Fault Diagnosis of Drone Propellers using Inner Loop Dynamics

The reliability of drones is essential for various civil and military applications. While a single failure can be tolerated through fault-tolerant control that allows for reduced functionality, some applications prioritize system availability, requiring the drone to perform fully for a specified duration. In this paper, we explore a method for detecting emerging drone faults, such as small cuts to the propellers, that could eventually lead to functional failures. These emerging faults have a minimal impact on drone dynamics, often going unnoticed by control systems. Our fault detection approach involves monitoring the inner control loops of the drone controller, where these emerging faults are significantly correlated. The method relies solely on the standard data streams already collected during normal flight, enabling real-time diagnosis without the need for additional sensors or hardware modifications, unlike other methods that may not operate in real time and often require additional sensors. Experimental results demonstrated promising performance across over 12 different scenarios, varying in fault severity and operating conditions, for a wide range of drones. This work aims to develop reliable predictive maintenance capabilities for drones that are critical to maintaining operational reliability.