基于时频分析和复数域的模拟电路故障诊断研究

发布时间：2018-08-17 18:42

【摘要】：面对电路的高速集成和规模化,人们迫切的需要研究更加先进的、高效的、智能的模拟电路故障诊断理论和技术,以此满足现代电子工业领域对电子系统安全性、可靠性及可测试性的高要求。其中,如何提取出故障的可辩特征以避免因故障之间的混叠造成故障的不可分已成为模拟电路故障诊断的研究焦点。时频分析方法能在时间和频率上同时提供信号的分布信息,能清晰的刻度任意时刻的信号频率和幅值。复数域分析通过把时域信号变换到复数域,用实部和虚部来联合刻画信号在复平面的分布状况。这两种方法都能很好的为故障诊断提供更多可辨的细节特征。本文围绕时频分析技术和复数域分析进行深入研究了模拟电路的参数故障诊断并提出了新的诊断方法。另外本文也对电子系统测试性验证做了一定的深入研究。本文取得的成果如下:(1)提出了基于集合经验模态分解和极限学习机(EEMD-ELM)的模拟电路故障诊断方法。通过研究希尔伯特-黄变换以及其经验模态分解变换的原理,将EEMD变换引入到模拟电路故障诊断中。由此提出了一种EEMD联合相对熵和峭度构建模拟电路故障特征向量,通过ELM进行单、多参数故障诊断的方法。首先通过分别采集电路正常状态和故障状态响应输出信号,然后使用EEMD把响应输出各自分解成一组本征模态函数(IMF),计算出所有电路状态IMF的峭度及电路正常状态IMF和故障状态IMF之间的相对熵,构建成故障特征向量以此作为ELM的输入样本进行故障的诊断。仿真结果表明该基于EEMD分解所建立的故障特征在ELM分类器中具有很低的计算代价和较高的诊断率。(2)对局部均值分解(LMD)的模拟电路故障特征向量优化进行了研究,提出了新的基于聚类方式的优化策略。本文深入的研究了LMD的原理及分解过程。并采用LMD技术把被测电路的响应输出信号分解成一系列的乘积函数(PF)信号;以各PF信号的能量熵、标准差和峭度等参数构造特征向量,这种特征向量的维度会随着信号分解的PF个数的增加而增加。因此本文提出一种新的特征优化策略对其进行降维,首先以每种故障模式的各特征项的聚类中心和半径为依据,通过优化策略从这些特征项中选出最优特征项,组成新的特征向量,以此作为LVQ神经网络的输入。仿真实验证明该策略能使故障特征向量维数和分类器计算量有效的降低,同时也能有效进行故障诊断。(3)提出了基于最小二乘圆拟合算法的复数域故障建模及故障诊断方法。实际模拟电路的故障响应是连续无穷的,但是在传统的故障诊断模型字典法中所存储的故障特征值是离散的,这必然造成字典中故障类型的不完备。为了提高字典的完备性和解决元件容差影响,通过深入研究斜率故障模型和复数域故障模型,并在模拟电路复数域故障建模理论基础上提出使用最小二乘圆拟合算法拟合出故障特征函数作为故障特征,并针对该模型提出了相应的故障诊断方法,仿真实验和实际电路实验都验证了该方法能很好的实现故障诊断。(4)基于蚁群算法的故障样本选择。在传统的测试验证实验中,基于故障率的随机测试样本选择方法往往忽略了具有较小故障率的传播故障的选择,但是传播故障一旦发生可能引起很严重的扩散故障。为解决这个问题,提高对传播故障的选择,本文利用有向图法和蚁群算法搜索传播故障的最优传播路径,然后为每个故障模块(器件)建立后续传播路径集,根据新建的样本选择策略,优化测试样本。模拟实验仿真证明该策略能提高传播故障被选择率,提高测试能力和降低使用风险。
[Abstract]:Facing the high-speed integration and large-scale of circuits, people urgently need to study more advanced, efficient and intelligent analog circuit fault diagnosis theory and technology, so as to meet the high requirements of electronic system security, reliability and testability in the field of modern electronic industry. Faults caused by overlapping between obstacles have become the focus of analog circuit fault diagnosis. Time-frequency analysis can provide the distribution information of signals at the same time in time and frequency, and can clearly scale the frequency and amplitude of signals at any time. Complex-domain analysis transforms the time-domain signals into complex-domain signals, using real and imaginary parts. Both of them can provide more distinguishable detail features for fault diagnosis. This paper focuses on time-frequency analysis and complex domain analysis to study the parameter fault diagnosis of analog circuits and proposes a new diagnosis method. The results are as follows: (1) A fault diagnosis method for analog circuits based on set empirical mode decomposition (EMD) and Extreme Learning Machine (EEMD-ELM) is proposed. In this paper, a method of constructing fault eigenvectors of analog circuits by EEMD combined with relative entropy and kurtosis is proposed. The single-parameter and multi-parameter fault diagnosis is performed by ELM. First, the output signals of normal state and fault state responses are collected respectively, and then the response outputs are decomposed into a set of intrinsic mode functions (IMFs) by EEMD. The kurtosis of the circuit state IMF and the relative entropy between the circuit normal state IMF and the fault state IMF are constructed as fault feature vectors, which are used as input samples of the ELM for fault diagnosis. The fault eigenvector optimization of analog circuits based on local mean decomposition (LMD) is studied and a new optimization strategy based on clustering method is proposed.The principle and decomposition process of LMD are studied in detail.The response output signal of the circuit under test is decomposed into a series of product function (PF) signals by LMD technology. The dimension of eigenvectors will increase with the number of PFs decomposed by signals. Therefore, a new feature optimization strategy is proposed to reduce the dimension of eigenvectors. The simulation results show that this strategy can effectively reduce the dimension of fault feature vectors and the amount of classifier calculation, and also can effectively diagnose faults. (3) A complex domain fault modeling method based on least square circle fitting algorithm is proposed. Fault response of analog circuits is continuous and infinite, but the fault eigenvalues stored in the traditional fault diagnosis model dictionary are discrete, which inevitably leads to incomplete fault types in the dictionary. Based on the theory of fault modeling in complex domain of analog circuit, the least square circle fitting algorithm is proposed to fit the fault characteristic function as the fault characteristic, and the corresponding fault diagnosis method is proposed for the model. The simulation and actual circuit experiments show that the method can be well implemented. Fault diagnosis. (4) Fault sample selection based on ant colony algorithm. In traditional test validation experiments, random test sample selection method based on fault rate often ignores the selection of propagation faults with small fault rate, but propagation faults may cause very serious diffusion faults once they occur. In this paper, the directed graph method and ant colony algorithm are used to search the optimal propagation path of propagation faults, and then a set of subsequent propagation paths is established for each fault module (device). According to the new sample selection strategy, the test samples are optimized. Simulation results show that the strategy can improve the selection rate of propagation faults and test performance. Strength and reduce use risk.
【学位授予单位】：电子科技大学
【学位级别】：博士
【学位授予年份】：2017
【分类号】：TN710

【参考文献】