【作者】 王冬云；

【导师】 张文志；

【作者基本信息】 燕山大学 ， 机械设计及理论， 2012， 博士

【摘要】 旋转机械故障诊断技术在企业中的应用能够及早发现设备故障、防止生产线停工、避免重大事故。转子-轴承系统是各种旋转机械中的关键设备，它的运行状态是否正常往往直接影响到整台机器的性能。转子-轴承系统故障诊断研究，对现代工业发展具有重要的意义。首先，针对故障轴承振动信号能量集中与调制的特点，提出了一种基于小波包能量与Hilbert变换的滚动轴承故障诊断方法。使用小波包变换对振动信号进行分解、重构、能量计算，并应用Hilbert变换对能量集中频段的重构信号进行解调和频谱分析，提取故障特征频率。同时采用故障特征参数自动搜索方法解决诊断过程中特征参数依靠人工计算的问题。经过对实际的滚动轴承实验数据的处理和分析，表明该诊断方法能够准确、快速地识别滚动轴承表面损伤的故障模式。其次，应用基于尺度变换的谐波小波包技术提取转子系统故障特征参数。采用谐波小波包对转子故障信号进行能量提取时，由于振动信号在各节点上的旋转倍频分布与转速相关，导致不同转速下旋转倍频与振动能量的节点分布没有统一的物理意义。为消除谐波小波包节点中的倍频分布在转速变化时表现出的随机性，运用尺度变换对原始振动信号进行重采样，再应用谐波小波包分解技术将重采样后的信号分解到给定层上，从而获得信号的频率特征，计算各频段的谐波小波包系数能量值作为转子故障特征参数。最后经过对实际的转子油膜涡动实验数据的处理和分析，表明该方法能够智能地提取不同转速下转子系统的故障特征，为故障诊断研究提供准确的数据支持。然后，设计了基于支持向量机与模糊聚类(SVM+FCM)方法的转子故障诊断方法。应用模糊c均值聚类方法对训练样本进行预选取，在保证分类精度的前提下减小了计算量，节省了训练时间；采用基于交叉验证的网格搜索法对支持向量机模型进行参数寻优，对C -SVM算法中的参数C与径向基核函数的尺度参数γ实施并行寻优，确保了诊断模型的最优性与参数寻优的高效性；在ZT-3转子实验平台进行转子不平衡、转子不对中、转子动静碰摩、转子油膜涡动等转子故障模拟，应用所测转子故障振动数据对基于尺度变换的转子故障特征参数提取方法与基于SVM+FCM的转子故障诊断方法的有效性进行了验证。最后，应用LabVIEW软件与MATLAB软件相结合的方法开发了转子-轴承故障诊断软件平台。该平台应用MATLAB Script节点对接LabVIEW与MATLAB软件，实现了友好的用户界面与强大的工程计算能力的统一。开发的软件平台具备时域分析、频谱分析、故障诊断、诊断结果保存、查询历史数据等功能，能够为转子-轴承系统提供实时、准确的诊断。更多还原

【Abstract】 The application of fault diagnosis technic with rotating machine is encouraged tomonitor facility status, ensure production line working regularly, and prevent from majoraccident. Rotor-bearing system is one of the most widely used elements in rotatorymachines. The running status of the rotor-bearing system is important to the performanceof the rotatory machinery. The fault diagnosis of rotor-bearing system has greatsignificance for development of contemporary industry.Firstly, a fault diagnosis study based on wavelet packet energy and Hilbert transformis put forward. That is relative to the modulation and energy concentration of faulted ballbearing vibration signal. The vibration signal of ball bearing is decomposed andreconstructed using wavelet packet transform. And energy calculation of every frequencyband is also done. Selects the frequency band with maximal energy. Then, analyses thesignal of the frequency band applying Hilbert transform. Finally, extracts the characteristicfrequency of fault signal. At present, the computation of fault features is accomplishedartificially. In this paper, a new method which can select fault features automatically ispresented. Through processing and analyzing the practical ball bearing experiment data, itis shown that the fault diagnosis study can diagnose different running states of ballbearings due to surface damage timely and exactly.Secondly, a method based on harmonic wavelet packet is proposed to extract featuresof rotor faults. Due to the distribution of rotating frequency doubling in every node relyingon rotor rotation rate, so there is no unified physical meaning of them in different rotatingrates. To eliminate the effect of rotor rotating rate to doubling frequency features, applingthe scale transform theory resamples the original signal firstly. And then decomposes thesignals with harmonic wavelet packet and computers energy of each node. Throughprocessing and analyzing the practical oil-whirl experiment data, it is shown that the studycan extract rotor faults features in diverse rotation rates intelligently. This providesaccurate data surpporting for fault diagnosis.Thirdly, put forward a method of rotor fault disgnosis based on support vector machine and fuzzy c mean clustering. Sample datas to train SVM are pre-selected withthe fuzzy c means clustering. It is useful to reduce time consuming in computation, andensure the classification accuracy. The grid search method based on cross–validation ischosen to determine model parameters. The model is optimal and efficent due tocaculating parameter C andγmeanwhile. Unbalance experiment of rotor,misalignment experiment of rotor, rubbing experiment of rotor, and whirling experimentof rotor are carried out on ZT-3 exprimental instrument. Through analysing the vibrationsignal of rotor fault, it is proved that the feature extraction method based on scaletransform and the fault diagnosis method based on SVM+FCM are efficient.Finally, fault diagnosis platform of rotor-bearing system based on LabVIEW andMATLAB is developed. MATLAB script is applied in the fault diagnosis platform. Theuser interface is frendly. And the engineering caculation capbility of the platform is strong.The platform provides function of time domain analysis, frequency domain analysis,intelligent fault diagnosis, result saving, and historical data inquirying. The fault diagnosisplatform is real-time and accurate.更多还原