基于神经网络的大规模模拟电路子网络级故障诊断方法
详细信息    本馆镜像全文|  推荐本文 |  |   获取CNKI官网全文
摘要
本论文主要研究基于神经网络的大规模模拟电路子网络级故障诊断方法。在已有知识的基础上,提出了完整的诊断方法和实现方案,通过计算机仿真证实了该方法并且快速、准确。针对如何对大多数电路的模拟故障分析仿真这个环节我们是由易及难的拓展,先是简易的电路,进而向更为复杂的电路方向发展。这个设计方法是根据多方的资料汇集出来的一种方法,也是比较新颖和有深度的一个方法,是对电路故障诊断方法的一个新探讨,具有广阔的应用前景,为大规模模拟电路故障诊断提供了新的理论依据和检测方法,并有希望研制成一套高效的检测设备。
    随着大规模集成电路技术的迅速发展及日益广泛应用,为了维护各种器件及设备,人们必须借助计算机来找出电路的故障,模拟电路故障诊断已成为大规模集成电路课题中令人瞩目的一个课题,是一个具有重要理论意义和应用价值的课题。随着微电子技术的发展,电子系统越来越复杂。一个电子系统通常由大量集成器件和分立元件组成,因此,对故障的诊断变得越来越难。一些已有复杂模拟电路的诊断方法只适用于特定条件下(如开路、短路等)的电路诊断,却很难发现由电路中的电子器件的容差变化引起的软故障。迄今为止,很少有文献对软故障即容差电路的故障诊断给出系统而有效的方法,特别是大规模电路的故障诊断。这是本文研究工作的主要内容。
     近年来,人工神经网络作为信息软处理的最新技术在诊断领域中得到了广泛的应用。因为神经网络具有自学习能力、泛化能力、自适应能力、联想记忆及非线性映射能力,使其故障检测的方法引起了人们的高度重视,已有人利用它来检测模拟电路的故障。但神经网络故障识别的准确性依赖于特征量的有效性和网络结构参数的合理选择。小波分析是一种全新的时-频两维分析技术,其对高频突变信号和低频缓变信号的分析有着独特的优点,可以利用信号在小波变换的多尺度刻划下所表现行为的不同特点来检测出突变的故障信号。在保证较为突出的故障特征情况下,具有用到的特征量少等优点。若数据量不大时,小波变换的时间较短,则小波分析可用
    
    
    于信号的实时分析。若把小波分析的特征提取和神经网络的模式识别技术有效地结合起来应用于电路故障诊断,则是一种非常有前途的检测方法。本文正是基于这一想法,提出了一种故障诊断方法。
    本文的研究工作分为两部分:第一部分提出神经网络诊断模拟电路故障的理论和实现方法。在这一部分首先介绍了小波变换与神经网络的基本知识,然后在此基础上提出了电路故障诊断的分析方法,并对其可行性进行了分析。第二部分是针对大规模电路故障检测方法的具体研究。这一部分首先选定了具体采用的检测方法,详细介绍了本课题研究方案的原理、组成及设计,最后进行了模拟诊断仿真实验,证明该方法的可行性。
    针对传统BP算法易陷入局部最小、收敛速度慢等弊端,本文采用了基于Levenberg-Marquardt算法和动量法相结合来训练神经网络获得网络权值,形成良好的网络结构。而对于大规模电路,所含元件较多,传统的方法工作量过于庞大,也存在组合数多的缺点。在这个关键问题上,本文运用了网络分析中的置换原理撕裂网络,采用分级诊断,首先将大规模电路中的故障子网络诊断出来,然后在此基础上利用神经网络仅对故障子网络诊断与学习。该方法使整个电路的故障诊断得到了简化,加快了故障诊断的速度。
    本文第四章在已有理论基础上,提出了完整的大规模模拟电路故障诊断方法和实现方案。首先对大规模电路进行正确的子网络划分,采用电压激励测电压,结合灵敏度分析,利用电压比较的方法定位出故障子网络。而对于故障子网络,利用PSPICE蒙特卡罗仿真分析,不断改变电路系统的参数值,形成一个个模拟故障状态,产生反映故障状态的特征量,完成故障仿真,然后用MATLAB语言读取PSPICE仿真所得数据。接着对采集到的模拟仿真信号进行小波分析算法的计算机仿真,构造适当的小波函数及分解阶数进行特征提取,从而为进行电路的故障诊断做铺垫。最后根据提取的特征量作为神经网络的输入进行学习和训练获得网络权值,经过训练后的神经网络即可正确进行故障诊断。
    文中第五章对大规模模拟电路故障诊断方法进行实例仿真,完成电路的故障诊断任务,并能在较短的时间内获得正确的诊断结果,证明该方法具有很高的成功率和鲁棒性。此方法缺点在于提取样本数据时,我们仍需要
    
