节点文献
基于现代信号分析和神经网络的滚动轴承智能诊断技术研究
Study on Intelligent Diagnosis Technology of Rolling Bearing Based on Modern Signal Analysis and Neural Networks
【作者】 陆爽;
【导师】 张子达;
【作者基本信息】 吉林大学 , 机械设计及理论, 2004, 博士
【摘要】 滚动轴承是各种机电设备中的重要组成部件,其主要特点是其寿命的随机性较大,目前还无法准确预测其寿命的长短,并且它的工作状态直接影响到设备正常运行的质量。因而掌握滚动轴承运行的工作状态以及故障的形成和发展是目前机械故障诊断前沿领域中所研究的重要的课题之一。利用轴承及其组件的随机振动信号对其工作状态进行监测和诊断是目前机械故障监测与诊断研究中最常用的方法。在以机械设备振动信号为参量的机械运行状态监测与诊断中,当轴承出现故障时,所测取的设备中机械设备振动信号不但有轴承故障信号,而且还包含其它机械部件的振动信号,是典型的非平稳、非线性、非Gauss信号,并且在滚动轴承早期的故障产生阶段,其故障特征信息很微弱,很容易被机械中其它零部件运行中引起的振动信号和大量的随机机械振动信号所淹没。如何从所测取的振动信号中提取轴承微弱故障信号的特征信息并用来识别轴承的运行状态成了一个工程实践中面临的重要问题。本文基于现代信号分析中的时频分析、时序分析、主分量分析、高阶谱分析、小波分析、分形理论和人工神经网络等数学方法和工具对滚动轴承运行的工作状态及故障特征进行较为深入和系统的研究,给出了一些新的解决问题的方案,其主要的研究成果和相应的结论如下:1. 应用传统谱分析和现代谱分析技术对滚动轴承的振动信号的频谱进行了详细的分析。试验结果明确指明:无论用哪种频谱分析方法,由于轴承的初始故障状态产生的信号一般比较微弱,因而试图通过所测取的振动信号直接进行频谱分析从而得到轴承故障特征频率的企图是徒劳的。2. 应用包络分析方法对滚动轴承的振动信号进行了研究。试验结果表明:轴承如果不只是存在一个损伤点的情况下,试图对包络谱图给出合理的解释几乎是不可能的。而一般情况下轴承当使用到一定寿命时,或者由于某种原因引起损伤时,无论是外圈、内圈、滚动体还是保持架,只出现单个损伤的情况是很少见的。经常是多个同类型故障同时出现,而且不可能等间隔分布。尽管包络分析方法在轴承的故障特征频率诊断中可以<WP=161>作为一种诊断方法应用,但一般情况下应与其它方法共同使用,形成多信息综合判断准则,如果单独使用则故障诊断的准确率较低。3. 详细讨论了STFT和WVD的理论和应用中的问题,建立了仿真软件和分析软件,并利用Wigner-Ville分布建立轴承状态时频谱图对其正常和异常信号进行分类。明确指出:利用滚动轴承振动信号的时频特征谱图建立基准图形信息库是一种可行的滚动轴承故障诊断的新方法。4. 研究了高阶谱在非平稳、非Gauss、非线性信号特征提取中的良好的性质与作用。给出了详细的算法,建立了仿真软件和分析软件,并利用双谱建立轴承状态谱图对其正常和异常信号进行分类。明确指出:利用轴承信号的双谱特征谱图建立基准图形信息库是一种可行的滚动轴承故障诊断的新方法。5. 利用分形原理和复杂信号分类的模式,详细研究了分形维数—盒维数的原理,明确给出盒维数的计算方法和实现程序,并以盒维数建立了轴承故障特征维数。明确指出:相同工作状态下的滚动轴承振动信号具有相近的盒维数,不同故障模式下的盒维数具有不同的数值,存在明显的可分性。可以利用盒维数来有效地识别滚动轴承的故障状态。分形理论可以为滚动轴承的故障监测和诊断提供一种准确可靠的新的和实用方法。6. 提出了一种基于径向基函数神经网络的智能诊断系统。径向基核函数采用Gauss函数作为基本核函数,其方差因子参数为极坐标中的方向的函数,Gauss核函数的重要特点是保留集中在原点的信号自分量,抑制远离原点的交叉分量。仿真结果表明: RBF网络不但训练所需的时间短而且训练效果非常好,对正弦信号的预测误差在1%左右,其精度比BP网络高得多。7. 提出了一种基于AR模型特征参数建立径向基函数神经网络的智能诊断系统。给出了详细的算法,建立了仿真软件和分析软件。明确指出:只要AR模型的特征参数估计合理、准确,网络训练集规模选择合理, 该智能网络就能以90%以上的准确率完成从滚动轴承振动信号空间到滚动轴承故障状态空间的非线性映射,完成滚动轴承状态模式识别的任务。8. 提出了一种基于主矢量的主特征参数建立径向基函数神经网络的智能诊断系统。给出了详细的算法,建立了仿真软件和分析软件。明确指出:只要网络训练集规模选择合理,主特征值的特征参数估计合理、准确,该智能网络就能以90%以上的准确率完成从滚动轴承振动信号空间到滚动轴承故障状态空间的非线性映射。完成滚动轴承状态模式识别的任务。 <WP=162>9. 提出了一种基于小波分析的小波减噪方法,给出了具体的算法和分析软件。通过仿真和试验研究表明,小波减噪方法是处理滚动轴承微弱故障信号的有效工具,尤其对隐含其中的微弱振动信号的检测能力很强,是实现对复杂滚动轴承故障信号信噪分离的较理想的工具。将小波减噪方法应用于滚动轴承的早期故障诊断,显著地提高了信噪比,取得了很好的效果。同时指出:小波分析是机械微弱信号检测的新方法和有力的工具。10. 提出了一种基于小波能量特征向量和RBF网络相结合的小波神经网络智能诊断系统。给
【Abstract】 Rolling bearing is a important part of all kinds of mechanical and electrical equipments. The longevity can not be predicted for it’s great randomicity. The working state of rolling bearing determines whether the equipment can be used normally. Therefore , mastery of the working state and the form and development of the fault of rolling bearing is one of the important research in the advanced field of mechanical fault diagnosis. The random vibration signals of bearings and components , which are employed in monitoring and diagnosing their working