滚动轴承振动信号特征提取与状态评估方法研究

发布时间：2018-03-21 02:02

本文选题：滚动轴承　切入点：振动信号　出处：《哈尔滨工业大学》2015年博士论文　论文类型：学位论文

【摘要】：滚动轴承是众多旋转机械的关键性部件,被人们称为机器的关节。其在极端环境下,受各种因素的影响,是整个旋转机械系统中可靠性最差的零部件,成为“水桶短板”,直接影响整个机械设备的运行可靠性。滚动轴承运行时,其性能一般会从正常状态逐渐衰退直至完全损坏。如果能在轴承损坏过程中检测到它的性能退化程度或同时检测到故障位置及性能退化程度,就能够变传统的定时或事后维修为视情维修,实现轴承的主动维护。这样可最大限度利用轴承寿命,降低维护保障成本,避免进一步导致事故发生,造成巨大的损失。该研究方向侧重于设备整体性能的研究,是从理念和方法上对现有故障诊断技术的拓展与深入。本文以滚动轴承为研究对象,对滚动轴承非平稳振动信号进行分析与处理。以实现一种滚动轴承不同状态(正常状态,内环、滚动体、外环故障状态及其不同性能退化程度)智能评估方法为目标,在深入研究振动信号特征提取与约简方法的基础上,进一步深入研究超球结构多类支持向量机的优化问题,以及如何建立滚动轴承多状态评估模型这一关键技术问题。论文的主要工作包括:(1)基于集合经验模态分解(EEMD)的振动信号时频分析方法研究。对比分析验证了基于EEMD的Hilbert谱时频分析方法具有时间和频率分辨率高、抗模态混叠的特点。并针对EEMD分解时,加入噪声幅值大小和集合平均次数这2个重要参数的选取问题,从能量标准差的角度,研究EEMD方法中加入白噪声的准则;针对滚动轴承振动信号经改进的EEMD分解后得到的固有模态函数(IMF)仍含有虚假分量和对滚动轴承故障不敏感的IMF分量问题,研究一种峭度值结合归一化相关系数的IMF提存算法。实验研究验证了所提方法的有效性,为后续进一步提取特征打下坚实的基础。(2)振动信号多域特征提取与约简方法研究。为了精细刻画滚动轴承运行状态,体现滚动轴承振动信号的全局特征以及局部特征,研究时域、频域和时频域的多域特征提取方案。其中,时频域特征提取方面提出了基于改进EEMD敏感IMF分别结合时域指标、频域指标、自回归模型和奇异值分解的方法。基于此构造了滚动轴承单个样本的特征向量以及各状态的特征向量矩阵,并建立了滚动轴承各状态特征库。针对高维特征之间存在相关性和冗余性的问题,研究流形学习算法,结合支持向量机(SVM)通过实验对比分析,确定了对滚动轴承特征约简最有效的方法。(3)智能分类方法及故障智能诊断方法研究。超球结构多类SVM虽具有一系列优点,但其分类精度与普通SVM相比并不高。针对此问题,研究分类规则,提出改进方案,并对关键区域提出了新的决策准则。同时,针对经验确定超球结构多类SVM核参数选取范围的问题,推导超球球心间的距离计算公式,提出将球心间的距离作为分离指数确定核参数的最优选取范围,达到了降低训练时间消耗的目的。深入研究了滚动轴承不同运行状态的智能故障诊断方法。建立了超球结构多类SVM智能诊断模型,并进行了大量实验,验证了所提方法的有效性。(4)滚动轴承状态评估方法研究。针对滚动轴承故障智能诊断方法只能判断轴承故障状态的从属关系,不能对损伤程度和故障变化进行量化描述,以此来定量评估其性能状态的问题,从SVM分类原理、滚动轴承结构及传感器安装位置的振动传播机理角度分析,提出基于SVM的补偿相对距离的评估指标;从改进超球结构多类SVM原理、特征向量的方向、各状态超球的位置关系多方面分析,又提出夹角余弦距离补偿广义最小距离的评估指标,建立了智能评估模型。通过滚动轴承不完备振动数据和全寿命完备数据两方面的实验研究,对比分析了各评估模型的性能。
[Abstract]:The rolling bearing is the key component of large rotating machinery, known as the machine joints. In extreme environments, affected by various factors, is the worst of the rotating parts reliability of mechanical systems, a "bucket short board", directly affects the reliability of the whole machine. The rolling operation, the performance will gradually decline from the normal state until completely damaged. If to its performance in detection of bearing damage in the process of degradation degree or detected fault position and performance degradation degree, can change the traditional timing or maintenance for maintenance, maintenance to achieve active bearing. It can maximize the use of bearing life, reduce maintenance costs, avoid further lead to accidents, resulting in huge losses. The research direction focuses on the overall performance of the equipment from the idea and the method Further exploration on the existing fault diagnosis technology. In this paper, the rolling bearing as the research object, the rolling bearing of non-stationary vibration signal analysis and processing. In order to achieve a rolling bearing state (normal, inner ring, rolling body, outer ring fault and not the same degree of degradation) intelligent assessment methods. Based on the method of vibration signal feature extraction and reduction deeply on the optimization problem of further research on the hyper sphere structure of multi class support vector machine and evaluation model which is a key technical problem of how to establish the multi state of rolling bearing. The main work includes: (1) based on ensemble empirical mode decomposition (EEMD) method the vibration signal time-frequency analysis. Comparative analysis verified that the EEMD Hilbert spectrum time-frequency analysis method with time and frequency resolution based on the characteristics of anti mode mixing. And for EEMD decomposition When the problem of selecting add noise amplitude and the average number of this set of 2 important parameters, standard