Fault-tolerant strategies for multi-rotor parcel delivery

• Main Author: Tan, Jun Kiat
• Other Authors: Mir Feroskhan
• Format: Thesis-Doctor of Philosophy
• Language: English
• Published: Nanyang Technological University 2024
• Subjects: Engineering; Fault tolerance; Drone
• Online Access: https://hdl.handle.net/10356/180356
• Institution: Nanyang Technological University

Drones have revolutionized various industries, offering versatile applications from surveillance to agriculture and disaster response. Specifically in urban aerial delivery, their agility streamlines parcel transportation, reducing environmental impact. However, deploying drones in cities poses challenges regarding fault tolerance. Mechanical faults like actuator or sensor failures risk stability, while software issues can disrupt control and functionality. Dynamics faults alter flight capabilities and environmental factors like weather impact performance. To counter these, strategies involve mechanical solutions redistributing control inputs or facilitating emergency landings, software solutions ensuring stability despite faults, and introducing redundancy through backup components like actuators, sensors, and power systems, collectively fortifying drone reliability and safety in complex urban settings. With multi-rotors being vulnerable to a spectrum of faults, it presents a challenge to further strengthen them as pivotal assets in aerial delivery by enabling fault tolerance capabilities on top of their existing agility and adaptability. Therefore, the motivation of this thesis is to enhance drones' fault tolerance capabilities on 3 fronts. The first is to address environmental faults in the form of imagery noises that occur during aerial parcel delivery, specifically during landing operations. The second is to address hardware faults in the form of rotor failure during aerial parcel delivery, specifically when the drone is in flight. The last is to address dynamics faults in the form of change in dynamics such as the change in mass, size, and center of mass due to different parcel requirements. Following the motivation, the objective of this thesis is to devise and apply a range of methodologies to systematically tackle faults and failures within an aerial system. This overarching goal can be segmented into three specific sub-objectives. The first is to devise a software solution to address positional inaccuracy due to imagery noise produced by the external interaction during drone landing. The second is to devise a hardware solution to address in-flight rotor failure, a common mechanical fault in drones. And the last is to devise a hardware solution to address the changes in the center of mass, size, and, mass of a parcel, which is classified under dynamics faults. The first objective involves implementing a pixel-based path-planning algorithm to enhance an aerial system's resilience against image noise, allowing precision tasks like landing on moving platforms. This method integrates an extended Kalman filter and Nonlinear Model Predictive Control into the control architecture, proving more effective than existing approaches through experimental validations. The second objective addresses rotor failures on a hexacopter by introducing a geometric morphing mechanism that adjusts rotor arms to redefine thrust location, ensuring near-zero static hover post-failure. The system employs event-triggered sliding mode control and Multivariable Nonlinear Terminal Sliding Mode Control to maintain stability during single or adjacent rotor failures, validated experimentally. Finally, the third objective establishes a decentralized, modular aerial delivery system adaptable to various parcel sizes and masses. It utilizes a decentralized controller with a Graph Neural Network-based policy and a module placement planning workflow driven by neural network-trained measuring scales, showcasing effectiveness in flights with different parcel configurations in experimental tests. The proposed pixel-based path-planning algorithm enables the UAV to land on both static and moving platforms. For moving platforms, the UAV is capable of detecting and tracking the platform as it moves at a speed of 1.5 m/s, maintaining vertical separation, and successfully landing once the landing sequence is triggered. This method demonstrates higher tracking accuracy, achieving about a 75% reduction in RMS error compared to existing literature. Additionally, the morphing hexacopter, equipped with an MNTSMC strategy, can sustain flight even with two adjacent rotor failures - an event that would typically result in the crash of a conventional hexacopter. The event-triggered morphing sequence, powered by a trained feed-forward neural network, can adjust the arms via servos to reach 50% of the desired angle within 0.8 seconds. Further analysis reveals that a higher event-triggering gain leads to a faster morphing response. Moreover, in the decentralized modular aerial delivery system, a dynamic scale with neural network logic can accurately measure parcel parameters, including mass, size, and center of mass, and determine the optimal placement of the modules. Flight experiments show that a shift in the center of mass in a conventional multi-rotor layout significantly impacts the system's flight performance, particularly in terms of power efficiency when analyzing thrust variation. These experiments also demonstrate how modular systems can be reconfigured to recover power efficiency and maneuverability by adjusting rotor placements. Achieving fault tolerance in diverse aerial platforms poses a formidable challenge due to the absence of a universal solution for various scenarios. Nonetheless, prioritizing fault tolerance is crucial for enhancing drone safety and reliability, particularly in densely populated urban environments, and exploring diverse strategies in this regard expands the potential applications of drones across different domains.