Predictive Maintenance with Neural Network Approach for UAV Propulsion Systems Monitoring

The number of registered Unmanned Aerial Vehicle (UAV) for both commercial or personal use and UAV market value has been increasing rapidly in the past 5 years, yet so is the number of accidents involving UAVs. Machine failure makes up almost 50% as the cause of accident, with almost 40% of the failures caused in the propulsion systems. Thus, it is crucial to have a monitoring system with predictive maintenance strategy to identify abnormalities in the UAV propulsion system and predict its Remaining Useful Life (RUL). Neural network and its ability to model complex relation between time series data is used to solve the RUL prediction problem, by estimating the current and predicting the upcoming Health Indicator (HI) of the propulsion system. This paper proposes a method with neural network approach for the implementation of predictive maintenance on propulsion systems in UAV. The method consists of three main procedures that are abnormality detection, HI calculation, and RUL prediction of the propulsion system. As an illustrative example, we model the degradation process of a quadrotor UAV propulsion system by gradually adding load imbalances to its propulsion system. The implementation of our method in the experiment data has shown good and effective results.