【作者】 杨娟；

【导师】 赵荣珍；

【作者基本信息】 兰州理工大学 ， 机械制造及其自动化， 2010， 硕士

【摘要】 故障诊断技术主要包括三部分的研究内容,即信号处理、特征提取、故障辨识。特征提取,是对系统的动态信号预处理后得到的信息进行分析,提取与系统状态有关的数据,再分析这些数据,提取其中与系统状态相关性较大的敏感特征,是故障诊断技术的关键,理论研究的热点之一。特征提取的完善和正确与否,直接影响到诊断是否成功和诊断结果的准确性。常用的信号特征提取方法存在以下两点不足：(1)难以准确描述系统非线性及信号的非平稳信号特征；(2)难以解决信号特征的定量评价问题。本文采用信息熵分析方法,对机械信号特征提取和定量评价方法进行研究。针对转子实验台的数据模型,运用信息熵方法对故障类别、程度分别给出客观的量化评价方法。主要工作内容及获得的研究结论如下：1)在认真查阅国内外大量文献的基础上,分析信息熵测度在故障诊断中的研究现状和存在的问题。对信息熵的性质和时域的奇异谱熵、频域的功率谱熵和时-频域的小波能谱熵进行了较为系统的研究和探讨。2)以轴承-转子系统动力学模型为基础,研究了它的常见的故障的机理。设计了有限冲激响应低通滤波器,对采集的信号进行滤波,为特征提取提供原始数据。同时把基于AR模型最大熵法的功率谱算法应用到故障类别的诊断中,分辨率比传统的方法高,谱线平滑,同时抗噪的能力很强,很适合轴承-转子的数据处理。3)研究了盲信号处理的Fast-ICA算法和AMUSE算法,通过实例仿真分析可得,由于Fast-ICA算法的分离性能指标较小和自身的优点,所以选择Fast-ICA算法分离后的数据作为奇异谱熵的样本数据。提出了基于Fast-ICA的奇异谱熵算法。通过对实验信号的分析表明,该算法能有效评价转子振动状态的特征指标。4)基于信息融合思想,对描述振动信号能量的三种信息熵测度,即时域的奇异谱熵、频域的功率谱熵、时频域的小波能谱熵,在根据广义集合概念对上述三种信息熵测度进行特征级的信息融合基础上,提出了一种基于时空(场)广义信息熵集合的轴承-转子系统故障辨识方法。分析表明,该方法拥有在三维特征空间图形化描述故障状态域、使不同故障类别间显示出显著差别的性能,它对提高辨识故障的准确率具有参考价值。信息熵理论在故障诊断领域带来了研究热潮,在特征的量化提取方面值得进一步深入地研究。更多还原

【Abstract】 Fault diagnosis consists of three parts of the content, that is, signal processing, feature extraction, fault identification.Feature extraction, is a dynamic signal preprocessing system analysis of the information obtained by extraction with the system state of the data, then analyze these data, extract the system state-dependent with greater sensitivity Feature, is the key of fault diagnosis and one of the focuses of theoretical research.Whether feather is extracted completely and correctly or not, directly affects the success of diagnosis and diagnostic accuracy. Signal feature extraction method commonly used the following two shortcomings:(1)Difficult to accurately describe the nonlinear and non stationary signal characteristics;(2)Difficult to solve the signal characteristics of the quantitative evaluation.In this paper, I use entropy analysis method to study the mechanical signal feature extraction and quantitative evaluation methods.Rotor test bed for the data model,using information entropy for the fault type, degree are given an objective quantitative evaluation.Main contents and conclusions of the study were as follows:1))Based on searching a lot of domestic and foreign literature current situation and existing problems of information entropy measure in fault diagnosis are analyzed.The nature of information entropy and spectral entropy of the singular time domain, frequency domain power spectral entropy and time-frequency domain of wavelet energy spectrum entropy are studied and disscussed systematically.2))Its common failure mechanism is studied based on bearing-Rotor System Dynamics model.A finite impulse response low-pass filter is designed,which filters the collected signals,to provide raw data for the feature extraction.While the AR model based on maximum entropy method of power spectrum algorithm is applied to fault diagnosis categories,which has higher resolution than traditional methods,smooth spectrum,strong ability of noise immunity,and fits bearing-Rotor data processing very much.3)Fast-ICA and AMUSE algorithms which are dealed with the blind signal are studied.Based on the simulation analysis,the data separated from Fast-ICA algorithm is selected as singular spectrum entropy of the sample data because of smaller separation performance Its own merits.Singular spectrum entropy algorithm based on Fast-ICA algorithm is proposed.Through the analysis of experimental signals,the algorithm can effectively evaluate the state of rotor vibration characteristics of indicators.4) In the view of information fusion,application methods of three kinds of information entropy measures,which describe vibration signal energy, such as the singular spectrum entropy in the time domain, power spectrum entropy in the frequency domain and wavelet energy spectrum entropy in the time-frequency domain, In accordance with Generalized sets concept by the feature information fusion about the above-mentioned three kinds of information entropy measures,a new fault identification method in the bearing-rotor system which base on Generalized sets concept of time spatial of the information entropy is proposed.The analysis shows that the method has the Function of three-dimensional feature space graphical representation of fault state domain, demonstrates significant differences between different fault types,and enhances the accuracy of identification fault with a reference value.Entropy theory in the field of fault diagnosis of fever brought in the quantitative extraction of features worthy of further study.更多还原