Non-destructive Evaluation of the Condition of a UAV´s Propellers by means of Acoustics

The ever-increasing usage of multi-copters goes along with the necessity of guaranteeing a safe operation before, during and after the flight. As of today, Maintenance, Repair and Operations (MRO) aspects are not sufficiently considered, however, they will quickly gain importance as UAV operation becomes more professional, more frequent and more competitive. The key element of the UAV's propulsion system are the propellers, which tend to be easily damaged by strikes or during the handling of the vehicles. The consequence is a reduced thrust and higher vibration, stressing the system as well as reducing its performance. Since the main source of sound is the propulsion system, we propose the use of acoustics as a means to detect damaged or imbalanced propellers at an early stage and without impairing UAV operation. In this paper, we present the concept for such non-destructive testing of a multi-copter. The fault diagnosis aims at identifying different system conditions such as an undamaged reference and an impaired propeller. The method is based on a machine learning algorithm with a neural network architecture. A prototype based on the structure of a single propeller is designed for a first experimental approach and the generation of data. In order to identify relevant features, this set-up is used to systematically explore the impact of recording procedures and setups on a successful diagnosis. Pertinent time and frequency domain features are then analyzed and the most promising ones are identified. From the results obtained, rules are derived for the implementation of an acoustic monitoring system for UAVs