Abstract: A fault diagnosis method of combining RS(rough set) and improved QPSO-RBF NN(quantum-behaved particle swarm optimization-radial basis function neural network) is proposed for valve fault,in which wavelet packet energy spectrum from vibration signals of bonnet is taken as fault characteristic parameters.Firstly,attributes of characteristic parameters are reduced by RS theory in order to delete redundant attributes and simplify the inputs of RBF NN,then QPSO algorithm with mutation operator is introduced into the learning process of RBF NN to improve its existing learning algorithms and enhance its predictive ability.The simulation results of valve fault on 6135D type diesel engine show that the method enhancs accuracy and efficiency of fault diagnosis.