基于流形学习的滚动轴承故障诊断若干方法研究

发布时间：2018-05-24 21:34

本文选题：滚动轴承 + 故障诊断　；参考：《大连理工大学》2013年博士论文

【摘要】：滚动轴承是机械设备中使用量最多的关键零部件,保证滚动轴承正常运行是设备维护工作的重要内容。但滚动轴承工作状态复杂,转速变化大,承载方式多样,运动形式多变,这些都会对滚动轴承故障诊断产生不利的影响,从而降低各种传统诊断方法的效能。为此,本文以滚动轴承故障振动信号为研究对象,将流形学习方法与现代信号处理理论结合,对滚动轴承故障诊断过程中遇到的降噪、特征提取、故障源分离、性能监测问题进行研究。论文的主要研究内容及结论如下： 1.论述了开展滚动轴承故障诊断的意义,分析了滚动轴承故障诊断技术不同发展阶段的特点。对滚动轴承故障诊断中遇到的降噪、特征提取、故障源分离、性能监测问题的研究现状进行了综述。在总结滚动轴承故障诊断技术发展趋势的基础上,介绍了非线性流形学习理论及其在故障诊断中的应用。 2.针对滚动轴承故障信号降噪问题,提出了一种基于最大方差展开(MVU)算法的对偶树复小波(DTCWT)降噪方法。利用MVU提取DTCWT细节信号空间的信号子空间,去除噪声子空间实现降噪。DTCWT具备平移不变性和完全重构性,能克服常规小波变换平移敏感性和非完全重构的缺陷。MVU流形算法能有效提取高维数据空间的非线性结构,克服线性结构的不足。结合二者优势的DTCWT_MVU降噪方法,比传统降噪方法具有更高的信噪比,能更好地提取故障信号的非线性冲击成分,减少降噪后信号波形的失真。仿真和工程实际信号验证了该方法的有效性。 3.对于滚动轴承故障特征提取问题,提出了一种基于张量流形学习的时频故障特征提取方法。在HHT时频特征的基础上,利用张量流形学习方法提取信号的非线性张量流形时频特征,定义了时频特征参数,将张量流形时频特征参数与概率神经网络相结合,准确实现了轴承故障样本分类。张量流形学习能有效提取高维时频特征组合的内蕴非线性特征,与HHT时频特征参数相比,张量流形时频特征参数能减少特征信息的冗余,更有效地区分不同类型故障样本,降低神经网络的迭代次数,提高故障分类的准确性。 4.对于滚动轴承故障源分离问题,提出了一种基于流形学习的滚动轴承故障源盲分离方法。利用EMD分解构造了多通道测试信号,估计测试信号的信源数,建立最优测试信号的选择标准,综合利用峭度、稀疏度、互信息标准选择最优测试信号,通过提取最优测试信号的KPCA流形成分作为ICA算法的输入,有效分离出故障源。该方法解决了欠定盲分离过程中最优测试信号的选取问题,利用流形学习增强了ICA的分离能力,使其能从故障信息微弱的单通道信号中分离出冲击特征明显的故障源。 5.针对滚动轴承性能退化监测问题,提出了一种基于流形学习和模糊聚类的性能监测方法。利用小波包分解确定监测信号的敏感频带,在此基础上提取信号的低维流形特征作为模糊聚类的数据样本,以样本的隶属度值作为性能指标,监测轴承性能退化规律。与基于单特征及线形多特征的监测方法相比,该方法能有效体现滚动轴承全寿命性能退化周期的四个阶段,反映滚动轴承各部件性能退化的统一规律,提前预知轴承早期故障。 6.使用LabVIEW和MATLAB混合编程的方式开发了基于流形学习的滚动轴承故障分析诊断系统。介绍了系统开发的软硬件环境和结构方案,通过实例演示了系统的基本功能,验证了系统的有效性。
[Abstract]:Rolling bearing is the most important component in mechanical equipment. It is an important part of the maintenance work to ensure the normal running of the rolling bearing. But the working state of the rolling bearing is complex, the rotational speed is varied, the bearing mode is varied, and the form of movement is changeable. All these will have an adverse effect on the rolling bearing fault diagnosis, thus reducing a variety of factors. In this paper, this paper takes the rolling bearing fault vibration signal as the research object, combines the manifold learning method with the modern signal processing theory, and studies the noise reduction, feature extraction, fault source separation and performance monitoring problems encountered in the fault diagnosis of rolling bearings. The main contents and conclusions of this paper are as follows :
1. the significance of rolling bearing fault diagnosis is discussed, and the characteristics of different development stages of rolling bearing fault diagnosis technology are analyzed. The research status of noise reduction, feature extraction, fault source separation and performance monitoring problems encountered in rolling bearing fault diagnosis are summarized. The basis for summarizing the development trend of rolling bearing fault diagnosis technology is summarized. The nonlinear manifold learning theory and its application in fault diagnosis are introduced.
2. a dual tree complex wavelet (DTCWT) denoising method based on the maximum variance expansion (MVU) algorithm is proposed to reduce the noise of rolling bearings fault signal. Using MVU to extract the signal subspace of the DTCWT detail signal space, the noise subspace is removed to realize the flat shift invariance and complete reconstruction of the noise reduction.DTCWT, and the conventional wavelet transform can be overcome. The defect.MVU manifold algorithm can effectively extract the nonlinear structure of the high dimensional data space and overcome the shortage of linear structure. The DTCWT_MVU denoising method combining the two advantages has a higher signal to noise ratio than the traditional noise reduction method, and can better extract the nonlinear impact component of the obstacle signal and reduce the noise reduction after reducing the noise. The distortion of signal waveform is verified by simulation and engineering practical signals.