    
    考虑其它子网络的影响。
    本文研究的重点之一在于选择合适的特征样本作为神经网络的输入,即将小波变换后的特征量进行主要分量分析选择最优特征量进行神经网络的学习和训练;另一重点是如何选择合适的神经网络拓扑结构和算法进行简单、快速、正确的故障诊断。
    全文最后做了总结,归纳了本文所做的工作和结论,同时也提出了几点待解决问题,指出了今后的主要研究和发展方向:(1) 对于故障子网络,在提取样本数据时,我们仍需要考虑其它子网络的影响,使之更适应大规模电路;(2) 本文提出的方法是对大规模电路故障检测的新探索,目前还没有完成一套实际的诊断仪器,所以应对此进行更深入研究;(3) 实际诊断时,我们针对的是简易电路,如何由易到难拓展到复杂电?
In this paper a fault diagnosis of sub-networks in large-scale analog circuits using neural network is proposed. The author proposes an intact diagnosis technique and its implement scheme on the basis of the existing knowledge. The computer simulations validate this approach’s swiftness and nicety. We expand circuits from easy to difficult aiming at how to simulate and analysis most circuits. This method is presented by collecting much information; it is also a novelty and an in-depth study. It is a new grope for analog circuits fault diagnostic technique. This method is so promising and widely applicable that it provides a new theoretic proof and detection approach. It may be executed to a high-efficient facility.
    With the rapid development and application of the LSI technology, in order to maintain various components and devices, we must detect the fault by the aid of computer. So study on a fault diagnostic approach is a significant subject on both theory and application. Electronic systems are getting more and more complicated with the rapid development of microelectronics techniques. An electronic system contains a large number of integrated circuit components and separate elements so that it becomes more and more difficult to detect those complicated systems and devices. The traditional approaches of fault location are performed only if the faults of the circuits are those hard faults,such as open-circuit,short-circuit,etc. Those soft faults aroused by the tolerance of circuit components cannot be easily discovered. At present, few references present a systematic and effective method on the soft faults. That is the capital mission of research work in the dissertation.
    In recent years, artificial neural network as a newest technique of the information soft processing is applied widely in the field of diagnosis. For the capabilities of self-study, robust, self-adaptive, associate memory and nonlinear mapped traits, people pay much attention to its diagnostic technique. But neural network’s accuracy relies heavily on the effective of the features and the reasonable choose for the parameter of the network’s framework. Wavelet transform is a fire-new time-frequency planar analytic technique. Its unique
    
