Real-Time UAV Fault Detection and Classification Using Measurement Data from the PADRE Database

The growing number of applications for unmanned aerial vehicles operating in close proximity to humans calls for strict safety requirements. To ensure reliability and safety, fast and effective diagnosis of possible drone faults is needed. In this paper, a study on the detection of damage to the propellers of a multirotor using inertial sensors was conducted. Measurement data from the publicly available PADRE repository were used. Frequency features and raw measurement data applicability were tested. Two different approaches to data processing during fault detection and classification were tested. The first is to analyze the signals after collecting the entire set of data needed for treatment by the neural network. The second approach processes the data after every single acquisition of sensor measurements. The performance results and processing time of each solution are recorded for analysis. By effectively selecting parameters of the proposed approaches, the processing time and accuracy of UAV fault classification can be significantly improved, as verified over different classes of propeller defects. Early fault detection is essential for safe operations of multirotor drones.