【作者】 于江林；

【导师】 戴光；

【作者基本信息】 大庆石油学院 ， 化工过程机械， 2009， 博士

【摘要】 滚动轴承被广泛用于石化、冶金和铁路等行业的重要设备上,是旋转设备和交通工具中易损坏的机械零件之一,约有30%的故障是由于滚动轴承的损坏造成的,并产生巨大的经济损失。因此,对滚动轴承工作状况进行实时监测和故障诊断的研究越来越受到人们的重视。目前,国内外将振动和声学方法用于滚动轴承故障诊断研究多数是接触式的,即所用传感器是与被测滚动轴承座或相连结构的表面相互接触,这种检测方式对低速和移动等特殊条件下运行的滚动轴承有一定的局限性。为此,本文运用声学理论和现代信号分析方法深入研究滚动轴承故障的非接触多传感器声学诊断方法,这对滚动轴承故障的早期诊断和预防事故的发生有重要的理论意义和实用价值。文中分析了滚动轴承的常见故障类型及形成机理,阐述了滚动轴承声发射源产生的机理,得出滚动轴承的磨损故障和表面损伤故障的形成和扩展过程均会产生声发射,故障部位与滚动轴承构件在运动过程中的相互摩擦和碰撞,也是产生声发射的重要来源。同时,对于非接触声学检测,由于声波在空气中以纵波传播,且主要受空气吸收衰减作用的影响,空气吸收系数与频率的平方成正比,所以,低频成分的声波传播的距离较远,而高频成分衰减很快,这些特性有利于非接触式滚动轴承检测的诊断和分析。通过建立滚动轴承非接触声发射检测实验系统,完成了无故障滚动轴承以及三种故障类型滚动轴承的非接触式声发射检测实验。通过接触式与非接触式滚动轴承检测信号的对比实验表明,由于声波在空气中的传播衰减,其幅度与撞击数在数量上后者均小于前者,但二者的撞击数都与不同故障类型的理论碰撞频率值相等,这为非接触声发射检测的故障模式识别提供依据。采用小波分析方法,对不同故障模式滚动轴承非接触声发射信号进行分解,依照各频段能谱分布,得出滚动轴承非接触声发射信号的特征频段,该频段的能谱系数反映了声发射源特征,根据这一特性,可实现对检测声发射信号的去噪。建立基于小波和EMD的局部Hilbert边际谱诊断方法识别滚动轴承故障频率,提取无故障滚动轴承,外圈故障滚动轴承和滚子故障滚动轴承的边际谱峰值频率,混合故障滚动轴承边际谱涵盖前三种峰值频率。上述方法,实现了对滚动轴承非接触声发射信号的去噪和有效性识别,为后续的研究工作奠定了基础。建立滚动轴承非接触多传感器声发射检测实验和分析方法。为了获取完整的周期性滚动轴承故障声发射信号,文中利用相关性分析方法对多个传感器得到的不同声发射信号进行分析,确定多传感器阵列中各个信号间的关系——相互关联或者是相互独立。根据信号触发时间与滚动轴承不同故障类型的理论碰撞频率间的关系,去除重叠和无效部分,并以此建立非接触多传感器声发射信号的周期性重构方法。对二个和三个传感器移动非接触检测实验数据进行周期性声信号重构,通过与单传感器重构的周期性声信号进行对比分析表明,建立的多传感器周期性声信号重构方法可获得较为完整的滚动轴承周期性声信号。通过上述研究工作,建立了声传感器阵列中相关传感器间重叠信息辨识与提取方法,实现了基于多传感器的滚动轴承故障周期性声信号重构,获取了移动中滚动轴承故障的完整周期性信息,为进一步对滚动轴承不同类型故障诊断提供了依据。滚动轴承故障类型识别方法的研究是本文的又一重要内容。本文提出基于滚动轴承不同故障周期性特性的声发射撞击数故障模式识别方法;该方法可实现单一故障滚动轴承的模式识别。同时,建立基于小波包和BP神经网络的滚动轴承故障模式识别方法,该方法有效地突出故障特征,提高了故障诊断的有效性和准确性。对于滚动轴承故障的多传感器非接触声发射检测,依据重构的完整周期性声发射信号波形图,利用模糊识别理论,结合不同滚动轴承故障的声发射信号特性,建立滚动轴承多传感器声发射信号重构波形图模糊识别方法。对不同故障模式滚动轴承声发射信号波形图进行分析,提取波形图上的5个不同特征要素——幅度电压值、上升沿宽度、波形宽度、捌点和最大余波宽度,建立了模糊识别算法。通过对已知故障类型进行训练,得到不同故障类型特征矩阵。利用该方法对滚动轴承不同故障的非接触多传感器重构周期性信号进行识别,验证了该方法的可行性,实现了对滚动轴承不同类型故障的早期诊断与预报。更多还原