    
    merit to high frequency burst signal and low frequency lazy signal is that it can detect the burst faulty signal using its different behavior to different scales. It can acquire the eigenvectors as few as possible on the condition that it can show the fault behavior. If the data is not much, wavelet transform can be very fast, so wavelet transform can be used at real time. If the features extraction of wavelet transform combines with the pattern recognition of neural network to diagnose the fault,it will be a promising method. The author just proposes a method based on this idea.
    The research work in this paper is divided into two sections: Section1 discusses the theoretical basis and implement approach of fault diagnosis of analog circuits based on neural network. In this section, firstly, the basic knowledge of wavelet transform and neural network is introduced, and then the analyzing means of the circuit fault diagnostic system is proposed, and its feasibility is analyzed. In section2, the large-scale analog circuits fault diagnostic method is described elaborately. This section confirms the detection method at first, and then introduces the theory, makeup and design of the method. At last, circuit simulation by using computer is presented, conclusions are given that it is an effective method.
    BP algorithm encounters local minimum、slow convergence speed and convergence instability, so we adopt a means based on Levenberg-Marquardt algorithm joint with momentum algorithm to train a neural network and obtain the network weights in this paper, forming robust network architecture. For the large-scale analog circuit, the workload of the traditional method is very large, which has many components and need many combinations. On the point, we tear the network in this paper based on the substitution the
引文
[1] Qian Zhihong, Zhao Xiaoming, Dai Yisong. En-In models study of automotive IC modules. Proc. IEEE International Vehicle Electronics Conference, Changchun, China, Sept.6-9, 1999, 99EX 275: 188-192
    [2] Richard. MD, et al. Neural classifiers Estimate Bayesian a posteriori Probabilities Neural Computation, 1991, 3: 461-483
    [3] 王巍,崔海英,黄文虎. 基于故障树最小割集和最小路集的诊断方法研究数据采集与处理. 1999,14(1):26-29
    [4] Patterson H.F.A.and Iverson D.L. An intergrated approach to system design, reliability and diagnosis. NASA Ames Research Cente, 1990, 102861: 1-2
    [5] Iverson D.L.and Patterson H.F.A. A diagnosis system using object-oriented fault tree models.NASA Ames Research Center, 1990, N90-27313: 1-2
    [6] Bander J.W. et al. A non-linear L1 optimization algorithm for design modeling and diagnosis of networks. IEEE Trans.CAS 1987,34(2):174-181
    [7] Bander J.W. et al. Fault isolation in linear analogue circuits using the L1-norm, Proc. IEEE int. Symp. CAS, Rome, Italy, 1982: 1140-1143
    [8] Lowe D, et al. Optimal Feature Extraction and the Bayesian Decision in Feed forward Classifier Networks. IEEE Trans PAMI, 1991, 13: 355-364
    [9] 何怡刚,罗先觉,邱关源. 基于神经网络的线性电路故障诊断非线性L1范数优化方法. 电子测量与仪器学报,1998,12(1):18-22
    [10] Lin P.M. and Elecher Y.S. Analog circuits fault dictionary –new approaches and implementation. Int. J. Circuit Theory and Appl. 1985, 13: 149-172
    [11] Elecherif Y.S. and Lin P.M. An isolation algorithm for the analog fault dictionary. ADA 127020, 1983
    [12] 雍正正. 模拟电路故障诊断:模糊字典法. 数据采集与处理. 1998,13(3):210-213
    [13] Mackey D J C. The Evidence Framework Applied to Classification Networks. Neural Computation, 1992, 4: 720-736
    [14]R Hecht-Nielsen.Theory of backpropagation network.Proc.of International Joint Conf. on Neural Network Vol. I, Washington D.C. 1989, 593-605
    
    [15] 程正兴. 小波分析算法与应用. 西安:西安交通大学出版社,1998
    [16] 胡昌华,张军波,夏军等. 基于MATLAB的系统分析与设计-小波分析. 西安:西安电子科技大学出版社,1999
    [17] Amari S I. A theory of adaptive pattern classification. IEEE Transon Electronic computer. Eo 16, 1967, 299-307
    [18] Hecht-Nielsen R.The theory of backpropagation neural network.Inreview, 1988
    [19] Bryson A, HOYC. Applied Optional Control. New York, Blaisdell, 1969
    [20] 阎平凡,张长水.人工神经网络与模拟进化计算. 北京:清华大学出版,2001
    [21] Maidon Y, ervis B W, Dutton. N, et al. IEE Proc[J]. Circuits Devices Syst.1997, 144(3): 149-154.
    [22] 丛爽. 面向MATLAB工具箱的神经网络理论即其应用. 合肥:中国科学技术大学出版社,1998
    [23] D.J.C. Mackay, “A practical Bayesian Framework for Backpropagation Networks” Neural Computation. 4(3):448-472, 1992
    [24] 胡守仁. 神经网络导论. 长沙:国防科技大学出版社,1993
    [25] 袁曾任. 人工神经元及其应用. 北京:清华大学出版社,1997
    [26] C. M. Bishop. Neural Networks for Pattern Recognition. Oxford University Press, 1995
    [27] Alessandro Sperduli and Antonina Starta. Speed up learning and network optimization with extended Backpropagation Neural Networks. 1993, 6: 365-385
    [28] 楼顺天. 基于MATLAB的系统分析与设计-神经网络. 西安:西安电子科技大学出版社,1998
    [29] 程相君. 神经网络原理及其应用. 北京:国防工业出版社,1995
    [30] 闻新,周露. MATLAB神经网络应用设计. 北京:科学出版社,2000
    [31] 赵建. 模拟电路故障诊断的研究. 上海海运学报,2000,21:61-65
    [32] M. Catelani and M. Gori, “On the Application of Neural Networks to Fault Diagnosis of Electronic Analog Circuits.” Measurement 1996, 17:321-326
    [33] 张友纯. 子网络故障模型及应用. 微电子报,2001,31.(3)
    
