基于谱峭度及原子分解的滚动轴承故障诊断方法研究

发布时间：2018-03-09 20:03

本文选题：滚动轴承　切入点：谱峭度　出处：《上海大学》2014年博士论文　论文类型：学位论文

【摘要】：滚动轴承是最为广泛应用的旋转机械通用零部件之一，其工作状态的优劣直接关系到整台机组乃至整条生产线的生产质量和安全，因此，滚动轴承故障诊断技术的研究具有十分重要的意义。特征提取是故障诊断技术的关键环节，本文针对滚动轴承故障特征提取现有方法的不足，深入研究了谱峭度及原子分解两种有效的故障特征提取方法： (1)谱峭度为经典滚动轴承共振解调方法提供了有效的自适应频带选择工具，然而，双参数同时精确定位的短时傅利叶变换谱峭度方法由于其巨大的计算量限制了实用性，而固定带宽的单参数定位Protrugram方法失去了带宽自适应性。本文根据滚动轴承故障振动信号的调幅特性，提出包络定位FFT谱峭度方法，分步实现了中心频率与带宽的定位，解决了计算量与自适应性之间的矛盾。通过三种滚动轴承故障诊断方法的对比研究，验证了新方法的有效性、自适应性以及实用性。 (2)有限冲击响应（FIR）滤波器快速谱峭度方法提供了快速定位带宽与中心频度的近似方法，但由于其滤波器建立在傅利叶变换基础上，对非平稳信号特征提取具有局限性。本文提出小波包滤波器组快速谱峭度方法，并根据滚动轴承故障信息分布频带广泛的特点，利用同步平均降噪原理，提出将指定同一分解层的部分或者全部子带包络谱的累积包络谱方法，有效增强了轴承有用信息，提高了对滚动轴承故障特征的识别能力。 (3)同一类特征原子组成的字典难以适应实际信号由多种物理现象混合而成的复杂性，使得信号分解结果稀疏度不足、物理解释困难。本文根据滚动轴承振动信号特征，构造由余弦包（CP）原子与小波包原子（WP）组成的混合字典，并提出快速CPWP混合原子分解匹配追踪算法，提高了分解结果的稀疏性，，增强了物理解释性。通过对滚动轴承的故障诊断，表明CPWP混合原子分解能够有效提取到冲击成分与载波成分，全面反映了滚动轴承故障特征。 (4)构造与实际复杂变化信号一致的参数化原子具有很大的困难。本文根据滚动轴承故障冲击的循环平稳与随机性双重特性，利用匹配追踪对原子构造的宽松条件以及对信号的近似表达，提出从故障信号中提取特征波形，构造非参数化字典的非参数化原子分解诊断方法，从匹配度、稀疏度以及频率分辨率几方面研究了新方法的优越性。通过对滚动轴承实测信号的诊断分析，并与谱相关密度方法以及包络解调方法进行对比，验证了该方法的有效性。
[Abstract]:Rolling bearing is one of the most widely used universal parts of rotating machinery. Its working condition is directly related to the production quality and safety of the whole unit and even the whole production line. The research of rolling bearing fault diagnosis technology is of great significance. Feature extraction is the key link of fault diagnosis technology. In this paper, two effective fault feature extraction methods, spectral kurtosis and atomic decomposition, are studied in order to overcome the shortcomings of the existing methods for fault feature extraction of rolling bearings. Spectral kurtosis provides an effective adaptive frequency band selection tool for the classical resonance demodulation method for rolling bearings. However, the short-time Fourier transform spectral kurtosis method, which can be accurately located by two parameters at the same time, has limited its practicability due to its huge computational complexity. In this paper, according to the amplitude modulation characteristic of rolling bearing fault vibration signal, the envelope positioning FFT spectrum kurtosis method is proposed, which realizes the location of center frequency and bandwidth step by step. The contradiction between computation and self-adaptability is solved, and the validity, adaptability and practicability of the new method are verified by the comparative study of three fault diagnosis methods for rolling bearings. The fast spectral kurtosis method of finite impulse response (FIR) filters provides an approximate method for fast location bandwidth and center frequency, but the filter is based on Fourier transform. In this paper, a fast spectral kurtosis method for wavelet packet filter banks is proposed. According to the wide frequency distribution of fault information of rolling bearings, the principle of synchronous average noise reduction is used. An accumulative envelope spectrum method with partial or all sub-band envelope spectrum assigned to the same decomposition layer is proposed, which can effectively enhance the useful information of bearings and improve the ability to identify the fault characteristics of rolling bearings. 3) the dictionary composed of the same kind of characteristic atoms is difficult to adapt to the complexity of the actual signals mixed by many physical phenomena, which makes the results of signal decomposition insufficient in sparsity and difficult in physical interpretation. In this paper, according to the characteristics of the vibration signals of rolling bearings, A hybrid dictionary consisting of cosine wrapped atom and wavelet packet atom is constructed, and a fast CPWP hybrid atomic decomposition matching tracing algorithm is proposed, which improves the sparsity of decomposition results and enhances the physical interpretation. The results show that the CPWP hybrid atomic decomposition can extract the impact and carrier components effectively and reflect the fault characteristics of rolling bearings. It is difficult to construct a parameterized atom which is consistent with the actual complex variation signal. In this paper, the cycle stability and randomness of rolling bearing fault impact are analyzed. Based on the loose conditions of atomic construction and approximate expression of signals by matching tracing, a nonparametric atomic decomposition diagnosis method is proposed to extract characteristic waveforms from fault signals and construct nonparametric dictionaries. The superiority of the new method is studied in terms of sparsity and frequency resolution. The effectiveness of this method is verified by the diagnosis and analysis of the measured signals of rolling bearings and the comparison with the spectral correlation density method and the envelope demodulation method.
【学位授予单位】：上海大学
【学位级别】：博士
【学位授予年份】：2014
【分类号】：TH133.33;TH165.3

【参考文献】