype photo in Fig. 8.

The control system of the ROV is based on two computing units - a real time micro-controller board and an embedded industrial PC, inertial and pressure sensors, brushless electronic speed controllers and the vision system which can play a double role of inspection and guidance. The real time unit is used for generating PWM waveforms, reading sensors and running time critical control loops. The industrial PC has a Linux based software architecture incorporating advanced control algorithms based on the vision system measurements. The control system is able to augment manual control of the ROV, done by the operator, with algorithms which support steady scanning of the water tank walls by keeping constant distance and attitude, important for building mosaics.

Fig. 7Design of the underwater remotely operated vehicle

Fig. 8Photo of the prototype

4.2. Distance and attitude measurement

The primary task of the underwater inspection robot is scanning the surface of the tank walls in order to recognise and register defects (mainly cracks). To be able to measure objects of interests in the image, the distance to the surface and the robot attitude are needed. These data are not only useful for the photo measurements but also for the robot control system which has to automatically keep the robot moving at a desired distance, perpendicularly to the wall. Determining the position of the camera in reference to the scanned surface is possible thanks to the use of four parallel laser beams, casting spots on the photographed object.

Fig. 9Formation of the laser dot image in the pinhole camera model. The image of the laser dot s is projected onto the camera sensor, through a lens with focal length f. The four parallel laser beams are located in a rectangular pattern of size a×b, symmetrically around the camera axis

The beams are parallel to the optical axis and arranged in a symmetrical rectangular pattern. Only three spots are necessary to calculate distance and attitude. The redundant forth laser beam was introduced to increase robustness of the system, by allowing one spot to remain undetected or reject a detected outlier. There is always a risk that the laser beam will hit a crack or dent and the spot will be invisible or distorted. An important feature of the vision system is an underwater housing of the inspection camera equipped with a dome port, which minimises image distortion and gives maximum possible field of view. It allows for close object distance which decreases in-water light path and thereby improves image brightness and contrast [9].

Fig. 10Geometry of the distance and attitude measurement

When a new image frame is captured it is then binarised and the centroids of the laser dots are calculated. Using Tales formula one can determine the distance $z_s$ to each dot using the following equations:

where $s$ denotes the distance from the optical axis to the laser dot determined by pattern size $a$, $b$ and $s\mathrm{\text{'}}$ the same distance on image plane while $f$ is the focal length of the optical system. For clarity all those quantities are presented in Fig. 9. When the dot distances are known one can determine the distance along the optical axis by simple average and the attitude angles $\alpha$ and $\beta$ (see Fig. 10) according to equations:

4.3. Experimental validation

To validate the proposed measurement method, the robot was placed in a test tank, a calibration of the underwater vision system was carried out and a set of images was acquired for various distances and orientation of the calibration table representing the analysed surface. Results obtained by means of the calibration table were considered as reference results and were compared with measurements obtained by the developed vision algorithm. An application was developed using the OpenCV library, which can detect corners in the checkerboard pattern and calculate its distance and orientation in reference to the camera frame. The same software includes simple image processing and analysis algorithms which allow for the detection of the laser dot pattern projected onto the surface. The calculations based on the developed algorithm, presented above, are compared with the ground truth in real-time [10]. Some of the acquired images and results obtained during experimental tests are depicted in Fig. 11.

In an exemplary case for the object at a distance of 632.8 mm calculated from the reference calibration table, vision-laser based measured distance amounted to 632.9 mm (see Fig. 11(a)). At 2132.4 mm object’s distance, vision-based measured distance was 2152.2 mm (see Fig. 11(b)). When regarding exemplary calculation of two angles of rotation of the object: a) the angle of rotation about $X$-axis amounted to 29.6 deg in the case of the reference measurement and 29.3 deg as a result of the developed vision-based measurement (see Fig. 11(c)); b) the angle of rotation about $Y$-axis 27.4 deg calculated using calibration table and 28.3 deg from vision-based measurement (see Fig. 11(d)).

The image from the inspection camera is not only used on-line but also archived on a hard drive inside the robot. It is then used to assess the condition of the scanned walls by detecting mechanical defects like cracks. After downloading the images to the control panel they are processed by a multistage computer vision algorithm, designed to automatically detect cracks. Firstly, the images are calibrated and the perspective distortions are removed. Then the image is pre-processed to remove noise and analysed with region segmentation and morphological operations were employed to detect the cracks. An exemplary result of the crack detection algorithm on an image of a wall surface is shown in Fig. 12.

The developed crack detection algorithm will be extended by underwater by image processing and image enhancement methods [11-17] and the images will also be used for mosaicing [18-20] to create a panoramic view of the tank surface.

Other problems, like negative influence of ROV self-noise on detection aspect in water environment [21] or ROV control [22-25] will be considered as well.

Fig. 11Exemplary measurements in a test tank with our custom vision software: a) close, b) far, c) rotation about X-axis, d) rotation about Y-axis

a)

b)

c)

d)

Fig. 12Crack detection on the wall surface: a) input image, b) detected crack

a)

b)