Failure Detection and Diagnosis of a Small Unmanned Aerial Vehicle using Amplitude of Peaks

This work presents an innovative technique called SAC-AM based on Chaos theory to classify failures in a small Unmanned Aerial Vehicle (UAV). The average of the amplitude of the peaks in accelerometer signals are used as parameters to determine the chaotic behavior and consequently to detect and diagnose the failures. A comparison with Machine Learning techniques is presented as reference. A dataset of accelerometer data collected from a small UAV is used to evaluate all algorithms under analysis. Three different failures are presented in the dataset, and a total of 13 features were taken under consideration. As a result, SAC-DM achieved an accuracy of 100%, while Support Vector Machine achieved a maximum of 98.6%, and Random Forest achieved 98 %.