Model-Free Health Monitoring of Multirotor UAVs Using Frequency-Response Analysis

Monitoring reliability in the propulsion systems of Multi-Rotor Unmanned Aerial Vehicles (UAV's) is crucial for safe operational practices in applications that are sensitive to safety, including inspections of infrastructure, delivery operations, and emergency response. Propulsion system degradation in these applications can lead to an unstable UAV flight condition, potentially causing loss of control during the mission with even slight degradations of components like propellers, Electric Motors (EM), Electronic Speed Controllers (ESC), etc. In contrast to current diagnostic methods that require additional sensors, complex high-fidelity models, or platform specific machine learning frameworks, which all contribute to increased UAV system complexity and reduced applicability across different platforms. This paper introduces a model free propulsion fault detection method based upon the application of frequency response analysis to measure the system's response to a chirp signal that is introduced into the attitude controller and then compared between the healthy and faulty responses of the propulsion system. The use of a chirp signal allows for the collection of both healthy and degraded system responses without any modification of the UAV's hardware or software. A Differential Bode plot is created through the subtraction of Healthy vs Faulty Frequency Responses. The subtraction of the two frequency responses increases the magnitude of the fault signature relative to the normal dynamics. It is shown experimentally that this process provides a 10 dB increase in sensitivity over a traditional Bode plot diagram, thus providing an opportunity for earlier identification and confirmation of the performance decline. Additionally, the phase-clustering method utilizes a simple metric to classify the identified faults as propeller, motor, or ESC-related faults. Since no additional sensors are required, no additional computation is required; and no changes are made to the physical structure of the UAV, the method is lightweight, platform independent, and very well suited for real-time UAV health monitoring and fault-tolerant flight control systems.