【作者】 彭畅；

【导师】 柏林；

【作者基本信息】 重庆大学 ， 机械电子工程， 2014， 博士

【摘要】 轴承作为旋转机械中广泛使用的关键零部件之一，其运行状态直接关系整台机械设备的工作性能，开展基于振动信号分析的轴承状态监测与故障诊断的相关研究并及时准确地识别故障萌发与演变，对确保设备平稳运行、减少甚至避免重大安全事故具有相当重要的意义。本课题以滚动轴承及滑动轴承故障振动信号为研究对象，针对滚动轴承故障信号易受噪声干扰的影响拓展并丰富了峭度图理论在振动信号降噪以及故障特征提取中的应用，并基于定向循环平稳分析理论研究了滑动轴承在油膜失稳状态下的故障信号特征，最后开发了虚拟式旋转机械轴承测试诊断系统实现了理论创新成果在工程实践中的应用。本文的具体研究内容介绍如下：本文首先介绍了峭度图理论中涉及的峭度统计理论、谱峭度系数等数学基础，并详细阐述了传统峭度图算法以及基于COT的阶比峭度图算法在故障信号降噪和最优解调频带参数确定方面的优势。针对传统四阶矩累积量谱峭度系数易受信号奇异点影响而不能真实估计信号峰态程度水平的问题，定义了Moors谱峭度、Hogg谱峭度以及Crow-Siddiqui谱峭度等三种鲁棒性谱峭度系数，并提出了能消除信号奇异点干扰的鲁棒性峭度图算法。滚动轴承仿真及实测故障信号分析结果验证了鲁棒性峭度图与传统峭度图相比，能显著提高滤波效果并增强原始信号中的故障瞬态冲击特征。传统的基于STFT、FIR滤波器以及第一代小波包变换的峭度图算法中存在时频分辨率取舍、运算效率以及小波函数库有限等缺点，本文提出了基于尺度自适应冗余提升小波包变换的峭度图算法，通过基于原始信号本身构造更新器和预测器来实现信号分解与重构的自适应功能，有效地提高了峭度图算法的自适应特性以及滤波后故障信号的准确性。随后利用时间尺度分解进一步增强了峭度图滤波后信号的瞬态冲击故障特征，并将基于固有时间尺度分解的Hilbert谱作为包络解调分析的有益补充，通过与Hilbert-Huang变换的对比验证了其在表征故障信号时频能量信息分布中的优越性。旋转机械振动信号往往呈现循环平稳特性，本文将机械故障诊断领域中用于分析单通道实信号的循环平稳统计量理论进行了拓展，定义了能用于分析双通道融合复信号的低阶定向循环平稳统计量并提取出了滑动轴承油膜失稳故障信号中周期性时变特征以及转子振动状态信息等。通过对比分析揭示了传统全谱分析以及定向Wigner分布与定向循环统计量之间的内在联系，即全谱分析实质上就是一阶定向循环平稳统计量（定向循环均值），而定向Wigner分布实质上属于二阶定向循环统计量（定向循环谱相关函数）。此外试验结果分别对比分析了加速度振动信号与位移振动信号中所包含系统振动状态信息。根据轴承振动监测需求分析，基于NI Labview平台开发了轴承振动信号测试、分析与特征提取于一体的虚拟式旋转机械轴承测试系统，为取得的理论创新成果在工程实践中的应用提供了强有力的支持。文章最后对全文工作及主要创新点进行了总结，并展望了后续的研究方向。更多还原

【Abstract】 Bearings are one of the most widely used elements in the rotating machinery andtheir performance of great necessity to the operation of the mechanical equipment. Inorder to timely identify the deterioration in the condition of the bearings, it is significantto carry out the vibration based condition monitoring and fault diagnosis. In this thesis,the kurtogram based method is firstly developed and improved for the de-noising andfault characterization of the vibration signals buried in strong noise for the rollingelement bearings. Subsequently, the directional cyclostationarity analysis basedalgorithm is proposed to reveal significant features of the vibration signature for thejournal bearings with oil film instability faults. Finally, the integrated and virtualmeasurement and diagnosis system for the bearings of rotary machines is developed tofacilitate the practical application of the achievement in this thesis. The main researchesare listed as follows:Firstly, the mathematical basis including the kurtosis statistics and spectral kurtosiscoefficients of the kurtogram is reviewed. Subsequently, the specific algorithm of theconventional kurtogram as well as the COT based order-kurtogram is introduced. It isstated that the kurtogram is able to de-noise and identify the optimal demodulation bandfor the following analysis of the vibration signal of the rolling element bearings. Sincethe traditional spectral kurtosis coefficient is easy to be influenced by the singularvalues in the original signal, the robust spectral kurtosis coefficients including theMoors spectral kurtosis, Hogg spectral kurtosis and the Crow-Siddiqui spectral kurtosisare defined. The robust kurtogram is also proposed to eliminate the interference of thesingular values. The simulation test and practical application has verified that theproposed robust kurtogram is able to improve the filtering effect and enhance theimpulsiveness of the fault signal.Since the filtered signals will be inputted to the distribution of the spectral kurtosis,it is necessary to select the optimal filter in order to improve the ability of the kurtogram.The STFT suffers from the trade-off between the time resolution and frequencyresolution according to the Heisenberg uncertainty principle. The FIR filter-bank suffersfrom the low operation efficiency and WPT from the limit of the wavelet functionlibrary. Therefore, an improved kurtogram is proposed based on the scale-adaptiveredundant lifting wavelet packet transform which can construct the updater and predictor according to the signal itself as well as to adaptively decompose the originalsignal. The results have proved that the proposed method is able to improve theadaptivity of the conventional kurtogram and ensure the accuracy of the signal filteredby the kurtogram. Moreover, the intrisic time-scale decomposition based Hilbertspectrum has been proposed to supplement the envelope analysis. Compared with theHilbert-Huang transform, practical test has verified the superiority of the ITD basedHilbert spectrum when characterizing the distribution of the time-frequency energy ofthe fault signal.Cyclostationarity defined to represent the correlation features of periodicphenomenon has been widely used to characterizing the fault vibration signal obtainedfrom the rotating machinery. Based on the conventional cyclic statistics, the directionalcyclic parameters including the directional cyclic mean, directional cyclic correlation,and directional spectral correlation density are defined to represent the complex-valuesignal integrated from double channel. It is found that the directional cyclic statistics areable to extract the periodically varying characteristics as well as the information relatedto the vibrating condition of the journal bearing supported rotor under oil film instabilityfault. In addition, the relationship between the conventional full spectrum, thedirectional Wigner distribution and the directional cyclic statistics is revealed. The fullspectrum can be interpreted as the first order cyclic statistics, namely the cyclic mean.The directional Wigner distribution belongs to the second order cyclic parameters(directional cyclic correlation function). Moreover, the acceleration signal basedcondition monitoring is compared with the displacement signal based conditionmonitoring when describing the system vibration.According to the requirements analysis of the vibration based condition monitoringon bearings, the NI Labview based virtual measurement and diagnosis system for thebearings of rotary machines is developed. The system which is able to conduct dataacquisition, signal analysis and fault diagnosis has facilitated the practical application ofthe achievement in this thesis.更多还原