基于最佳测试点选取的改进mRMR-SVDD模拟电路故障诊断

发布时间：2018-05-05 14:02

本文选题：模拟电路 + 故障诊断　；参考：《南京邮电大学》2017年硕士论文

【摘要】：随着科学技术的发展,电子电路取得了长足的发展,各式各样的电子设备给人们的日常生活带来极大的便利。一旦模拟电路发生故障,整个电路的性能会发生严重的畸变,甚至会带来毁灭性灾难。对模拟电路进行故障诊断与预防,是保证整个电子电路系统正常运行的关键,对于改善电子设备性能意义重大。本文从测试点提优、最优特征子集提取、故障模式分类三个方面对模拟电路诊断技术进行优化,具体如下:(1)修正灰色关联分析的测试点选择模型由于实际可供提取的模拟电路故障样本数量有限,反映的故障信息不够丰富,模拟电路具有灰色系统的特征,可以将模拟电路系统看成灰色系统。本文在阐述灰色关联分析模型的基础上,分析该模型的不足并对其进行改进,以灰色熵作为衡量不同测试点敏感度的标准,提出一种修正灰色关联分析的故障测试点选择技术。(2)平衡权重因子的mRMR原则最优故障特征选择模型本文对Wiener级数的理论知识进行阐述,给出优化低阶Wiener核计算量的方式,给出基于Wiener核的特征提取技术。其次,介绍信息学中mRMR原则,提出其不足并进行改进。最后,构建平衡权重因子的mRMR原则模拟电路故障特征选择模型。(3)自适应梯度算法的组合核SVDD故障分类模型该分类模型选用高斯核函数与多项式核函数的组合最作为SVDD核函数,新的核函数具有较好的泛化能力和分类精度。在寻找组合核函数的权重系数的过程中,利用改进梯度算法,解决梯度下降算法在极值点处局部最小化问题。
[Abstract]:With the development of science and technology, electronic circuits have made great progress. Once the analog circuit fails, the performance of the whole circuit will be seriously distorted and even bring about disaster. Fault diagnosis and prevention of analog circuits is the key to ensure the normal operation of the whole electronic circuit system, and it is of great significance to improve the performance of electronic equipment. This paper optimizes the analog circuit diagnosis technology from three aspects: test point optimization, optimal feature subset extraction and fault mode classification. The test point selection model of the modified grey correlation analysis is not rich enough because of the limited number of fault samples that can be extracted from the analog circuit, and the analog circuit has the characteristics of grey system. The analog circuit system can be regarded as a grey system. On the basis of describing the grey relation analysis model, this paper analyzes the deficiency of the model and improves it. The grey entropy is used as the criterion to measure the sensitivity of different test points. This paper presents an optimal fault feature selection model based on the mRMR principle, which modifies the fault test point selection technique of grey relational analysis. The optimal fault feature selection model is based on the theory of Wiener series, and the way to optimize the computation of low order Wiener kernel is given. A feature extraction technique based on Wiener kernel is presented. Secondly, introduce the principle of mRMR in informatics, put forward its deficiency and improve it. Finally, the mRMR principle of balance weight factor is constructed to simulate the fault feature selection model of circuits. (3) the combined kernel SVDD fault classification model of adaptive gradient algorithm. The classification model selects the combination of Gao Si kernel function and polynomial kernel function as SVDD kernel function. The new kernel functions have better generalization ability and classification accuracy. In the process of searching for the weight coefficient of the combined kernel function, an improved gradient algorithm is used to solve the local minimization problem at the extremum of the gradient descent algorithm.
【学位授予单位】：南京邮电大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN710

【参考文献】