Simulations and Results
6.2 Simulation Setup
6.2.2 Target Detection
As discussed in Section 6.2.1, targets are detected using infrared cameras. Since the targets are humans in the sea, who may be suffering from hypothermia, a target is said to be located if the camera detects a temperature of 32°C or higher; this is the temperature at which mild hypothermia becomes moderate hypothermia [NHS, 2013]. The search algorithms make use of the evaluation of the objective function at a series of waypoints, but the infrared camera is still active during the travel times, so that targets are not missed in between waypoints. In order to make sure this is feasible, a calculation must be done, which estimates how much time would be saved from the normal operating time. According to Gavrilets (2003), the fuel in the helicopter lasts around 9 minutes, but information on the battery storage and the power output of the avionics system has not been provided. Nevertheless, a reasonable justification for keeping the camera switched on can still be obtained. Although the battery capacity was not given, for the sake of this calculation a capacity of 78Wh at 19V is assumed, as this is a typical battery capacity for this type of vehicle [Qi, Zhao, Jiang & Han, 2006]. Now, if the avionics system with the infrared camera is able to run for 9 minutes (the same time as the fuel lasts), then the average current in the circuit can be estimated by taking into account power losses. These losses are estimated to be 70% and therefore the maximum total current for the avionics system to last 9 minutes is
19.16A mins 60 1h mins 9 19V 78Wh 0.7 Time Discharge Capacity 0.7 I (6.2)According to FLIR (2013), the power dissipation at steady state is about 1.2W, and the input voltage is 4-6V, so assuming a minimum voltage of 4V, this means that the maximum current required from the battery is power/voltage = 1.2/4 = 0.3A. Subtracting this from the original current obtained in Equation (6.2), the maximum total current without the camera that guarantees a battery lifetime of 9 minutes with the camera is I = 18.86A. Substituting this back into Equation (6.2) and rearranging, the discharge time required without the camera is
Chapter 6 Simulations and Results 91 9s mins 9 mins 9.14 0.152h A 18.86 19V 78Wh 0.7 I Capacity 0.7 Time Discharge
In other words, if the avionics system without the camera lasts more than 9 minutes and 9 seconds then the inclusion of the camera does not reduce the discharge time to anything lower than 9 minutes, which is the time limit imposed by fuel consumption. Since it is implied that the battery lasts longer than the fuel, it is assumed that the battery lasts long enough to include the camera, since it only needs 9 seconds more than the fuel. Therefore, it is assumed that the major factor in determining a suitable mission time is the fuel consumption, so the simulations are run for 9 minutes. If a target is detected by a vehicle en route to a waypoint, the vehicle stores this location but still continues to the waypoint in case another target is found there. This is done so that the intended point generated by the search algorithm is still visited, but if no other target is found by the time the vehicle reaches the waypoint, then the target location is incorporated into the search algorithm instead of the waypoint.
Given that the fuel lasts 9 minutes, the actual search phase is only carried out for 7 minutes to give enough time for the capture phase. In this phase, the UAVs are instructed to move to their corresponding landing positions directly above the USV. As the UAVs are instructed to fly back towards the USV, they are instructed to fly at different altitudes to reduce the risk of collisions. As the search phase is over at this point, flying at different altitudes does not affect the results. Once the UAVs are directly above their landing spots, they are lowered slowly onto the platform for refuelling. A time of 2 minutes has been chosen to complete the capture phase to take into account the possible distance from the platform, the time taken to position the UAVs correctly and lower them slowly onto the platform. For example, a UAV could potentially be 282.8m away from the platform (distance from centre to corner of square), and while the UAVs travel at 10m/s, they still need to slow down so that they can be positioned and oriented correctly, which can take a total time of up to about a minute.. The UAVs are then lowered slowly onto the platform, which typically takes up to 30 seconds. An extra 30 seconds is then added as a safety margin to take into account other factors such as weather conditions and possible collision avoidance manoeuvres.
6.3 Standard Search Patterns
This section presents the implementation and results for the three standard search patterns discussed: Parallel Sweep, Sector Search, and Expanding Square. As these methods are commonly used in real search and rescue missions, they have been tested as a benchmark for the more advanced heuristic techniques. The individual results for these methods are given in Appendix C1.
