滚动轴承振动信号非平稳、非高斯分析及故障诊断研究

发布时间：2018-06-05 03:42

本文选题：滚动轴承 + 非平稳分析　；参考：《西安电子科技大学》2014年博士论文

【摘要】：滚动轴承是旋转机械的重要组成部件,其性能状态对机械设备运行及效率起至关重要的作用。基于振动分析的滚动轴承智能故障诊断已成为国内外学者的研究热点,但相关研究主要针对固定工况运行环境,难以满足工程实践中载荷变化和转速波动的变工况故障诊断需要。本文基于滚动轴承振动信号的非平稳、非高斯特性,对特征参数提取、特征向量优化、变工况故障诊断和性能退化评估等问题展开研究。主要工作概述如下:1.研究振动信号的连续小波变换,提出一种基于最小香农熵和奇异值分解的Morlet小波参数优化方法。最小香农熵意味着小波系数稀疏,保证小波波形和信号之间的较高相似度,奇异值分解可以检测序列的周期特性,二者相结合能提取更有效的故障特征信息。研究最优Morlet小波系数的常用统计参量性质,标准差、均值、均方根和无穷范数在不同轴承状态下差异性显著,作为特征参数获得可靠的故障诊断结果。最后用比较实验证明Morlet小波参数优化的有效性和可靠性。2.研究振动信号的小波包分解,提出一种差异性和相似性相结合的特征向量优化方法。振动信号小波包分解获得多个带宽相同的子频带,但有效故障特征信息只分布在少量子带中,特征向量存在冗余信息。选择Daub8小波,根据子带宽度和谐波频率估算小波包分解层数,并将子带能量作为参数构造特征向量。基于Fisher线性距离测度,差异性优化选出不同轴承状态下距离较大的特征向量行向量,相似性优化选出特征向量内距离较小的行向量。优化特征向量具有较大的类间差异性和类内相似性,在突出故障特征信息同时抑制了干扰成分。比较实验表明文中优化方法的辨识精度优于文献方法。3.对振动信号进行小波降噪研究,提出一种基于短时过零率的工况鲁棒早期故障诊断方法。过零率只与信号通过零点的频度有关而与波形或幅度无关,对工况改变导致的振动信号波形变化鲁棒,也能在一定程度上表征信号的频域信息。确定小波函数、分解层数和阈值策略后,研究小波降噪信号的短时过零率特点,其在不同故障状态下的差异性明显,在故障相同但工况不同时又具有较大的相似性,是一种工况鲁棒的特征参数。使用任意一种工况的数据训练模型,都能正确辨识当前工况和其它三种工况的故障类型,实现工况鲁棒的早期故障诊断。4.研究滚动轴承振动信号的非高斯特性,提出一种基于双谱主成分分析的智能故障诊断方法。先对振动信号的双谱特性进行研究,其幅度和分布特性在不同故障类型时具有明显的差异性,在故障相同但工况不同时又具有一定的相似性。使用主成分分析方法提取双谱中的有效特征信息,取其幅值作为特征参数,实现了不同工况和不同故障程度的轴承状态判别。此外,零载荷工况数据训练的模型,能辨识其它三种不同工况的故障类型,具有工况鲁棒的故障诊断功能。5.对滚动轴承性能退化评估进行研究,提出一种基于隐马尔可夫模型距离的性能退化评估指标。先设计滚动轴承加速度疲劳寿命试验,并自制数据采集系统记录6205轴承性能退化过程的振动加速度信号。研究常用诊断指标在性能退化过程中的变化规律,发现滚动轴承性能退化过程经历六个不同阶段,将其命名为:正常状态、早期故障、中度故障、严重故障、预警阶段和轴承失效。振动信号的均方根作为特征参数训练隐马尔可夫模型,并将初始寿命时刻模型作为基准点,计算性能退化过程模型与基准点之间的距离,结果表明隐马尔可夫模型距离是一种有效的性能退化评估指标。
[Abstract]:Rolling bearing is an important component of rotating machinery. Its performance state plays an important role in the operation and efficiency of mechanical equipment. The intelligent fault diagnosis of rolling bearing based on vibration analysis has become a hot spot of research at home and abroad. However, the related research is mainly aimed at the operating environment of fixed working conditions, and it is difficult to meet the load change in engineering practice. Based on the non-stationary and non Gauss characteristics of the vibration signals of rolling bearings, this paper studies the problems of feature extraction, eigenvector optimization, variable condition fault diagnosis and performance degradation evaluation. The main work is summarized as follows: 1. the continuous wavelet transform of vibration signals is studied, and a kind of continuous wavelet transform is proposed. The Morlet wavelet parameter optimization method based on the minimum Shannon entropy and singular value decomposition. The minimum Shannon entropy means that the wavelet coefficients are sparse, and the high similarity between the wavelets and the signals is guaranteed. The singular value decomposition can detect the periodic characteristics of the sequence. The combination of the two can extract more effective fault feature information. The optimal Morlet wavelet system is studied. The properties of the common statistical parameters, the standard deviation, the mean value, the root mean square and the infinite norm are significant in the different bearing states. As the characteristic parameters, the reliable fault diagnosis results are obtained. Finally, the validity and reliability of the Morlet wavelet parameter optimization are proved by comparative experiments. The difference of the wavelet packet decomposition of the vibration signal is studied by.2.. The eigenvector optimization method combining nature and similarity is used. The wavelet packet decomposition of vibration signals obtains multiple subbands with the same bandwidth, but the effective fault feature information is only distributed in a small number of subbands, and the eigenvectors have