state is the common used method in the study of mechanical fault monitoring and diagnosis. Monitoring and diagnosing of mechanical working state that based on the parameter of vibration signals, the signals captured from fault bearings are composed of not only itself but also other parts. The typical unstable, nonlinear and non-Gauss signals are so faint that usually covered with vibration signals generated by other parts and quantities of random signals. How to capture the fault features from faint signals of fault bearing efficiently and recognize working state correctly are the two linchpins of solving the problem in the field of engineering.In this dissertation, the working state and fault features of rolling bearing which is based on mathematic method and means of modern signal analysis such as time-frequency analysis, time-sequence analysis, principal components analysis, higher-order spectrum analysis, wavelet analysis, fractal theory and neural networks is deeply studied. Some new solution is presented. The main research achievement and conclusion are as follows: 1. Vibration signal spectrum of rolling bearing is analysed detailedly with traditional and modern spectrum technology. The experimental result shows that the initial fault signal is very faint. As a result, spectrum analyzing from capturing signals and thus obtaining characteristic frequency of rolling bearing is in vain, whichever spectrum analyzing method is used.2. Vibration signal of rolling bearing which is based on envelopment analysis <WP=164>methods is studied. The experiment results verify that it is difficult to explain the envelopment chart reasonably if the rolling bearing possesses more than one damage point. Actually , seldom does inner and outer rings, roller or retainer appear single damage when attains its longevity or damaged by other reasons. The properties of damage are some fault type of the same kind and non-equal distances distribution. Although envelopment analysis is available in diagnosing fault features of frequency , it should be used with other efficient methods to form a multi-information integrated recognizing system,otherwise,the accuracy rate would be too low.3. STFT and WVD theory and their application are discussed detailedly. Furthermore, simulation and analysis software are presented. Normal and abnormal signals are classified by time-frequency spectrum that in the light of Wigner-Ville distribution. It shows clearly that a new method of reference graphic database which established by time-frequency spectrum is feasible in diagnosing fault of
【Key words】 Rolling bearing; Fault diagnosis; Time-Frequency analysis; Higher-Order spectrum; Wavelet transform; Wavelet packet; Principal components analysis; Time-Sequence analysis; Fractal theory; Power spectral analysis; Neural Networks; Intelligent diagnosis;