Chapter 6 Simulations and Results
92
6.3.1 Parallel Sweep
The first common search method to be tested is the Parallel Sweep (PS). This search pattern is implemented by dividing the search area into four different regions, and assigning one UAV to each region, with each of the four UAVs being deployed from the USV in the middle of the search area. This illustrated in Figure 6.5.
Figure 6.5: Parallel Sweep Implementation
The search pattern, as shown in Figure 6.5, is implemented in each region, although the “long sweeps” alternate between horizontal and vertical in adjacent regions so that the collision avoidance algorithm does not take over. For example, if the long sweeps are horizontal in Regions 1 and 2, then it is possible that the 2 vehicles could fly towards each other, and even though it is unlikely that they could collide, the collision avoidance algorithm might instruct the vehicles to perform avoidance manoeuvres. Implementing the search in this way ensures that the area is still searched uniformly. The results of this algorithm are shown in Table (6.1), and a typical run is shown in Figure 6.6.
Table 6.1: Parallel Sweep – Results
Algorithm Targets Detected % Coverage Detection Time (s) Mean Median St. Dev. Mean Median St. Dev. Mean Median St. Dev.
PS 5.58 6 1.79 46.48 46.45 0.13 86.51 61.10 58.64
Region 3
Region 4
Region 2 Region 1
Chapter 6 Simulations and Results
93 Figure 6.6: Parallel Sweep
Because of timing issues related to fuel consumption, the tracks for the parallel sweep are separated by 20m, which is greater than the detection diameter of the infrared camera. This has been done so that the UAVs carry out a more distributed search. The whole area cannot be covered within the given time limit, as shown by the coverage in Table (6.1), and therefore, not all of the targets are detected. Here, the median time to detect the first target is lower than the mean time, which indicates that there are some runs that take an unusually long time to detect a target, and as a result, the mean time increases. This can easily occur if the targets are located far from the origin. From this particular run, it can be seen that the Parallel Sweep algorithm is likely to find targets, but it is also likely to miss some because of the spacing between the tracks. Specifically, because the UAVs cover a large distribution of points, they are likely to find some targets but because the space between the tracks is larger than the detection diameter of the infrared camera, the UAVs are unlikely to detect all of the targets.
6.3.2 Sector Search
The next common search pattern to be tested is the Sector Search (SS). The Sector Search, like the Parallel Sweep search, is implemented by dividing the search area into four regions, as illustrated in Figure 6.7. -250 -200 -150 -100 -50 0 50 100 150 200 250 -250 -200 -150 -100 -50 0 50 100 150 200 250 East Position [m] N o rt h P o s it io n [ m ]
Chapter 6 Simulations and Results
94
Figure 6.7: Sector Search Implementation
This approach is used as it is not advised to carry out the sector search when multiple vehicles are searching the same area at similar altitudes [IAMSAR, 2008]. Note that the second search, as illustrated in Figure 5.4, is applied here, with the initial Sector Search being represented by the solid black lines, and the second part of the search being represented by the dotted black lines. The results of this algorithm are summarised in Table (6.2).
Table 6.2: Sector Search – Results
Algorithm Targets Detected % Coverage Detection Time (s) Mean Median St. Dev. Mean Median St. Dev. Mean Median St. Dev.
SS 5.27 5 1.74 43.32 43.32 0.06 82.88 60.90 71.60
A typical run of this algorithm is shown in Figure 6.8. Again, due to fuel consumption, the entire area cannot be covered, as shown by the coverage in Table (6.2). Compared to the Parallel Sweep, it can be seen that the Sector Search gives lower coverage, does not detect as many targets, and takes longer to detect targets. The reason for the lower coverage is that the UAVs visit their respective centre points several times and therefore other parts of the search space are missed, as opposed to Parallel Sweep, which does not visit the same points more than once. As a result, the UAVs are likely to miss some of the targets unless the targets are located near any of the four centre points. The results for the detection time are very similar to those of Parallel Sweep but overall, Sector Search does not perform as well.
Region 1
Region 4 Region 2
Chapter 6 Simulations and Results
95 Figure 6.8: Sector Search