    [34] 张萍,王桂增,周东华. 动态系统的故障诊断方法[J]. 控制理论与应用, 2000,17(2):153-158
    [35] 张定会,邵惠鹤. 基于神经网络的故障诊断推理方法[J]. 上海交通大学学报,1999(5):619-621
    [36] Funahashi K I. On the approximate realization of continuous mapping by neural networks, Neural Networks, 1989
    [37] M. T. Hagan, H .B. Demuth. and Mark Beale. Neural Network Design Boston. MA:PWS Publishing, 1996
    [38] 吴今培. 智能故障诊断与神经网络. 北京:科学出版社,1997
    [39] 王晓东,杨德生. 神经网络在火箭容错控制中的应用. 航天控制,1996-0110
    [40] 傅周东. 液压系统泄漏故障诊断技术研究:[博士学位论文]. 杭州:浙江大学,1996
    [41] 赵纪元. 基于小波理论神经网络的实用诊断技术研究[D]. 西安: 西安交通大学,1997
    [42] 杨士元. 模拟系统的故障诊断与可靠性设计. 北京:清华大学出版社,1993,5-9
    [43] N R Karayiannis and A N Venetsanopoulos. Fast learning algorithms for neural networks. IEEE Trans. Circuits Syst., 1992, 39(7): 453-473
    [44] 唐人享. 模拟电子系统的自动故障诊断. 北京:高等教育出版社,西北工业大学学报,1991,214
    [45] 陈圣俭, 蔡金燕. 容差模拟电路子网络级故障诊断的新方法. 电子测量与仪器学报,1998,6(2):61-63
    [46] 罗先觉. 一种基于灵敏度分析的模拟电路子网络级诊断算法[J]. 电子测量与仪器学报,1996,10(1):34-39
    [47] 崔锦泰. 小波分析导论. 西安:西安交通大学出版社,1997
    [48] 飞思科技产品研发中心. MATLAB6.5辅助小波分析与应用. 北京:电子工业出版社,2003
    [49] 朱建良,周欣荣. 模拟电路故障诊断的一种方法[J]. 哈尔滨电工学院学报,1993,16(3):235-239
    
    [50] 赵雅兴. 电子线路PSPICE分析与仿真. 天津:天津大学出版社,1995.
    [51] 邹锐. 模拟电路故障诊断原理及应用. 武汉:华中理工大学出版社,1998
    [52] 包伟,蔡宜三,利用PSPICE仿真研究PWM开关电源. 电工电能新技术,1995,(2):25-29
    [53] 杨喜伟. 一种新的模拟电路故障字典法[J],电子测量与仪器学报,1993,7(3):49-56
    [54] 袁志发,孟德顺. 多元统计分析. 杨陵:天则出版社,1991:201-208.
    [55] 黄凤岗,宋克欧. 模式识别. 哈尔滨工程大学出版社,1991:129-134
    [56] 韩力群. 人工神经网络理论、设计及应用. 北京:化学工业出版社,2002:53-54
    [57] 陈艳峰,丘水生,伍言真等. MATLAB与PSPICE相结合用于开关功率变换器仿真的方法. 电机与控制学报,1999,3(2):99-102

© 2004-2018 中国地质图书馆版权所有 京ICP备05064691号 京公网安备11010802017129号

地址:北京市海淀区学院路29号 邮编:100083

电话:办公室:(+86 10)66554848;文献借阅、咨询服务、科技查新:66554700