基于改进小波神经网络的模拟电路故障诊断研究

发布时间：2018-06-24 18:56

本文选题：故障诊断 + 神经网络　；参考：《湖南师范大学》2015年硕士论文

【摘要】：模拟电路故障诊断技术的研究开始于1960年,目前已在国内外取得了大量有效的科研成果,逐步形成了完善的系统理论,在电路理论中占据非常重要的地位。同时,随着电子工业的飞速发展,电器设备的集成度越来越高,日趋模块化和功能化。但是由于模拟电路自身存在的非线性、连续性、元器件参数容差等特性使得模拟电路故障诊断的难度非常大。采用传统的模拟电路故障诊断方法已难以满足实际工程应用的需求,所以亟需探求新的现代化模拟电路故障诊断技术。诸如神经网络、小波分析、模糊理论、遗传算法等人工智能技术的出现和发展,形成了这一领域新的研究方向。针对模拟电路故障诊断的模糊性和不确定性等问题,采用人工智能新技术的现代模拟电路故障诊断方法为常规方法所不能解决的各类问题带来了新的解决思路。本文系统地分析了几类传统的模拟电路故障诊断方法以及基于智能理论的现代模拟电路故障诊断方法。在此基础上,研究了将BP神经网络、小波分析、小波包分析等理论应用于模拟电路故障诊断中的方法,并引入改进粒子群优化算法优化神经网络的连接权值,达到加快网络收敛速度和提高诊断正确率的目的,进一步提升网络的性能。主要工作有:(1)阐述了模拟电路故障诊断课题的背景意义及当前国内外的发展状况,总结了传统的故障诊断技术以及近年来发展较快的智能故障诊断技术;(2)系统的研究了神经网络、小波分析、小波包分析等理论知识,探索了将这几种技术应用于模拟电路故障诊断中的方法,并选取待测电路进行了仿真分析,用实例证明了该方法的有效性与可行性;(3)对模拟电路故障诊断中最为关键的技术——特征向量的提取进行了详尽的分析与研究。应用小波多分辨分析和小波包分析等技术提取故障特征,并进一步探索将两种方法提取的故障特征向量融合成新的特征向量,作为故障诊断的故障集。通过对待测电路做实例研究的诊断结果表明了此方法的优异性;(4)将粒子群算法引入基于小波神经网络的模拟电路故障诊断中,利用改进粒子群优化算法对小波神经网络的连接权值进行适当的优化,加快了神经网络的收敛速度,并且训练后的网络具有较好的鲁棒性。
[Abstract]:The research of analog circuit fault diagnosis technology began in 1960. At present, it has made a lot of effective scientific research achievements at home and abroad, and gradually formed a perfect system theory, which occupies a very important position in the circuit theory. At the same time, with the rapid development of electronic industry, the integration of electrical equipment is becoming more and more high, increasingly modular and functional. However, because of the nonlinearity, continuity and component tolerance of analog circuits, the fault diagnosis of analog circuits is very difficult. Traditional analog circuit fault diagnosis method is difficult to meet the needs of practical engineering applications, so it is urgent to explore a new modern analog circuit fault diagnosis technology. The emergence and development of artificial intelligence technology such as neural network, wavelet analysis, fuzzy theory and genetic algorithm have formed a new research direction in this field. In view of the fuzziness and uncertainty of analog circuit fault diagnosis, the modern analog circuit fault diagnosis method using new artificial intelligence technology brings a new solution to all kinds of problems that can not be solved by conventional method. In this paper, several traditional analog circuit fault diagnosis methods and modern analog circuit fault diagnosis methods based on intelligent theory are systematically analyzed. On this basis, the methods of applying BP neural network, wavelet analysis and wavelet packet analysis to fault diagnosis of analog circuits are studied, and an improved particle swarm optimization algorithm is introduced to optimize the connection weights of neural networks. It can speed up the convergence of the network and improve the diagnostic accuracy, and further improve the performance of the network. The main works are as follows: (1) the background significance of analog circuit fault diagnosis and the current development situation at home and abroad are expounded. The traditional fault diagnosis technology and the intelligent fault diagnosis technology developed rapidly in recent years are summarized. (2) the theoretical knowledge of neural network, wavelet analysis, wavelet packet analysis and so on are studied systematically. The methods of applying these techniques to the fault diagnosis of analog circuits are explored, and the circuits to be tested are selected for simulation analysis. The effectiveness and feasibility of this method are proved by an example. (3) the extraction of eigenvector, which is the most important technique in analog circuit fault diagnosis, is analyzed and studied in detail. Wavelet multi-resolution analysis and wavelet packet analysis are used to extract fault features. Furthermore, the fault feature vectors extracted by the two methods are fused into new feature vectors, which can be used as fault sets for fault diagnosis. The results show that the method is excellent. (4) the particle swarm optimization algorithm is introduced into the fault diagnosis of analog circuits based on wavelet neural network. The improved particle swarm optimization algorithm is used to optimize the connection weights of wavelet neural networks, which accelerates the convergence speed of neural networks, and the trained neural networks have better robustness.
【学位授予单位】：湖南师范大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TP183;TN710

【参考文献】