Preliminary study of actuator fault detection for RUAVs using neuro-fuzzy system

Abstract

With the ever-increasing use of Rotary Unmanned Aerial Vehicles (RUAVs) for various purposes, a fast and accurate fault detection system is key to detect unknown faults, to ensure fault tolerance and for safe operations. In this paper, a data-driven fault detection algorithm based on an Adaptive Neuro-Fuzzy Inference System (ANFIS) was used to detect single actuator faults of quadrotor flights through sensor readings from inertial measurement units. Sensor readings were collated and pre-processed to generate a training dataset to train the neuro-fuzzy model, and then the trained model was evaluated using four-fold cross-validation and a set of performance metrics. The results presented illustrate the generalisation performance ANFIS model and its effectiveness in detecting actuator faults.