Quantitative evaluation of multi-rotor UAV propulsion system reliability

Abstract

The study investigates the reliability assessment of unmanned aerial vehicle (UAV) propulsion systems using quantitative approaches to predict the failure rates of these crucial components. The research is driven by the need for reliable UAV operations, given the significant consequences of propulsion system failures on safety and performance. This project aims to fill the gap in current research, which mainly uses linear models, by introducing the use of the Weibull distribution, a mixed non-linear Weibull model (NWMM) with Bayesian estimation, and artificial neural networks (ANNs) to accurately represent the complex failure patterns observed in UAV systems. By studying UAV-FD dataset, with a specific emphasis on the electric speed controller (ESC) data, it became evident that the deterioration patterns of UAV components are not linear. This facilitates understanding of the effects of defects on motor dependability, while also laying the groundwork for the creation of predictive models that can provide insights for maintenance schedules and influence enhancements in UAV system design. The study employed multiple models to assess UAV-FD and test bench datasets, showing that the Weibull distribution is capable of describing motor degradation under various fault conditions. This promotes using non-linear analysis to assess UAV reliability. Future initiatives include run-to-failure experiments and flight testing will collect failure data to improve the NWMM and investigate NN models' capacity to represent complex relationships.