A GRU network framework towards fault-tolerant control for flight vehicles based on a gain-scheduled approach

Abstract

This paper introduces a novel approach to enhance the fault-tolerant control of flight vehicles by incorporating a gated recurrent unit (GRU) neural network within a gain-scheduled framework. Gain scheduling, a well-established technique for achieving active fault-tolerant control, enables the selection of control gains from a predefined set based on specific faults. An improved terminal sliding mode controller is derived, and appropriate parameters are chosen to formulate a gain-scheduled fault-tolerant controller. Based on it, training data is generated by comprehensive simulations on a Winged-Cone configuration flight vehicle. The GRU neural network architecture is designed and trained to function as the flight controller. The effectiveness of the proposed GRU network controller is demonstrated through a series of simulations.