redundant information. The Daub8 wavelet is selected to estimate the number of wavelet packet decomposition layers based on the width and harmonic frequency of the subband and the energy of the subband. The feature vector is constructed as a parameter. Based on the Fisher linear distance measure, the row vector of the feature vector with a larger distance in different bearing States is optimized and the row vector with a smaller distance within the feature vector is selected. The comparison experiment shows that the identification accuracy of the optimization method is better than the literature method.3. to study the wavelet denoising of the vibration signal. A robust early fault diagnosis method based on the short-time zero crossing rate is proposed. The zero crossing rate is only related to the frequency of the zero point of the signal, which is independent of the waveform or amplitude, and the working condition is changed. The change of the vibration signal waveform is robust, and can also represent the frequency domain information of the signal to a certain extent. After determining the wavelet function, the decomposition layer number and the threshold strategy, the short time zero crossing rate characteristic of the wavelet denoising signal is studied. The difference is obvious in the different fault state, and it has the larger similarity in the same fault but not at the same time. It is a robust characteristic parameter. Using the data training model of any working condition, the fault types of the current and other three operating conditions can be identified correctly, and the early fault diagnosis.4. is robust to study the non Gauss characteristics of the vibration signal of the rolling bearing and an intelligent fault based on the bispectrum principal component analysis is proposed. The method of diagnosis is to study the bispectrum characteristic of the vibration signal first. Its amplitude and distribution characteristics have distinct difference when the fault types are different, and are similar in the same fault but not at the same time. Using the principal component analysis method to extract the effective feature information in the bispectrum and take its amplitude as the characteristic parameter. Bearing state discrimination of different working conditions and different fault degrees. In addition, the model of data training of zero load condition can identify the other three different types of fault. The fault diagnosis function.5. with robust working condition is used to evaluate the performance degradation of rolling bearings, and a performance degradation evaluation based on Hidden Markov model distance is proposed. First design the rolling bearing acceleration fatigue life test, and record the vibration acceleration signal of the 6205 bearing performance degradation process by the self-made data acquisition system, and study the change law of the common diagnostic index in the process of performance degradation, and find that the performance degradation process of the rolling bearing has gone through six different stages, which is named as the normal state, Early fault, moderate fault, serious fault, early warning stage and bearing failure. The root mean square of the vibration signal is used as a feature parameter to train hidden Markov model, and the initial life time model is used as a reference point to calculate the distance between the performance degradation process model and the base point. The results show that the hidden Markov model distance is an effective way. Performance degradation assessment indicators.
【学位授予单位】：西安电子科技大学
【学位级别】：博士
【学位授予年份】：2014
【分类号】：TH133.33

【参考文献】