基于深度特征学习的电子电路故障诊断

发布时间：2018-10-13 12:54

【摘要】：电子电路故障诊断一直是电路领域里的研究热点。随着电子电路规模和集成度进一步提高,以及电子电路模拟部分的元件的非线性、容差效应问题的影响,电子电路故障特征呈现复杂的非线性情况,对于电子电路故障的特征提取和诊断技术提出更高的要求。对此,本文提出一种基于深度特征学习的电子电路故障诊断方法。由于电路故障特征的提取实质在于找到故障响应信号的非线性表达,而深度学习技术恰是一种对数据逐层地非线性映射表达实现特征提取的技术。因此本文结合电路故障特征提取与深度学习技术的特点,展开了对电路故障的深度特征提取的研究。本文的研究主要集中在以下2点:(1)提出一种基于SAE-SOFTMAX的电路故障特征提取方法。该方法通过构建堆叠自动编码器(SAE)和SOFTMAX分类器的深度学习框架,结合无监督的预训练和有监督的全局微调,优化堆叠自动编码器的特征提取性能,实现了电路故障特征的深度提取。(2)给出两种基于SAE特征提取的电子电路故障诊断模型。包括基于SAESOFTMAX的诊断模型和基于SAE-SVM的诊断模型。前者融合SAE与SOFTMAX分类层实现电路故障特征的快速提取和诊断,后者以SAE提取特征、与鲁棒特性强的SVM结合构成电路故障诊断模型。最后两例电路仿真实验表明:基于SAE-SOFTMAX的电路故障特征提取技术具有很好的效果,较传统的小波和主元分析法所提取具有明显的技术优势、特征评估指标高;本文提出两种电路故障诊断模型均未错分。对于基于SAE-SOFTMAX的诊断模型,在诊断效果及性能都优于传统的神经网络诊断模型。
[Abstract]:Fault diagnosis of electronic circuits has been a hot topic in the field of circuits. With the further improvement of the scale and integration of electronic circuits, as well as the nonlinearity of components in the analog part of electronic circuits and the effect of tolerance effect, the fault characteristics of electronic circuits present complex nonlinear situations. Higher requirements for feature extraction and diagnosis of electronic circuit faults are put forward. In this paper, a method of electronic circuit fault diagnosis based on depth feature learning is proposed. Because the essence of circuit fault feature extraction is to find the nonlinear expression of fault response signal, the depth learning technique is just a kind of feature extraction technology for the data layer by layer nonlinear mapping expression. Therefore, combining the characteristics of circuit fault feature extraction and depth learning technology, the research of circuit fault depth feature extraction is carried out in this paper. This paper focuses on the following two points: (1) A fault feature extraction method based on SAE-SOFTMAX is proposed. This method optimizes the feature extraction performance of stack automatic encoder by constructing the depth learning framework of (SAE) and SOFTMAX classifier, combining unsupervised pre-training and supervised global fine-tuning. The depth extraction of circuit fault features is realized. (2) two fault diagnosis models of electronic circuits based on SAE feature extraction are presented. It includes diagnosis model based on SAESOFTMAX and diagnosis model based on SAE-SVM. The former fuses SAE and SOFTMAX classification layer to realize the fast extraction and diagnosis of circuit fault features, the latter uses SAE to extract features and combines with robust SVM to form a circuit fault diagnosis model. The last two circuit simulation experiments show that the circuit fault feature extraction technology based on SAE-SOFTMAX has a good effect, and has obvious technical advantages over traditional wavelet and principal component analysis methods, and has a high feature evaluation index. In this paper, two fault diagnosis models are presented. For the diagnosis model based on SAE-SOFTMAX, the diagnostic effect and performance are better than the traditional neural network diagnosis model.
【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】：TN707

【相似文献】