【作者】 杨宇；

【导师】 于德介；

【作者基本信息】 湖南大学 ， 机械工程， 2005， 博士

【摘要】 机械设备的诊断过程包括诊断信息获取、故障特征信息提取和状态识别三部分。其中,故障特征提取和状态识别是诊断的关键。本文将时频分析的新方法—经验模态分解(Empirical Mode Decomposition,简称EMD)和模式识别的新技术—支持向量机(Support Vector Machine,简称SVM)相结合应用于旋转机械故障诊断当中。EMD方法基于信号的局部特征时间尺度,可把信号分解为若干个内禀模态函数(Intrinsic Mode Function,简称IMF)之和,分解出的各个IMF分量突出了数据的局部特征,对其进行分析可以更准确有效地把握原数据的特征信息。此外,由于每一个IMF所包含的频率成分不仅仅与采样频率有关,更为重要的是它还随着信号本身的变化而变化,因此EMD方法是一种自适应的时频局部化分析方法,它从根本上摆脱了Fourier变换的局限性,具有很高的信噪比,非常适用于非平稳、非线性过程。针对旋转机械故障振动信号的非平稳特征,本文将EMD方法引入旋转机械故障特征提取当中,对其基本理论进行了研究,对其边界效应提出了解决方案,并在此基础上提出了五种基于内禀模态函数的故障特征提取方法。支持向量机有比神经网络更好的泛化能力,且能保证找到的极值解就是全局最优解,同时它还较好地解决了小样本的学习分类问题。针对机械故障诊断中难以获得大量典型故障样本的实际情况以及支持向量机优良的分类性能,本文采用支持向量机作为分类器对旋转机械的工作状态和故障类型进行了分类,并对支持向量机在小样本故障诊断中的应用进行了较为全面的研究。对实验数据的分析结果表明,EMD和SVM相结合可有效地应用于旋转机械故障诊断当中。 本文主要工作包括: (1)讨论了传统时频分析方法在信号处理中的应用,指出了其缺陷,并在此基础上介绍了时频分析的新方法—Hilbert-Huang变换,它包括EMD方法和Hilbert变换两部分。对仿真信号的分析结果表明,EMD方法的分解效果优于小波方法;Hilbert谱具有比小波谱更高的分辨率;Hilbert-Huang变换所得到的Hilbert边际谱具有比FT谱更高的分辨率。 (2)采用EMD方法和Hilbert变换对信号进行时频分析时会产生边界效应,针对这一问题,本文采用径向基函数网络对信号进行了延长。对仿真信号的分析结果表明,该延拓方法能有效地抑制边界效应。 (3)在分析传统统计模式识别方法和人工神经网络分类器的缺陷的基础上,针对机械故障诊断中难以获得大量典型故障样本的实际情况以及支持向量机优良的分类性能,将支持向量机引入旋转机械故障诊断当中。采用神经网络和支持向量机两种技术进行了小样本试验研究,研究结果表明支持向量机无论在训练速度更多还原

【Abstract】 The process of machinery fault diagnosis includes the acquisition of information and extracting feature and recognizing conditions of which feature extraction and condition identification are the priority. A novel method of time-frequency analysis, Empirical Mode Decomposition (EMD) and the comparatively recent development of pattern recognition techniques, Support Vector Machines (SVMs), are combined and applied to the rotating machinery fault diagnosis. EMD is based on the local characteristic time scale of signal and decompose the complicated signal into a number of Intrinsic Mode Functions (IMFs). By analyzing each IMF component that involves the local characteristic of the signal, the characteristic information of the original signal could be extracted more accurately and effectively. In addition, the frequency components involved in each IMF not only relates to sampling frequency but also changes with the signal itself, therefore, EMD is a self-adaptive time frequency analysis method that is applicable to non-linear and non-stationary processes perfectly thus overcoming the limitations experienced by the Fourier Transform (FT). In addition the EMD has a high signal-noise ratio (SNR). According to the non-stationary vibration signal characteristics of rotating machinery EMD method is introduced into rotating machinery fault diagnosis. The EMD method is improved upon and five types of feature extraction methods based on IMFs are proposed. SVMs have better generalization than Artificial Neural Networks (ANNs) and guarantee the local optimal solution is exactly the global optimal solution. SVMs can solve the learning problem of a smaller number of samples. Due to the fact that it is difficult to obtain sufficient fault samples in practice, SVMs are introduced into rotating machinery fault diagnosis due to their high accuracy and good generalization for a smaller sample number. The experimental results demonstrate the proposed diagnosis approach in which EMD and SVM are combined is effective.The outline of the work is as follows:1. The applications and limitations of conventional time frequency analysis method in signal processing are briefly discussed. A new theory of time frequency analysis method, Hilbert-Huang Transform (HHT), which includes EMD and Hilbert transform, is introduced. The analysis results from simulation signals show the decomposition effect of EMD is superior to that of the wavelet method; Hilbert spectrum obtained by HHT has a higher resolution than wavelet spectrum; Hilbert marginal spectrum has ahigher resolution than that obtained by FT.2. To target the disadvantage that the end effects will occur when EMD method and Hilbert transform are used to analysis signals, Radial Basis Function (RBF) Networks are introduced to prolong the original signal. The analysis results from simulation signals show that the proposed method effectively restrains the end effects.3. The limitations of the conventional statistical pattern recognition methods and ANNs classifier are targeted. SVMs are introduced into rotating machinery fault diagnosis due to the fact that it is hard to obtain enough fault samples in practice. This dissertation offers a comparison between two classification algorithms, ANNs and SVMs for cases where only limited training samples are available for diagnosis. The results show that SVMs have better performance than ANNs both in training speed and recognition rate.4. The concept of IMFs energy entropy is proposed and an approach of fault feature extraction based on IMFs energy entropy is put forward. The energy of vibration signals in different frequency bands will change when faults occur, thus one or more kinds of energy variation in frequency components indicate fault occurrence. The analysis results from roller bearing vibration signals show that the fault diagnosis approach based on IMFs energy entropy and SVMs can extract fault features effectively and classify working condition and fault patterns accurately.5. The concepts of Hilbert marginal energy spectrum and local Hilbert marginal energy spectrum are proposed and an approach of fault feature extraction based on local Hilbert marginal energy spectrum is put forward. Hilbert spectrum offers a complete time-frequency distribution, Hilbert marginal energy spectrum offers an energy distribution in the frequency domain and local Hilbert marginal energy spectrum offers an energy distribution in intrinsic frequency band. The analysis results from roller bearing vibration signals show that the fault diagnosis method based on local Hilbert marginal energy spectrum can extract fault features effectively and classify working condition and fault patterns accurately.6. Hilbert marginal spectrum reflects the amplitude distribution in the frequency domain accurately. To highlight the changes of energy in the intrinsic frequency band, an approach of fault feature extraction based on Hilbert marginal spectrum is proposed. The analysis results from roller bearing vibration signals shows that the fault diagnosis approach based on Hilbert marginal spectrum can extract fault features effectively and classify working condition and fault patterns accurately.7. Targeting the characteristics that periodic impulses usually occur whilst the更多还原