强噪声干扰下行星轮系振动信号分析及其故障诊断技术研究

发布时间：2018-01-10 10:08

  本文关键词：强噪声干扰下行星轮系振动信号分析及其故障诊断技术研究　出处：《中国矿业大学》2017年博士论文　论文类型：学位论文

  更多相关文章： 行星轮系 振动信号 预处理降噪 特征提取 多传感器 故障诊断

【摘要】：大型复杂机电装备涉及煤炭、航天、钢铁、船舶、工程机械等重要制造行业,是我国制造业和工业发展的重要基础。随着工业化进程的不断推进和科学技术的快速发展,大型复杂机电装备日益趋向复杂、可靠、高效、智能化方向发展。由于行星轮系具有众多优点,现已成为大型复杂机电装备传动系统的重要组成部件。但是行星轮系一般工作于大载荷、强干扰、高污染的恶劣工况,其故障时有发生,直接影响机电装备的传动效率,严重时会导致整个机电装备失效,甚至人员伤亡等恶劣后果。因此,对行星轮系进行故障诊断研究具有非常重大的意义。但是在真实工况条件下所测得的行星轮系振动信号一般包含强噪声干扰,并且其特殊的结构及工作方式,也致使行星轮系的故障诊断具有自身的特点和难点。本文以强噪声干扰下的行星轮系为研究对象,通过对预处理降噪、故障特征信息提取、特征降维处理、多传感器融合诊断进行深入研究,形成基于振动信号分析的行星轮系故障诊断技术,为保障行星轮系以及机电装备传动系统安全运行提供理论支撑和技术解决方案。主要内容包括:(1)针对行星轮系的具体结构及工作方式,分析了进行行星轮系故障诊断研究的特点和难点;并进行了真实工况条件下强噪声干扰获取实验和不同行星轮系故障状态的模拟实验,获得了真实工况条件下的强噪声干扰和不同行星轮系故障状态的多传感器信号。(2)针对行星轮系在真实工况中所遭受的强噪声干扰,提出了一种结合双树复小波变换(Dual-Tree Complex Wavelet Transform,DTCWT)和循环奇异能量差分谱的预处理降噪方法。利用DTCWT具有较少频率混叠和频率泄漏的优点,将包含强噪声干扰的原始振动信号分解到多个具有不同频率特性的信号中;通过对奇异值分解降噪原理分析,基于级联循环、逐次滤除噪声的思想提出了循环奇异能量差分谱降噪方法,根据不同频带噪声干扰分布特点设置不同的终止条件实现了对各频带信号的降噪处理。利用所提出的预处理降噪方法,能够有效消除真实工况强噪声干扰,保留行星轮系产生的有效信号成分。(3)针对行星轮系所产生的振动信号具有非线性、非平稳、强耦合的特性,在充分研究经验模态分解(Empirical Mode Decomposition,EMD)和总体经验模态分解(Ensemble Empirical Mode Decomposition,EEMD)的基础上,研究了一种自适应噪声的完备总体经验模态分解(Complete Ensemble Empirical Mode Decomposition with Adaptive Noise,CEEMDAN)算法。并且针对CEEMDAN存在添加高斯白噪声次数过多、计算耗时等缺点,提出了一种改进的自适应噪声的完备总体经验模态分解(Improved Complete Ensemble Empirical Mode Decomposition with Adaptive Noise,ICEEMDAN)算法。一方面在求取本征模态函数(Intrinsic Mode Function,IMF)分量过程中,连续检测所添加高斯白噪声对IMF分量的影响,控制在求取IMF分量过程中添加的高斯白噪声次数,有效减少了非必要高斯白噪声的添加;另一方面,引入排列熵对IMF分量进行复杂性检测,根据排列熵值变化情况决定是否继续需要高斯白噪声的辅助分解作用。所提出的ICEEMDAN方法能够有效抑制信号模态混叠,保证信号分解质量和分解的完备性,并且有效减少非必要高斯白噪声的添加,具有相对更快的计算速度。(4)针对行星轮系产生的微弱故障特征信息,基于行星轮系特征频率及其相关成分信噪比建立了有效IMF分量提取准则,结合粒子群优化(Particle Swarm Optimization,PSO)算法和随机共振算法构建了自适应随机共振系统处理有效IMF分量的重构信号,能够有效提取行星轮系啮合频率及其边频带信息,可定性比较判断行星轮系状态。另外,为了量化各IMF分量中的行星轮系特征信息,基于信息熵定义,构建了多角度熵特征联合提取模型,实现了包含在各IMF分量中故障特征信息的多角度提取量化,形成了全面而综合的原始故障特征集合。(5)针对原始故障特征集合存在维数过大、信息冗余和无效特征干扰等情况,开展了基于核方法的高维特征降维研究。在对主元分析(Principal Component Analysis,PCA)、Fisher鉴别分析(Fisher Discriminant Analysis,FDA)和核主元分析(Kernel Principal Component Analysis,KPCA)研究基础上,研究了一种优化参数的核Fisher鉴别分析(Kernel Fisher Discriminant Analysis,KFDA)方法。根据核特征空间的类间散度矩阵和类内散度矩阵关系,定义了类对可分性和类别可分性,进而形成了基于类别可分性的KFDA核参数优化选取准则,解决了KFDA的核参数选择问题。并且利用优化参数的KFDA方法能够有效完成对原始故障特征集合的特征融合和降维处理,提取出敏感故障特征。(6)针对真实工况强噪声干扰下,基于单传感器进行故障诊断容易引起信息缺失、出现识别不确定性、故障诊断精度降低和诊断信任程度降低等情况,开展了多传感器融合诊断研究。基于所提取的敏感故障特征和极限学习机(Extreme Learning Machine,ELM)获得了单传感器产生的局部诊断结论,提出了基于ELM误差距离的基本信任函数分配方法,并建立了基于证据冲突检测的D-S证据理论融合规则。通过本文所建立的方法对强噪声干扰下的行星轮系进行多传感器融合诊断,可消除识别的不确定性,处理证据冲突情况,有效提高故障诊断精度和诊断信任程度。文章最后对论文的工作进行了总结,并对相关的研究技术进行了展望。
[Abstract]:Large-scale complex electromechanical equipment involving coal, steel, shipbuilding, aerospace, engineering machinery and other important manufacturing industry, is an important basic industry and manufacturing industry development in our country. With the rapid development of the industrialization process of science and technology, large-scale complex electromechanical equipment to more complex, reliable, efficient, intelligent direction because. The planetary gear has many advantages, has become an important component of large-scale complex electromechanical equipment transmission system. But planetary gear trains generally work in large load, strong interference, harsh working conditions and high pollution, the fault occurred, directly affect the transmission efficiency of electrical equipment, which can lead to the mechanical and electrical equipment failure, even casualties bad consequences. Therefore, it has