Abstract: Aiming at nonlinearility，couplement and noise exsiting in measurements，QPSO (Quantum Particle Swarm Optimization) algorithm was introduced into parameter selection，and a fault diagnosis method was presented combined with radial basis network. Number of neighborhood and simplified dimension are key problems of manifold learning. Firstly，this method uses QPSO algorithm to select number of neighborhood，simplifying dimension and parameter in radial basis function optimally. Secondly，it uses ISOMAP to reduce nonlinear dimension of original parameter and it extracts features in low dimension manifold，and classifies fault. This method was introduced into aeroengine fault diagnosis. The results indicate that this method can classify every composite fault effectively，and the accuracy reaches 97.33%，the result of QPSO is superior to basic PSO (Particle Swarm Optimization) algorithm. Its accuracy is abviously superior to PCA(Principal Component Analysis)，KPCA(Kernel Principal Component Analysis)，and it resists strongly to noise.

Key words: Manifold learning；Quantum particle swarm optimization(QPSO)；Isomap；Principal component analysis；Kernel principal component analysis；Radial basis network；Fault classification

摘要： 针对测量参数存在的非线性、参数间的耦合性以及噪声干扰，将量子粒子群算法引入到流形学习的参数选择中，结合径向基神经网络，提出了一种故障诊断方法。邻域个数和约简维数是流形学习中的关键问题。结果表明:该方法首先利用量子粒子群算法优选邻域个数、约简维数和径向基函数的参数，再利用等距特征映射(ISOMAP)对原始参数进行非线性降维，提取其低维流形特征，从而进行故障分类。结果表明:该方法能够有效地对发动机各种复合故障进行分类，精度达到97.33%，量子粒子群优于基本粒子群优化的分类结果；其分类精度明显优于主元分析(PCA)、核主元分析(KPCA)方法，且有很强的抗噪能力。

关键词: 流形学习；量子粒子群优化；等距映射；主元分析；核主元分析；径向基网络；故障分类