3. for the problem of rolling bearing fault feature extraction, a time frequency fault feature extraction method based on tensor manifold learning is proposed. Based on the time-frequency characteristics of HHT, the tensor manifold learning method is used to extract the time-frequency characteristics of the nonlinear tensor flow manifolds of the signal, and the time frequency characteristic parameters are defined, and the time frequency characteristic parameters and the probability of the tensor manifolds are given. With the combination of neural network, the classification of bearing fault samples is accurately realized. The tensor manifold learning can effectively extract the intrinsic nonlinear characteristics of the combination of high dimensional frequency characteristics. Compared with the HHT time-frequency characteristic parameters, the tensor manifold time frequency characteristic parameters can reduce the redundancy of the feature information, divide the different types of fault samples more effectively, and reduce the neural network. The number of iterations can improve the accuracy of the fault classification.
4. for the problem of fault source separation of rolling bearings, a blind separation method of rolling bearing fault source based on manifold learning is proposed. The multi channel test signal is constructed by EMD decomposition, the number of source of the test signal is estimated, the selection standard of the optimal test signal is established, and the optimal test letter is selected by comprehensive use of kurtosis, sparsity and mutual information standard. By extracting the KPCA manifold component of the optimal test signal as the input of the ICA algorithm, the fault source is effectively separated. The method solves the problem of selecting the optimal test signal in the underdetermined blind separation process. The separation ability of the ICA is enhanced by manifold learning, and the impact feature can be separated from the weak single channel signal with the weak fault signal. An obvious source of failure.
5. in view of the performance degradation monitoring of rolling bearings, a performance monitoring method based on manifold learning and fuzzy clustering is proposed. The wavelet packet decomposition is used to determine the sensitive frequency band of the monitoring signal. On this basis, the feature of the low dimensional manifold of the signal is extracted as the data sample of the fuzzy clustering, and the membership degree value of the sample is used as the performance index. Compared with the single feature and linear multi feature monitoring method, the method can effectively reflect the four stages of the life performance degradation period of the rolling bearing, reflect the unified law of the performance degradation of the rolling bearing components, and advance the early prediction of the bearing barrier.
6. the fault analysis and diagnosis system of rolling bearing based on manifold learning is developed using the hybrid programming of LabVIEW and MATLAB. The software and hardware environment and the structure scheme of the system development are introduced. The basic functions of the system are demonstrated by an example, and the effectiveness of the system is verified.
【学位授予单位】：大连理工大学
【学位级别】：博士
【学位授予年份】：2013
【分类号】：TH165.3;TH133.33

【引证文献】