very important meaning to research on fault diagnosis of planetary gear planetary gear vibration test. The trust in the real conditions but General contains strong noise interference, and its special structure and working mode, has its own characteristics and difficulties of the fault diagnosis of planetary gear train. This paper also leads to noise of the planetary gear train as the research object, through the pretreatment of noise reduction, fault feature extraction, feature dimension reduction, multi-sensor fusion diagnosis study the formation of planetary gear fault diagnosis based on vibration signal analysis, provide theoretical support and technical solutions for the protection of the planetary gear transmission system and mechanical and electrical equipment safety operation. The main content includes: (1) the specific structure and operation mode of the planetary gear train, analyzes the characteristics and difficulties of gear fault diagnosis on the planet true; and make a simulation experiment under the condition of strong noise and obtain the different planetary gear fault condition, get real conditions Multi sensor signal in strong noise and different planetary gear fault state. (2) according to the strong noise of planetary gear train suffered in the real conditions, proposed a combination of dual tree complex wavelet transform (Dual-Tree Complex Wavelet Transform, DTCWT) and the odd cycle ability differential spectral preprocessing denoising method. Using the DTCWT has less of the mixed frequency aliasing and frequency leakage advantages, will contain the original vibration signal in strong noise interference is decomposed into multiple signals with different frequency characteristics; based on singular value decomposition and de-noising principle analysis, based on the cascade cycle noise, successive proposed cyclic singular energy difference spectrum denoising method based on noise reduction. Different band noise distribution characteristics of different set termination conditions realized on each band signal. By pretreatment denoising method proposed can effectively eliminate it. The real condition of strong noise, retain the effective signal component of planetary gear train generated. (3) with nonlinear vibration signal generated according to the non-stationary characteristics of planetary gear train, strong coupling, on the basis of the empirical mode decomposition (Empirical Mode, Decomposition, EMD) and ensemble empirical mode decomposition (Ensemble Empirical Mode Decomposition, EEMD). Based on the complete modal analysis of the overall experience of an adaptive noise decomposition (Complete Ensemble Empirical Mode Decomposition with Adaptive Noise, CEEMDAN) and CEEMDAN algorithm. For there are too many added Gauss white noise frequency, computation time and other shortcomings, put forward a complete overall empirical mode of an improved adaptive noise decomposition (Improved Complete Ensemble Empirical Mode Decomposition with Adaptive Noise ICEEMDAN) algorithm. On the one hand, to obtain the intrinsic mode function (Intrinsi C Mode Function, IMF) component process, continuous detection of the added influence of Gauss white noise on the IMF component, control the Gauss white noise added in the times of seeking IMF component process, effectively reduce the need to add non Gauss white noise; on the other hand, the introduction of permutation entropy complexity detection of IMF component, according to the permutation entropy