基于多特征量提取和极限学习机的轴承故障诊断方法研究

发布时间：2018-02-03 10:13

本文关键词： 滚动轴承排列熵极限学习机最大相关最小冗余故障诊断　出处：《昆明理工大学》2017年硕士论文　论文类型：学位论文

【摘要】：近年来,我国科技事业得到了进一步提升,信息技术的发展带动了工业化的进步,使得机械化生产日益普及。在机械的生产过程与应用中,机械系统的非计划停机与故障停机会给企业的生产发展及经济效益带来严重的损害,甚至存有人身安全隐患,造成安全事故。在旋转机械设备的众多零部件中,地位最高但又是最容易损坏的部件就是轴承。轴承是否能正常运行直接影响整个机械系统的性能与寿命,所以对滚动轴承故障诊断的方法进行研究具有重大意义。滚动轴承振动信号多表现为非平稳性与非线性,同时,受周边设备传来的噪声以及故障信息中所含有的短期冲击成分的影响,使得故障特征提取有一定的难度。特征提取不全面、特征值不明显,均会影响轴承的故障识别精度,带来误判甚至漏判现象。针对此问题,本文提出了基于多特征量提取和极限学习机的轴承故障诊断方法研究。论文的主要研究工作如下:(1)研究了滚动轴承故障的形成原因以及其振动频率,具体分析了轴承的结构和动力学特性,提出了几种典型振动的基本参数。对滚动轴承故障数据采集系统进行学习和研究,获得不同类型下的轴承振动数据,最后,以Logistic典型非线性系统为验证对象,进而验证排列熵用于检测非线性系统的动力学突变行为的可行性。(2)研究了基于排列熵和极限学习机的轴承故障类型诊断方法。通过使用多分辨奇异值分解对原始加速度振动信号进行分解,获得三层不同细节分量D1～D3,进一步结合排列熵在信息提取方面的优势,构造能表征原始加速度振动信号故障特征的特征向量,最后采用极限学习机的方法进行轴承故障类型识别,验证了该方法的可行性和有效性。(3)研究了基于优化MRMR和极限学习机的轴承故障类型诊断方法。对经过多分辨SVD降噪处理的原始加速度振动信号分别求取时域、频域和时-频域特征量构建混合域特征集,在特征选取上,采用加权MRMR的特征选取方式,以极限学习机分类正确率为依据,从含有18个特征的混合域中最后选取出8个最优特征向量。实验数据分析显示,故障辨别精度可达到97.5%,证明该方法可以有效的实现轴承故障类型诊断。以实际滚动轴承故障数据为例进行对比分析,结果表明基于优化MRMR和极限学习机的轴承故障智能诊断方法比基于排列熵和极限学习机的故障诊断方法的故障识别率高,这是由于相比于特征值单一且特征不明显的排列熵,混合域特征集更能多方面表征振动信号的内在特征,且经过优化的MRMR算法准则选出来的特征子集是最具代表性的。
[Abstract]:In recent years, China's science and technology has been further promoted, the development of information technology has led to the progress of industrialization, making mechanized production increasingly popular in the production process and application of machinery. The unplanned downtime and malfunction of the mechanical system will bring serious damage to the production development and economic benefit of the enterprise, even have the personal safety hidden trouble, cause the safety accident, in many parts and components of the rotating machinery and equipment. Bearing is the highest-ranking but most easily damaged component. Whether the bearing can run normally will directly affect the performance and life of the whole mechanical system. Therefore, it is of great significance to study the fault diagnosis method of rolling bearings. The vibration signals of rolling bearings are usually non-stationary and nonlinear, and at the same time. Affected by the noise from peripheral equipments and the short term impact components contained in the fault information, it is difficult to extract the fault features. The feature extraction is not comprehensive, and the feature value is not obvious. Will affect the bearing fault identification accuracy, leading to misjudgment or even miss the phenomenon, in view of this problem. In this paper, the method of bearing fault diagnosis based on multi-feature extraction and extreme learning machine is proposed. The main research work of this paper is as follows: 1) the cause of rolling bearing fault and its vibration frequency are studied. The structure and dynamic characteristics of the bearing are analyzed in detail, and the basic parameters of several typical vibration are put forward. The fault data acquisition system of rolling bearing is studied and studied, and the bearing vibration data of different types are obtained. Finally, the typical nonlinear Logistic system is used as the verification object. Furthermore, the feasibility of using permutation entropy to detect the dynamical catastrophe behavior of nonlinear systems is verified. The fault type diagnosis method of bearing based on permutation entropy and ultimate learning machine is studied. The original acceleration vibration signal is decomposed by using multi-resolution singular value decomposition. Three layers of different detail components D _ 1 and D _ 3 are obtained. Furthermore, combining the advantage of permutation entropy in information extraction, the eigenvector which can represent the fault characteristics of the original acceleration vibration signal is constructed. Finally, the bearing fault type is identified by the extreme learning machine. The feasibility and effectiveness of the method are verified. The fault type diagnosis method of bearing based on optimized MRMR and ultimate learning machine is studied, and the time domain of the original acceleration vibration signal after multi-resolution SVD de-noising is obtained respectively. In frequency domain and time-frequency domain, the feature sets in mixed domain are constructed. In feature selection, weighted MRMR is used to select features, and the classification accuracy of extreme learning machine is taken as the basis. Finally, 8 optimal feature vectors are selected from the mixed domain with 18 features. The experimental data show that the accuracy of fault identification can reach 97.5%. It is proved that this method can effectively realize the fault type diagnosis of bearing. Take the actual rolling bearing fault data as an example to carry on the contrast analysis. The results show that the intelligent fault diagnosis method based on optimized MRMR and LLM is higher than that based on permutation entropy and LLM. This is due to the fact that, compared with the permutation entropy with a single eigenvalue and no obvious feature, the feature set in the mixed domain can represent the intrinsic characteristics of the vibration signal in many ways. And the feature subset selected by the optimized MRMR algorithm criterion is the most representative.
【学位授予单位】：昆明理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TH133.33;TP181

【参考文献】