滚动轴承故障特征提取与早期诊断方法研究
发布时间：2018-11-05 14:20
【摘要】：滚动轴承是旋转机械中广泛应用的零部件之一,其安全运行与否对整个设备有至关重要的影响。本文以滚动轴承为研究对象,在总结滚动轴承现有故障诊断技术的基础上,采用现代信号处理技术,对滚动轴承的特征提取、早期故障检测和早期故障位置识别展开研究工作。本文的主要研究内容如下:(1)针对滚动轴承不同故障类型的振动信号所呈现的复杂度不同的特点,提出了基于多尺度符号动力学熵(MSDE)的滚动轴承故障位置特征提取方法。符号动力学熵在符号动力学滤波(SDF)的基础上,构建符号序列的状态模式矩阵和状态迁移矩阵,极大保留振动信号的故障信息。经过四种仿真信号测试,验证了符号动力学熵能够有效辨别信号动力学特性变化,并具有较高的计算效率。结合多尺度分析的概念,提出了MSDE的特征提取方法,使其能够在不同尺度下描述信号的复杂度。同时采用最大相关最小冗余进行特征优选和最小二乘支持向量机进行模式识别,经实验数据测试,基于MSDE的特征提取方法具有较高的识别精度和计算效率,实现了滚动轴承不同故障位置的准确区分。(2)针对多尺度模糊熵(MFE)对滚动轴承故障损伤程度特征提取能力不足的问题,提出了基于层次模糊熵(HFE)的特征提取方法。针对MFE中多尺度分析的缺点,提出了基于HFE的滚动轴承特征提取方法。层次分析能够同时提取信号中低频部分和高频部分隐藏的故障信息。通过白噪声和粉红噪声的测试,HFE能够更准确、全面描述时间序列复杂度的变化,且具有更好的稳定性。然后采用拉普拉斯分值对特征进行优选,并结合二叉树支持向量机实现滚动轴承不同故障程度的识别。同MFE方法进行对比分析表明基于HFE所提取的特征向量具有良好的可分性,实现了滚动轴承不同故障程度的准确区分。(3)为实现滚动轴承的早期故障检测,提出了基于SDF的性能监测指标M。监测指标M的计算过程不依赖于轴承的故障样本数据,且与有效值和峭度因子相比,监测指标M对轴承早期故障具有较高的敏感性和稳定性。将监测指标M与累积和相结合来完成轴承早期故障的报警,经两次轴承加速全寿命实验的测试,该方法能够有效地检测出轴承的早期故障。(4)针对滚动轴承早期故障特征极其微弱的问题,提出了基于内禀特征尺度分解(ICD)与优化品质因子的共振稀疏分解(ORSSD)的滚动轴承早期故障诊断方法。为了降低噪声的干扰,在总结局部均值分解(LMD)方法和本征时间尺度分解(ITD)方法优势的基础上,提出了ICD分解方法。以仿真信号为例,比较了ICD和LMD的分解性能。分析结果表明ICD具有更高的分解精度和计算效率,更加适合于分解滚动轴承的振动信号。针对共振稀疏分解中品质因子Q难以确定的问题,引入了特征频率比(CFR)进行品质因子Q的优选,根据最大的CFR值找出最优的高品质因子QH和低品质因子QL。将ICD与ORSSD相结合,提出了一种基于ICD和ORSSD的滚动轴承早期故障位置的识别方法。经仿真与实验信号测试,该方法能够有效提取早期故障特征,完成滚动轴承的早期故障诊断。
[Abstract]:Rolling bearing is one of the widely used components in rotating machinery, and its safe operation is of vital importance to the whole equipment. On the basis of summarizing the existing fault diagnosis technology of rolling bearing, this paper uses modern signal processing technology to study the feature extraction, early fault detection and early fault location identification of rolling bearing. The main contents of this paper are as follows: (1) The feature extraction method based on multi-scale symbol dynamic entropy (MSDE) based on multi-scale symbol dynamic entropy (MSDE) is presented in this paper. Based on the symbol dynamic filtering (SDF), the symbol dynamic entropy constructs a state pattern matrix and a state transition matrix of the symbol sequence, thereby greatly preserving the fault information of the vibration signal. Through the four simulation signal tests, it is verified that the symbolic dynamics entropy can effectively distinguish the change of signal dynamics and has higher computational efficiency. According to the concept of multi-scale analysis, a feature extraction method of MSDE is proposed, which can describe the complexity of signal at different scales. At the same time, the maximum correlation minimum redundancy is used to carry out pattern recognition and the least two-multiplication support vector machine to carry out pattern recognition, and the method has higher recognition precision and calculation efficiency based on the feature extraction method of the MSDE and realizes accurate discrimination of different fault positions of the rolling bearing. (2) Aiming at the problem of insufficient feature extraction ability of multi-scale fuzzy entropy (MFE) on fault damage degree of rolling bearing, a method of feature extraction based on hierarchical fuzzy entropy (HFE) is proposed. Aiming at the disadvantage of multi-scale analysis in MFE, a method for feature extraction of rolling bearing based on HFE is proposed. The hierarchical analysis can extract fault information hidden from low and high frequency parts of the signal at the same time. Through the test of white noise and pink noise, HFE can describe the change of time sequence complexity more accurately and comprehensively, and has better stability. then the characteristic is optimized by using the Laplacian value, and the recognition of different degrees of fault of the rolling bearing is realized by combining the binary tree support vector machine. Compared with MFE method, it is shown that HFE has good separability based on the feature vector extracted by HFE, which realizes the accurate differentiation of different fault degree of rolling bearing. (3) In order to realize the early fault detection of rolling bearing, the performance monitoring index M based on SDF is put forward. The calculation process of the monitoring index M is independent of the fault sample data of the bearing, and compared with the effective value and the kurtosis factor, The monitoring index M has high sensitivity and stability to early bearing failure. The monitoring index M is combined with the accumulation and the combination to complete the early fault alarm of the bearing, and the test of the full life experiment is accelerated through two bearings, and the method can effectively detect the early failure of the bearing. (4) The early fault diagnosis method of rolling bearing based on intrinsic feature-scale decomposition (ICD) and optimization quality factor is proposed in order to solve the problem of extremely weak fault feature in the early stage of rolling bearing. In order to reduce the interference of noise, an ICD decomposition method is proposed on the basis of summarizing the advantages of local mean decomposition (LMD) method and current time scale decomposition (ITD) method. The decomposition performance of ICD and LMD was compared with simulation signal. The results show that the ICD has higher decomposition precision and computational efficiency, and is more suitable for decomposing the vibration signals of rolling bearings. Aiming at the problem that the quality factor Q is difficult to be determined in the resonance sparse decomposition, the preference of the characteristic frequency ratio (CFR) to carry out the quality factor Q is introduced, and the optimal high-quality factor QH and the low-quality factor QL are found according to the maximum CFR value. Combining ICD with ORSSD, an identification method based on ICD and ORSSD for early fault location of rolling bearing is presented. Through simulation and experimental signal testing, this method can extract early fault features effectively and complete the early fault diagnosis of rolling bearing.
