Sensor failure detection, identification and accommodation using neural network and fuzzy voter

Abstract

Sensor failure detection, identification and accommodation (SFDIA) is a challenging and important problem on unmanned aerial vehicles (UAVs) or other safety critical applications. This paper proposes new SFDIA scheme. The new scheme combines advantages of hardware redundancy and analytical redundancy. Fuzzy voter and neural network (NN)-based sensor estimators are developed in the proposed scheme. The fuzzy voter is based on the fuzzy voting scheme. The new SFDIA scheme has a more reliable redundancy based on the FCC NN. The performance of the SFDIA scheme is validated by experiments of quadrotor with two gyro sensor modules. It can detect the sensors failure, pinpoint what sensor fails, and replace the sensor with the other normal sensor.