基于流形特征提取与优化的电子电路故障诊断

发布时间：2018-07-02 13:15

本文选题：特征提取 + 模式识别　；参考：《合肥工业大学》2017年硕士论文

【摘要】：随着电子设备中集成电路规模和复杂度的不断提升,系统功能不断完善,同时也带来了电路可测节点少、故障潜在信息难以获取、故障诊断准确率低等问题,对电子电路故障诊断技术的可靠性研究提出更高的要求。研究电路的故障诊断技术关键在于实现有效的故障特征提取和精准的故障模式识别,本文从这两个核心方面开展电路故障诊断技术的研究,分别提出了融合LLE和MDS算法的故障特征提取技术以及基于DCQGA的优化SVM分类器模型,并通过电路诊断实例对该方法进行了仿真验证。本文主要研究内容如下:首先,研究典型电子电路的建模原理及故障特性。通过对电子电路的软故障和硬故障的特性进行分析,构建电路故障模型,以故障电路可及节点的输出响应作为电路的原始多维故障特征数据。其次,提出融合线性和非线性流形学习算法的特征提取技术,即基于LLE和MDS算法的故障特征提取。该融合算法可以最大限度挖掘故障数据中潜在的低维流形结构,并保持数据的距离相似性不变,达到提取与优化故障特征的目的。利用该算法对电路的多维故障特征数据进行降维,结果表明降维后的故障特征分布区分性较明显。最后,建立一种用于电子电路故障诊断的优化分类器模型,即基于双链量子遗传算法的支持向量机分类器模型。DCQGA-SVM分类器具有双链寻优特征,可以加速寻优过程,增加获得全局最优参数的概率。通过UCI公共数据库中的两个经典分类数据集对分类器的可行性进行了测试,最后以电子电路故障诊断实例对优化改进分类器的收敛性能和故障诊断性能进行了验证。实验结果不仅验证了DCQGA-SVM良好的分类性能,而且验证了其在电路故障诊断中的有效性和实用性。
[Abstract]:With the improvement of the scale and complexity of integrated circuits in electronic devices, the system functions are constantly improved, and it also brings many problems, such as less measurable nodes of circuits, difficulty in obtaining potential fault information, low accuracy rate of fault diagnosis, etc. Higher requirements are put forward for the reliability research of electronic circuit fault diagnosis technology. The key to research the fault diagnosis technology of circuit is to realize effective fault feature extraction and accurate fault pattern recognition. In this paper, the research of circuit fault diagnosis technology is carried out from these two core aspects. A fault feature extraction technique based on LLE and MDS algorithm and an optimized SVM classifier model based on DCQGA are proposed respectively. The main contents of this paper are as follows: firstly, the modeling principle and fault characteristics of typical electronic circuits are studied. By analyzing the characteristics of soft and hard faults of electronic circuits, the fault model of circuits is constructed, and the output response of reachable nodes of fault circuits is taken as the original multidimensional fault characteristic data of circuits. Secondly, a feature extraction technique based on LLE and MDS is proposed, which combines linear and nonlinear manifold learning algorithms. The fusion algorithm can maximize the mining of potential low-dimensional manifold structures in fault data and keep the distance similarity of the data unchanged to achieve the purpose of extracting and optimizing fault features. The algorithm is used to reduce the dimension of the multi-dimensional fault feature data of the circuit. The results show that the fault feature distribution after dimension reduction is more distinct. Finally, an optimal classifier model for fault diagnosis of electronic circuits is established, that is, the support vector machine classifier model based on double chain quantum genetic algorithm. DCQGA-SVM classifier has the characteristics of double-chain optimization, which can accelerate the optimization process. Increase the probability of obtaining global optimal parameters. The feasibility of the classifier is tested by two classical classification data sets in UCI common database. Finally, the convergence performance and fault diagnosis performance of the improved classifier are verified by an example of electronic circuit fault diagnosis. The experimental results not only verify the good classification performance of DCQGA-SVM, but also verify the validity and practicability of DCQGA-SVM in circuit fault diagnosis.
【学位授予单位】：合肥工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN710

【相似文献】