The Application of Tolerant Rough Set Neural Network to Fighter Fault Diagnosis

Abstract

Conventional rough set theory is based on indiscernibility relation, which lacks the adaptive ability to data noise or data missing. Furthermore, it may present qualitatively whether or not the faults exist, but it cant compute accurately the value of the faults. Though the neural network has ability of approximating unknown nonlinear systems, but it cant distinguish the redundant knowledge from useful knowledge, so its classification ability cant catch up with the rough set classifier. This paper combines the rough set theory and the tolerant rough set neural network to diagnose the rudder faults of fighter, which solves well the problem of fault diagnosis and fault degree computation. Simulation results demonstrate the effectiveness of the proposed method.