deviation from the energy point of view, adding white noise of the EEMD method for intrinsic mode function criterion; vibration signal of rolling bearing was improved after the EEMD decomposition (IMF) still contain false component and is not sensitive to the rolling bearing fault component IMF, a kurtosis value IMF drawing algorithm combined with the normalized correlation coefficient. Experimental results verify the effectiveness of the proposed method, for further extraction and lay a solid foundation. (2) research on feature extraction and reduction method of vibration signal of multi domain features. In order to characterize the running state of rolling bearing, reflect the global features of the rolling bearing vibration signal and local characteristics, research on multi domain feature extraction in time domain, frequency domain and time-frequency domain. The time domain and frequency domain feature extraction are analyzed. To improve the EEMD sensitive IMF respectively based on the time domain index, frequency index, autoregressive model and singular value decomposition method. The structure of the eigenvector matrix feature vector of the rolling bearing and the single sample based on the state, and the establishment of the State Library of rolling bearing. According to the existing characteristics of relevance and redundancy between the high dimension problem study, manifold learning algorithm, combined with support vector machine (SVM) through experimental analysis to determine the most effective method of rolling bearing feature reduction. (3) research on intelligent classification method and intelligent fault diagnosis methods. Hyper sphere multi class SVM structure has a series of advantages, but its classification accuracy is compared with ordinary SVM high. To solve this problem, the research of classification rules, the improvement scheme is proposed, and a new decision criteria of key areas. At the same time, according to the empirical determination of hyper sphere multi class SVM nuclear structure parameter range The problem, the calculation formula of super sphere distance, the distance between the center of the proposed optimal separation as the index to determine the kernel parameters selection, to reduce the training time consumption. The further study of the intelligent fault diagnosis method of rolling bearings in different working state. A multi class SVM intelligent diagnosis model of hyper sphere structure and a large number of experiments, verify the effectiveness of the proposed method. (4) research on the evaluation method of rolling bearing. According to the bearing fault intelligent diagnosing method can determine fault state of affiliation, cannot be quantified description of damage and failure in order to change the performance state of the quantitative evaluation, from the principle SVM classification, analysis of vibration propagation mechanism of rolling bearing structure and sensor position, put forward the evaluation index of SVM compensation based on relative distance from the modified sphere; The structure of multi class SVM principle, characteristic vector direction, of the state sphere in many aspects, and put forward the evaluation index of cosine distance compensation generalized minimum distance, the establishment of intelligent evaluation model. Through the rolling bearing vibration data and incomplete life complete data on two aspects of the experimental research, comparative analysis of the performance the evaluation of the model.

【学位授予单位】：哈尔滨工业大学
【学位级别】：博士
【学位授予年份】：2015
【分类号】：TH133.33;TN911.7

【相似文献】