【作者】 田锡宇；

【导师】 王友仁；

【作者基本信息】 南京航空航天大学 ， 测试计量技术及仪器， 2009， 硕士

【摘要】 本文在系统研究人工免疫算法的基础上,将人工免疫算法与智能诊断方法相结合,对模拟电路的智能故障诊断作了深入的研究,完成了如下研究工作:(1)设计了一种基于克隆选择和聚类的模拟电路故障分类器,该算法的亲和力函数是借助模糊C均值算法的目标函数定义的,有利于区分克隆选择算法在模拟电路故障诊断中存在故障重叠的部分,有效的提高了故障诊断率,同时将克隆选择和聚类算法结合寻找故障的聚类中心,有效的提高了算法的收敛速度。(2)研究了一种基于免疫记忆网络理论与k近邻算法的模拟电路故障诊断新方法。首先,利用免疫记忆网络寻找各故障空间的最佳记忆抗体。在免疫记忆网络中根据浓度来选择记忆抗体,以促进记忆抗体在各故障空间的均匀分布。利用克隆和超级变异机制来增强抗体多样性,再利用浓度和期望值对抗体进行促进和抑制,以避免早熟现象的产生;然后,根据所得到的各故障空间的最佳记忆抗体,使用改进的阈值k近邻算法对抗原进行故障分类;最后,以带通滤波器为诊断实例,利用实际电路测试数据和仿真数据作为测试样本进行故障诊断性能评估,实验结果证明了新故障诊断方法具有高的故障诊断率。(3)开发了基于研华PCI1714UL数据采集程序,完成单通道探测激励源,单通道进行DMA高速数据采集,最高采样频率为10MHz,利用数据采集程序采集模拟电路的故障特征向量,以备后续验证故障诊断方法是否有效。更多还原

【Abstract】 In this dissertation,by combing AIA and computational intelligence fault diagnosis appraoch,an intelligent diagnosis method for analog fault diagnosis is deeply researched,The main achievement are included as followed:(1) This paper proposed a fault sorter based on Clone Selection algorithm and clustering, the algorithm’s affinity function is defined in virtue of fuzzy C-mean’s destination function.it is available to distinguish the parts of fault overlapped and improves the fault diagnosis precision.At the same time it make use of combining Clone Selection Algorithm and Clustering to search the fault clustring center,it is capable of curtail the algorithm’s convergence time.(2) A new method of analog fault diagnosis based on Immune Memory Network theory and k nearest neighbor algorithm was proposed. First, immune memory network was used to search best memory antibody in fault space. In order to equally distribute the memory antibody in fault space, the memory antibody in immune memory network were chosen according to concentration. The mechanism of clone and hyper-variation were used to increase the diversity of antibody, and methods including stimulating and suppressing antibody by concentration and living expectation were applied to avoid immaturity convergence. Second, KNN algorithm based on threshold of antibody fitness value and set of best memory antibody in fault space were integrated to classify the antigen. At last, the band-pass filter was taken as an example, both of data from real circuit and data from software simulation were provided to test the proposed fault classifier. The experimental results prove that the new method for analog circuit fault diagnosis increase the diagnosis precision.(3) Develop a data sample program based on Advantech PCI1714UL,it has some functions as follows: one channel detect the stimulus and another channel high rate data sample under DMA mode.The max sample frequency is 10 MHz. Data sample programming are used to get fault character vectors from analog circuit which will be used to validate whether the fault diagnosis method is avilable.更多还原