Since their inception, unmanned aerial vehicles (UAVs) or drones have demonstrate great potential for various applications due to their flexibility and versatility. Nonetheless, there are safety concerns about using UAVs in public areas, where a malfunction during flight can harm individuals, property, or the UAVs. Hence, it is important to develop safety standards and procedures to decrease the impact of such failures. The proposed study presents a new system for detecting failures in UAVs as the first stage of an autonomous emergency landing safety framework. The proposal explores the capabilities of the mean-shift clustering algorithm alongside the sensing of vibration data. The measurement of vibration signals was accomplished through multiple flight tests conducted under different faulty propeller scenarios using a customized hardware system. The collected data was analyzed to identify the optimal configuration of acceleration and gyroscope parameters for accurate failure detection in quadcopter propellers. The study revealed that the most accurate method to detect and ensure a safe emergency landing in case of a propeller fault or failure is to consider the gyroscope parameter in the vertical direction (gZ) and the accelerometer parameter in the same direction (aZ). This approach enables quick detection and timely engagement of the safety framework. Scatter plots and confusion matrices were generated based on the parameter set (gZ-aZ), and the mean-shift clustering algorithm was applied to categorize the vibration data into three clusters representing different fitness states: normal, faulty, and failure. The proposed system was validated through real-time flight experiments, successfully identifying faults and failures using the established clusters.
Mean-Shift Clustering for Failure Detection in Quadcopter Unmanned Aerial Vehicles
Studies Comp.Intelligence
Advances in Optimization Algorithms for Multidisciplinary Engineering Applications: From Classical Methods to AI-Enhanced Solutions ; Chapter : 35 ; 785-806
2025-04-25
22 pages
Article/Chapter (Book)
Electronic Resource
English