多电飞机电源系统智能机内测试诊断技术研究  

Research on Intelligent Built-in Test Fault Diagnosis of More-electric Aircraft Electrical Power System

作　　者：刘震[1] 林辉[1] 

机构地区：[1]西北工业大学自动化学院,陕西西安710072

出　　处：《兵工学报》2007年第11期1357-1362,共6页Acta Armamentarii

基　　金：航空科学基金资助项目(04F53036)

摘　　要：提出了一种基于混合神经网络模型的智能机内测试(BIT)方法,并将其应用在多电飞机电源系统智能BIT故障诊断当中。对BIT各个测点的信号进行功率谱分析,在不损失谱线特征的前提下,利用小波的多尺度分析对谱线数量进行压缩,提取出信号在频域的特征量,并同小波包分解后的频带能量组成诊断用的特征向量。针对原有广义学习矢量量化(GLVQ)神经网络的算法缺陷进行改进,并在此基础上提出了一种混合网络模型结构。将提取的特征向量作为混合网络的学习样本,经训练后对电源系统的故障进行诊断。结果表明,基于这种混合网络的智能BIT方法诊断精度高,对测量噪声也具有良好的鲁棒性,可以有效提高多电飞机电源BIT系统的诊断性能。An intelligent built-in test（BIT） technology based on generalized learning vector quantization（GLVQ） neural network was proposed and applied to the BIT system of more-electric aircraft electrical power system（MEAEPS）. A power spectrum analysis method was employed to get the characteristics of BIT signals in frequency domain. In order to reduce the dimension of eigenvectors, the spectrum characteristics of BIT signals were compressed by using the wavelet packet decomposition, and the energy of each frequency-band was computed to form the final eigenvectors, which were used as learning samples to train the GLVQ neural network. Since the original GLVQ algorithm suffered from several major problems, some modifications were made and a supervised LVQ layer was added to the GLVQ network, which made the boundaries among the fault classes more discriminative thanusing the GLVQ network alone. The proposed method was applied to the BIT system of MEAEPS, and the results showed that the proposed method is promising to improve BIT performance of MEAEPS.

关 键 词：自动控制技术 多电飞机 智能BIT 神经网络 小波包 故障诊断 

分 类 号：V242[航空宇航科学与技术—飞行器设计]