基于核方法的模拟电路故障诊断

Analog Circuit Fault Diagnosis Based on Kernel Methods

摘要:

核方法通过非线性映射将原始数据嵌入到高维特征空间,然后进行线性分析和处理,为基于知识的数据分析带来新的方法和模式;传统方法无法解决故障特征数据维数高、在故障样本交叠严重时多分类性能较差的问题,因此在电路故障特征数据预处理阶段,提出了分步骤分别在时域对电路输出电压波形进行小波包分析和在频域测量电路幅频特性的方法来提取电路故障特征；预处理后的故障特征向量只是8维向量,减少了SVM的训练时间;将该方法应用于国际标准电路中的CTSV滤波器电路的故障诊断,结果表明:该方法能突出不同故障的特性,故障诊断正确率达到98.57%(414/420)。

Abstract:

New methods and models for knowledge-based data analysis are brought by kernel method. It uses nonlinear mapping, the original data is embedded into a high dimensional feature space, then conduct the linear analysis and processing. The traditional methods cannot solve the problem of poor classification performance in the case of high dimensionality of fault feature data and serious overlap of fault samples, therefore in the circuit fault feature data pre-processing stage, we propose a method of extracting circuit fault feature.The method step by step, respectively, conduct wavelet packet analysis of Circuit Output Voltage Waveform in the time domain and measure amplitude-frequency characteristics of the circuit in the frequency domain; Preprocessed fault feature vectors is just eight-dimensional vector, thus reducing training time of SVM .When the method is applied in the fault diagnosis of the CTSV filter circuit in international standard circuit, the results have shown that this method can highlight the different characteristics of the fault, and the fault diagnostic correct rate of 98.57% is achieved.