【Abstract】 Bearings are generally used in important equipments of petrochemical, metallurgy and railway, etc. And as easily broken parts, almost 30% rotating equipments and vehicles fault accidents are originated from the bearings, which result in large lost. Thus, researchers pay more and more attention to the research on monitoring state of bearings on line and diagnose faults. At present, most methods combining vibration with acoustic testing on rolling bearing fault diagnosing are contacting detection; namely, the sensors used for testing are contacted with bearing seat or connected structure surface, which is not applicable for the testing of bearings under low rotating speed or moving bearings. The paper focus on the research of non-contacting rolling bearing fault diagnosing with acoustic theories and modern signals analysis method. All of these are valuable on theory and application for early diagnosis of rolling bearing fault and preventing failures.The common types of faults and their mechanisms in rolling bearings are analyzed in the paper. Especially, the mechanism that acoustic sources are generated is discussed, the result shows Acoustic Emission(AE) signals mainly occur in the formation and spread stage of wear and surface damage, also, large amount of signals can be found when there are friction and collisions at fault site. To acoustic waves, they propagate in the form of longitudinal wave in air. Their propagation is mainly influenced by the absorption attenuation, and absorption factor is proportional to the square of the frequency. Low frequency acoustic waves have relatively long propagation distance, but high frequency components decay quickly, which is favorable for the diagnosis and analysis of non-contacting rolling bearings with acoustic method.With established testing system for rolling bearing fault detecting, non-contacting acoustic testing experiments are carried out on non-fault bearings and bearings with three different faults. Through the comparative analysis between contacting testing and non-contacting testing, the result shows that for the attenuation when acoustic waves propagate in air, the amplitude and magnitude of signals from non-contacting testing are smaller than those from contacting testing, but the number of hits of them is closer to the theoretical value of components collision frequency under different fault. It provides the basis for fault pattern recognition in non-contacting bearing AE detection.After wavelet decomposition of AE signals from rolling bearings under different fault modes, typical bands are obtained according to spectrum distribution of every band, whose spectrum factors representing the acoustic sources characteristics. Based on this characteristic, the denoising of AE signals can be achieved. Local Hilbert marginal spectrum diagnosis method based on wavelet and EMD is advanced and can be used to recognize fault frequency, marginal spectrum peak frequency of non-fault bearing, bearing with faults in outer ring and bearing with faults in roller. The marginal spectrum frequencies of mixed fault bearing include the above-mentioned peak frequencies. All the above methods can be used to diagnose and recognize signals from non-contacting AE testing on rolling bearings, which lays the foundation for follow-up study.AE experiment and analysis method is established for multi-sensors non-contacting rolling bearing fault testing. In order to obtain complete and cyclic fault acoustic signals, large amount of signals from different sensors are analyzed with the method of correlation analysis, and the relationship is determined between signals from the sensors array, that is, they are corresponding or independent. According to the relationship between hit time and theoretical striking frequency of different faults, duplicated and ineffective parts are removed. After that, cyclic reconstructing method is established for multi-sensors acoustic signals. After reconstructing cyclic acoustic signals from two and three sensors moving non-contacting testing, and comparing with the reconstructing signal form single sensor inspection, the result shows that the established reconstructing method can be used to obtain complete periodic acoustic signals. On the basis of these, overlap information identification and extraction method is established for relevant sensors in acoustic arrays. Cyclic acoustic signals reconstruction is achieved for multi-sensors rolling bearing inspection, and integral cyclic information of moving bearings with faults is obtained. All of them provide basis for the denoising of bearing with different faults.The method of fault pattern recognition is one of the main contents in the paper. AE signals from non-fault rolling bearing are continuous, whose energy is small, and there are no obvious peak in the signals. While, bursting signals will be get during the detection of bearings containing faults, and the magnitude of peak energy in spectrum diagram is large. On the basis of this, the state whether there are faults in bearing can be identified. To the bearings with single fault, combining with the number of hits based on theoretical characteristic frequency, the type of the fault can be achieved. Rolling bearing fault diagnosis method can effectively recognize fault characteristics based on wavelet packet and BP neural network, so as to enhance the diagnosis efficiency and accuracy.Combining with AE signals characteristics of different bearing faults, pattern recognition method of multi-sensors signal reconstructing waveform is advanced based on fuzzy recognition theory. After the analysis of different waveforms belong to different faults, fuzzy recognition algorithm is achieved according to five different recognition elements in the waveform diagram, i.e., voltage of peak amplitude, risetime range width, waveform width, inflection point and largest remaining wave width. Through the exercise of known faults, fault signals from rolling bearings can be recognized properly.更多还原