【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】：2017
【分类号】：TH133.33
本文编号：2312341
[Abstract]:Rolling bearing is one of the widely used components in rotating machinery, and its safe operation is of vital importance to the whole equipment. On the basis of summarizing the existing fault diagnosis technology of rolling bearing, this paper uses modern signal processing technology to study the feature extraction, early fault detection and early fault location identification of rolling bearing. The main contents of this paper are as follows: (1) The feature extraction method based on multi-scale symbol dynamic entropy (MSDE) based on multi-scale symbol dynamic entropy (MSDE) is presented in this paper. Based on the symbol dynamic filtering (SDF), the symbol dynamic entropy constructs a state pattern matrix and a state transition matrix of the symbol sequence, thereby greatly preserving the fault information of the vibration signal. Through the four simulation signal tests, it is verified that the symbolic dynamics entropy can effectively distinguish the change of signal dynamics and has higher computational efficiency. According to the concept of multi-scale analysis, a feature extraction method of MSDE is proposed, which can describe the complexity of signal at different scales. At the same time, the maximum correlation minimum redundancy is used to carry out pattern recognition and the least two-multiplication support vector machine to carry out pattern recognition, and the method has higher recognition precision and calculation efficiency based on the feature extraction method of the MSDE and realizes accurate discrimination of different fault positions of the rolling bearing. (2) Aiming at the problem of insufficient feature extraction ability of multi-scale fuzzy entropy (MFE) on fault damage degree of rolling bearing, a method of feature extraction based on hierarchical fuzzy entropy (HFE) is proposed. Aiming at the disadvantage of multi-scale analysis in MFE, a method for feature extraction of rolling bearing based on HFE is proposed. The hierarchical analysis can extract fault information hidden from low and high frequency parts of the signal at the same time. Through the test of white noise and pink noise, HFE can describe the change of time sequence complexity more accurately and comprehensively, and has better stability. then the characteristic is optimized by using the Laplacian value, and the recognition of different degrees of fault of the rolling bearing is realized by combining the binary tree support vector machine. Compared with MFE method, it is shown that HFE has good separability based on the feature vector extracted by HFE, which realizes the accurate differentiation of different fault degree of rolling bearing. (3) In order to realize the early fault detection of rolling bearing, the performance monitoring index M based on SDF is put forward. The calculation process of the monitoring index M is independent of the fault sample data of the bearing, and compared with the effective value and the kurtosis factor, The monitoring index M has high sensitivity and stability to early bearing failure. The monitoring index M is combined with the accumulation and the combination to complete the early fault alarm of the bearing, and the test of the full life experiment is accelerated through two bearings, and the method can effectively detect the early failure of the bearing. (4) The early fault diagnosis method of rolling bearing based on intrinsic feature-scale decomposition (ICD) and optimization quality factor is proposed in order to solve the problem of extremely weak fault feature in the early stage of rolling bearing. In order to reduce the interference of noise, an ICD decomposition method is proposed on the basis of summarizing the advantages of local mean decomposition (LMD) method and current time scale decomposition (ITD) method. The decomposition performance of ICD and LMD was compared with simulation signal. The results show that the ICD has higher decomposition precision and computational efficiency, and is more suitable for decomposing the vibration signals of rolling bearings. Aiming at the problem that the quality factor Q is difficult to be determined in the resonance sparse decomposition, the preference of the characteristic frequency ratio (CFR) to carry out the quality factor Q is introduced, and the optimal high-quality factor QH and the low-quality factor QL are found according to the maximum CFR value. Combining ICD with ORSSD, an identification method based on ICD and ORSSD for early fault location of rolling bearing is presented. Through simulation and experimental signal testing, this method can extract early fault features effectively and complete the early fault diagnosis of rolling bearing.
【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】：2017
【分类号】：TH133.33
【相似文献】
相关博士学位论文 前4条
1 李永波;滚动轴承故障特征提取与早期诊断方法研究[D];哈尔滨工业大学;2017年
2 王晓龙;基于振动信号处理的滚动轴承故障诊断方法研究[D];华北电力大学(北京);2017年
3 刘尚坤;基于振动信号处理的旋转机械故障诊断方法研究[D];华北电力大学(北京);2017年
4 路振勇;航空发动机转子系统的动力学建模及非线性振动研究[D];哈尔滨工业大学;2017年
相关硕士学位论文 前2条
1 马艳丽;基于全矢MEMD的滚动轴承状态退化研究[D];郑州大学;2017年
2 金兵;基于信息融合与VPMCD的滚动轴承智能诊断研究[D];郑州大学;2017年
，本文编号：2312341
本文链接：https://www.wllwen.com/jixiegongchenglunwen/2312341.html
