【作者】 周强；

【导师】 李铁骊；

【作者基本信息】 大连理工大学 ， 轮机工程， 2006， 硕士

【摘要】 本文主要研究的内容为故障诊断的新方法探讨,即结合模糊神经网络和遗传算法的故障诊断方法,具体包括模糊理论、神经网络和遗传算法及他们的在故障诊断中的应用。 模糊理论包含方面较多,这里主要应用了隶属度和模糊推理。应用隶属度概念可以把设备故障的特征参数分级模糊化,再应用模糊推理可以直接进行设备的故障诊断。该诊断方法是应用MATLAB提供的模糊逻辑工具箱实现的。 神经网络的许多特性表明它适用于设备的故障诊断应用,BP网络是应用最广泛、最成熟的网络之一。这里以BP网络为应用模型,把模糊化后的故障特征参数作为网络的输入样本,对样本数据训练、仿真,应用MATLAB提供的神经网络工具箱函数实现了模糊神经网络对设备故障的诊断,结果证明了该诊断模型的有效性,但也暴露出BP算法易陷入局部极值的弱点。 最后应用遗传算法,以神经网络的权值和阈值为变量编码,随机生成初始种群,经过选择、交叉、变异等算子的操作,以适应度计算为依据,使种群不断进化直至满足要求。遗传算法的全局搜索能力弥补了BP算法的不足,使BP网络能跳出局部极值,而且缩短了网络的训练时间。这里的遗传算法优化是应用设菲尔德大学开发的遗传算法工具箱实现的。 本文以某制冷系统故障诊断为例,应用遗传算法优化神经网络权重,应用模糊理论把故障特征参数分级模糊化,再结合神经网络进行故障诊断,结果证明了基于模糊神经网络和遗传算法的故障诊断方法是有效的,具有一定的学术和工程应用价值。更多还原

【Abstract】 A new method of fault diagnosis is the main content of this paper. Concretely, Fuzzy Theory, Artificial Neural Network (ANN) and Genetic Algorithm (GA) and their application in equipment fault diagnosis are discussed in detail.In Fuzzy Theory, the characteristic parameters of equipment fault can be fuzzied by degree of membership, and then the fault diagnosis can be processed by the fuzzy inference. This fuzzy method is realized by the Fuzzy Inference System Toolbox of MATLAB.The characteristics of ANN indicate that it is suitable for the fault diagnosis. The popular BP network is adopted as the application model. Input data which are from fuzzy characteristic parameters are trained and simulated. The fault diagnosis of Fuzzy ANN is realized by the Neural Network Toolbox of MATLAB. The result is proved valid, but also proved easy to be trapped in the local extremum.In GA, with the weight and threshold of ANN encoded, then the initial population randomized, and the selection, crossover and mutation operated, the population can be optimized till requirement satisfied. The global search ability of GA covers the shortage of BP network, and save the training time. GA is realized by the Genetic Algorithm Toolbox developed by University of Sheffield, UK.A refrigeration system is cited here. The chosen fault characteristic parameters are fuzzied, and the weight and threshold of ANN are optimized, then the diagnosis is accomplished by BP network. The method of fault diagnosis based on fuzzy neural network and genetic algorithm is proved valid, and valuable in the academic and engineering application.更多还原