Efficient Ground Area Coverage using UAV Network

Abstract

Unmanned aerial vehicles (UAVs) have gained popularity recently. The flight control unit (FCU), onboard CPU, camera, and other sensors automate UAVs. They collectively form an ad hoc network called flying ad hoc networks (FANETs). Compared to other terrestrial ad hoc networks, FANET has some unique features that attract users more these days. Regarding applications, other forms of ad hoc networks have specific restrictions. They can only be used on the surface of the ground or water. The best part is that FANET can be used in a regular environment in hazardous locations. FANET is now integrated with terrestrial networks for reliable wireless communication. UAV’s agility, processing power, data collecting, and transmission improve network infrastructure. It’s useful for environmental and traffic monitoring, rescue operations, smart farming, and more. All the uses mentioned above go under the coverage applications, in which UAVs need to cover or scan a target ground region to complete a mission successfully. Finding a suitable collection of waypoints for the movement, a collision-free path to cover all the waypoints, establishing a stable route for data transmission, and an automated fault diagnosis approach to isolate faulty UAVs and erroneous data for a successful coverage operation are the objectives of the thesis. Simulation is performed to evaluate the proposed work using Network Simulator-3 (NS-3) and Python environment. An optimized aerial decomposition approach is proposed for determining a suitable set of waypoints for the movement of UAVs. The decomposition process is carried out by integrating the footprints created by a UAV. This work considers two distinct forms of UAV footprints: the camera footprint, which pertains to visual coverage, and the antenna footprint, which pertains to sensing coverage. The midpoint of each footprint is regarded as a waypoint for the trajectory of a UAV. An objective function is introduced to optimize the decomposition process by maximizing and minimizing the inside and outside ground area coverage, respectively. The proposed work is evaluated regarding the percentage of inside and outside area coverage using the generated set of waypoints. It improves the inside area coverage to 6.8% and reduces the outside area coverage to 67.59% approximately. A path planning approach is proposed to find a collision-free path. It aims to cover all the waypoints generated before. The proposed approach involves representing the collection of waypoints as a graph, whereby a UAV determines its subsequent intermediate waypoint by considering the environmental conditions at its present waypoint. Geometrical methods, such as the collision cone approach and triangle laws are used to determine the occurrence of collision with static and dynamic obstacles. The path planning algorithm’s performance in a collision-free environment is evaluated using the traveling salesperson algorithm (TSP) Next, the process of path planning involving collision detection and avoidance is conducted and compared to the FGM-I approach. Path length, execution time, and the total number of collisions required to be avoided are the parameters of the evaluation process. The proposed algorithm reduces the path length to 8.17% and the total number of collision avoidance to 40.3% approximately. Routing mechanisms of non-cluster-based and cluster-based approaches are proposed to find a stable and energy-enabled route to the ground control station (GCS). The non-cluster-based routing incorporates three key parameters in the route discovery process: the velocity threshold, energy threshold, and expected link stability time between two UAVs. However, in the case of the proposed cluster-based routing, the cluster head election process considers five parameters, including connectivity degree, link stability time, surplus energy, connectivity with backbone UAV, and speed differential. The clustering process is managed using an approach of reinforcement learning (RL) mechanism called State-Action-Reward-State-Action (SARSA). The routing mechanism is implemented through the inter and intra-cluster forwarding method. The performance evaluation of both methods encompassed the assessment of various parameters such as packet delivery ratio (PDR), delay, routing overhead, network lifetime, number of redundant clusters, and topology construction time. The non-cluster-based and clustering-based routing schemes reduce their routing overhead to approximately 11.6% and 13.7%. An automated fault diagnosis method is proposed to identify the different kinds of faults. In this work, UAVs cover a ground region filled with sensor devices and collect sensor data during the coverage mission. This work identifies two types of data faults: sensor faults and UAV data faults, and the classification of the faulty data is performed at the GCS. The fault detection process is performed in three stages: the sensor fault detection and elimination using a modified Z-score statistical test, the UAV data fault identification using a multivariate analysis of variance (MANOVA) test, and the classification of erroneous data using a centroid-based probabilistic neural network (PNN) at the GCS. The evaluation is performed using fault detection accuracy (FDA), false alarm rate (FAR), false positive rate (FPR), and false classification rate (FCR). The proposed PNN detects an accuracy of 95.3%, 92%, and 91% for permanent, intermittent, and transient UAV data faults, respectively.