change the decision whether to continue to assist the Gauss white noise decomposition. ICEEMDAN the proposed method can effectively suppress the signal modal aliasing signal decomposition and decomposition of the quality assurance system, and effectively reduce the need to add non Gauss white noise, with relatively faster computing speed. (4) the weak fault information according to the characteristics of planetary gear, planetary gear train frequency and related components of SNR based on the establishment of an effective IMF component extraction criteria based on particle swarm optimization (Particle Swarm Optimiza Tion, PSO) algorithm and stochastic resonance algorithm is constructed to reconstruct the signal adaptive stochastic resonance system effectively with the IMF component, which can effectively extract the meshing planetary gear train frequency and side band information, can be compared to determine the state of planetary gear train. In addition, in order to quantify the characteristics of planetary gear train information component of each IMF, based on the definition of information entropy, construct multi angle joint entropy feature extraction model, realizes the multi angle containing fault feature information in the IMF component extraction in quantification, formed a comprehensive and comprehensive collection of original fault feature. (5) according to the original fault feature set in high dimension, information redundancy and invalid feature interference etc., carried out to reduce the dimension of the high feature based on kernel method. In the analysis of PCA (Principal Component Analysis, PCA), Fisher (Fisher Discriminant Analysis discriminant analysis, FDA) and Kernel (kernel principal component analysis Principal Component Analysis, KPCA) on the basis of an optimized kernel Fisher discriminant analysis of parameters (Kernel Fisher Discriminant Analysis, KFDA) method. According to the kernel feature space between class scatter matrix and within class scatter matrix, we define a class of separable and separability, and the formation of the parameter of KFDA kernel the optimization selection criterion of separability based on solving the KFDA kernel parameter selection problem. And the use of KFDA optimization method can effectively complete the fusion feature set on the original fault feature and reduce dimension, extract the sensitive fault features. (6) according to the actual condition of strong noise, easy to cause the single sensor fault diagnosis the lack of information based on the recognition of uncertainty, fault diagnosis and diagnosis accuracy reduce the degree of trust decreased, launched a multi sensor fusion study based on diagnosis. The extraction of fault feature sensitive and extreme learning machine (Extreme Learning Machine, ELM) and obtain the local diagnosis of single sensor production, puts forward the basic belief function assignment method for ELM errors based on distance, and the establishment of evidence conflict detection D-S evidence theory fusion rules based on the method proposed in this paper. Based on strong noise the planetary gear train for multi sensor fusion diagnosis, which can eliminate the uncertainty of identification, handling of evidence conflict, effectively improve the accuracy of fault diagnosis and diagnosis of the degree of trust. At the end of the thesis sums up, and the related research technology is prospected.

【学位授予单位】：中国矿业大学
【学位级别】：博士
【学位授予年份】：2017
【分类号】：TH132.425
，

本文编号：1404839