Authors:
Adam Bondyra
;
Przemysław Ga̧sior
;
Stanisław Gardecki
and
Andrzej Kasiński
Affiliation:
Institute of Control, Robotics and Information Engineering, Poznan University of Technology, Piotrowo 3A, Poznan and Poland
Keyword(s):
Multirotor UAV, Fault Detection, Propeller Damage, Vibration Analysis, Random Decision Trees.
Related
Ontology
Subjects/Areas/Topics:
Informatics in Control, Automation and Robotics
;
Intelligent Control Systems and Optimization
;
Machine Learning in Control Applications
;
Robotics and Automation
;
Surveillance
Abstract:
In this paper, a fault detection and isolation (FDI) system for propeller impairments of the multirotor UAV is presented. The algorithm is based on the processing of signal vectors from the set of vibration sensors located close to the propulsion units. Axial and radial vibrations are measured using MEMS accelerometers that provide data for the feature extraction based on the Fast Fourier Transform (FFT). Characteristic fault signatures extracted from vibration signals are used to detect and localize damaged blades using the set of random decision trees. A method was evaluated with data gathered during numerous test flights and validated in relation to signal acquisition time and number of classifiers in the ensemble. Results show over 95% sensitivity in detecting and isolating faulty rotor states. The presented approach is an effective and low-cost solution, very versatile to implement in the arbitrary UAV.