【作者】 李传亮；

【导师】 王友仁；

【作者基本信息】 南京航空航天大学 ， 检测技术与自动化装置， 2011， 硕士

【摘要】 故障模式分类是模拟电路故障智能诊断方法的关键,它的研究对于提高故障诊断的准确率、保障诊断的有效性具有重要的意义。SVDD（支持向量域描述,Support Vector Domain Description）不仅具有参数少、精度高、全局最优等优点,而且效率高、扩展性强、有解决在线诊断问题的潜力,在模拟电路故障诊断中具有广阔的应用前景。传统模拟电路故障诊断方法中的参数识别诊断法和故障验证法,虽然性能不是十分理想、适用范围有限,但是依然有智能诊断方法替代的优势,对它们进行深入研究必定能够促进模拟电路故障诊断的较快发展。本论文以SVDD理论和遗传算法参数辨识为基础,研究了三种基于SVDD的模拟电路故障分类方法和两种基于参数辨识的模拟电路故障诊断方法:（1）基于离散度支持向量预选取SVDD的模拟电路故障分类方法。基于SVDD的模拟电路故障诊断方法中,训练样本数随着故障模式的增加而大幅增加,而SVDD应用于大规模数据集会遭遇计算时间和存储空间的瓶颈。针对这个问题,本文提出了样本离散度的概念,从理论上说明了离散度和支持向量的关系,并利用离散度对SVDD的支持向量进行预选取构成约简的训练样本集,以减少SVDD故障分类器的训练样本数。实验结果表明,离散度精确地反应了样本和支持向量之间的关系,与其他度量标准相比精度更高。基于离散度支持向量预选取的SVDD故障分类方法能够在保证分类精度的前提下,大幅提高故障分类器的训练效率、降低存储空间的需求。（2）基于二次映射SVDD的模拟电路故障分类方法。二次映射SVDD利用二次映射的方法,能够获得一个紧凑、对样本分布适用性更好的描述边界线。利用二次映射SVDD构造的故障分类器可以有效的减少故障模式空间的重叠、提高故障分类的正确率。故障诊断实验表明,基于二次映射SVDD的模拟电路故障分类方法有效的改善了SVDD故障分类的性能,提高了SVDD在模拟电路故障诊断中的适用性。（3）基于全样本SVDD的模拟电路故障分类方法。SVDD作为一种单值分类方法,忽略非支持向量所包含的样本信息对其多类分类的性能有较大影响。本文利用核密度估计和分类器模糊组合的思想,在传统SVDD故障分类决策规则中融入了非支持向量所包含的样本信息,提出了基于全样本SVDD的模拟电路故障分类方法。故障诊断实验表明,与传统基于SVDD的模拟电路故障诊断方法相比,基于全样本SVDD的模拟电路故障分类方法的精度得到较大幅度的提高,并且鲁棒性强、参数选取方便、诊断结果可靠。（4）基于参数辨识的模拟电路故障诊断方法,包括基于系统参数辨识的模拟电路模块级故障诊断方法、基于参数辨识的模拟电路故障验证诊断法两种。基于系统参数辨识的模拟电路模块级故障诊断方法根据系统参数和电路模块的对应关系,利用遗传算法辨识得到的系统参数对电路模块的故障进行定位并判别故障程度。基于参数辨识的模拟电路故障验证诊断法结合了故障验证法和参数辨识法的优点,利用遗传算法参数辨识验证故障元件集的同时得到故障元件的参数值。实验结果表明,两种诊断方法是有效的,适用于模拟电路参数性故障的诊断。本论文的研究工作受国家自然科学基金（60871009）、航空科学基金（2009ZD52045）以及江苏省研究生科研创新计划项目（CX10B₀98Z）的资助。更多还原

【Abstract】 Fault classification is the key of intelligent fault diagnosis methods of analog circuit.The study of it has great significance on improving the fault diagnosis accurancy, and ensuring the effectivity of the diagnosis methods. SVDD not only has few parameters, but also has high efficient and strong expansibility.It also has the potential to solve the problem of on-line fault diagnosis, and has broad prospects of application on analog circuit fault diagnosis. The traditional fault diagnosis methods of analog circuit, suach as parameter identify and fault verification, don’t have good performance in some applicable, but these methods still have advantages that the intelligent fault diagnosis methods don’t have.This paper proposes three kinds of analog circuit fault classification methods based on SVDD and two kinds of fault diagnosis methods based on parameter identification.（1） Analog circuit fault classification method based on SVDD with support vector pre-extracting. The number of training samples of SVDD will be very big when there are many failt patterns, then the time and space complexity will become very high.This paper proposes the definition of dispersion of sample.And the relation between dispersion of a sample and the possibility of support vector is analyzed in theory.Training samples are extracted according to this relation, by which the reduced training sets can be formed.Experiment results show that the proposed method can effectively reduce the number of training sets; also it has better performance than other support vector pre-extracting methods. Analog circuit fault classification methods based on SVDD with support vector pre-extracting based on dispersion can effectively reduce the time and space complexity of SVDD fault classifier.（2） Analog circuit fault classification method based on second mapping SVDD.Second mapping SVDD can get a more tight description boundary with high adaptability to the distribution of sample. Analog circuit fault classifier based on second mapping SVDD can effectively reduce the overlapping of fault pattern spaces, and improve the fault diagnosis accurancy. Experiment results show that the fault classifier of analog circuit based on second mapping SVDD has improved the performance of the fault classifier based on SVDD, and it can be widly used in anlalog circuit fault diagnosis.（3）Analog circuit fault classification method based on All Samples SVDD （AS-SVDD）.SVDD is a one-class classification method,and its performance will be low in multi-class classification if the sample information of non-support vectors are ignored completely.A fault classification method called All Samples SVDD is proposed, which merge the sample information of non-support vectors into the fault classification rule.Experiment results of analog circuit fault diagnosis show that the accurancy of AS-SVDD is highly improved compared with normal fault classifier based on SVDD.And AS-SVDD is more robust, the parameter slection is more easy, diagnosis results is more reliable.（4） Analog circuit fault diagnose methods based on parameter identification that based on genetic algorithm, including the method based on system parameter identification and the fault verification method based on parameter identification.The first one locate the fault of analog circuit module and estimate the grade of the fault according to the relasionship between system parameter and circuit module based on the system parameters that identified by GA.The second method integrate the advantages of fault parameter identify method and fault verification method,it use GA to verify the set of fault components and identify the parameter of the fault components. Experiment results show that these two methods are effective, and they are suitable for the diagnosis of parameter faults.The research of this paper is funded by Chinese National Natural Science Foundation （60501022）, Aeronautical Science Foundation of China （2009ZD52045）, Postgraduates Research and Innovation Program of Jiangsu Province（CX10B₀98Z）.更多还原