基于免疫阴性选择算法的转子故障数据分类方法研究

发布时间：2018-05-01 00:30

本文选题：转子系统 + 信息熵　；参考：《兰州理工大学》2013年硕士论文

【摘要】：旋转机械是工业部门中应用最为广泛的一类机械设备,其核心部件为转子-轴承系统。利用振动信号对转子-轴承系统的运行进行实时监测、分析与诊断,是保证旋转机械稳定、高效运作的重要措施。但随着旋转机械设备结构的大型化和工作环境的复杂化,传统的故障诊断方法已不能满足现代转子系统故障分类的需求,故智能故障分类方法在转子故障诊断研究中占有愈加重要的地位。然而,现有的智能故障分类方法很难解决通用性与高效性之间的矛盾,且在处理异常故障时还存在诸多问题。基于此,本研究以转子试验台模拟的故障数据为研究对象,借鉴生物免疫调节理论,研究了免疫阴性选择算法在转子故障数据分类中的应用。并且采用信息熵方法来定量的对故障信息进行特征提取,进而对数据进行了归一化处理,最后设计了适合典型转子故障识别的分类器。开展的具体研究工作与获得的研究结论如下： 1)在转子实验台上模拟了四种典型故障,分析了四种故障的机理。在此基础上分析了信号在时域的奇异谱熵、频域的功率谱熵、时频域的小波能谱熵和小波空间谱熵,并计算了四种故障信号的熵带,以四类谱熵为原始数据,对数据进行归一化处理,并建立了训练样本集和测试样本集。 2)分析了Forrest阴性选择算法、RNS算法与V-detector算法产生检测器的优缺点,进而对V-detector算法产生检测器阶段进行了两方面的改进：①改变接受和拒绝零假设的条件,使生成检测器的数目与预期覆盖率没有直接关系,在覆盖率提高时,检测器数目没有明显增加；②对生成的检测器集进行优化,使检测器间的重叠覆盖现象得到改善,且实现了降低“黑洞的数目”的目的。 3)针对V-detector算法在检测阶段只能识别自我和非我的缺陷,借鉴SVM中“有向无环图分类法(directed acyclic gragh)",设计出能够识别多种转子故障的分类器,并对比了改进后V-detector算法与V-detector算法的分类效果,表明前者的分类效果更优。 4)基于改进后V-detector算法流程,开发了一套基于MATLAB GUI的转子故障数据分类系统。此系统由三个子系统组成。子系统一：实现对振动信号的消噪分析,频谱分析,轴心轨迹分析等；子系统二：实现熵值数据的归一化；子系统三：实现两种算法分类的效果对比。在转子实验台上的应用效果理想。研究表明,免疫阴性选择算法作为人工免疫算法的核心算法,其在故障诊断辨识中的应用具有很大的研究空间和研究价值,该算法的思想为提高智能故障诊断质量提供了新思想和新方法。
[Abstract]:Rotating machinery is the most widely used type of mechanical equipment in the industrial sector, its core component is rotor-bearing system. Using vibration signals to monitor, analyze and diagnose the rotor-bearing system in real time is an important measure to ensure the stable and efficient operation of the rotating machinery. However, with the large-scale structure of rotating machinery and complicated working environment, the traditional fault diagnosis method can not meet the needs of modern rotor system fault classification. Therefore, intelligent fault classification plays an increasingly important role in rotor fault diagnosis. However, the existing intelligent fault classification methods are difficult to solve the contradiction between universality and efficiency, and there are still many problems in dealing with abnormal faults. Based on this, the application of immune negative selection algorithm in rotor fault data classification is studied based on the theory of biological immune regulation. And the information entropy method is used to quantitatively extract the fault information, and then the data is normalized. Finally, a classifier suitable for the typical rotor fault identification is designed. The specific research work carried out and the conclusions obtained are as follows: 1) four typical faults are simulated on the rotor test bench, and the mechanism of the four faults is analyzed. On this basis, the singular spectral entropy in time domain, power spectral entropy in frequency domain, wavelet spectrum entropy in time-frequency domain and wavelet space spectral entropy in time-frequency domain are analyzed. The entropy bands of four kinds of fault signals are calculated, and the four kinds of spectral entropy are taken as the original data. The data is normalized and the training sample set and the test sample set are established. 2) the advantages and disadvantages of Forrest negative selection algorithm and V-detector algorithm generation detector are analyzed, and then two improvements are made to the stage of V-detector algorithm generating detector: 1, which changes the condition of accepting and rejecting zero hypothesis. The number of generated detectors is not directly related to the expected coverage. When the coverage increases, the number of detectors is not significantly increased and the generated detector sets are optimized, so that the overlap coverage between detectors is improved. The aim of reducing the number of black holes is achieved. 3) aiming at the defect that V-detector algorithm can only recognize self and non-self in detection stage, a classifier which can identify many kinds of rotor faults is designed by referring to "directed acyclic graghs" in SVM. The classification effect of the improved V-detector algorithm and the V-detector algorithm is compared, which shows that the former has better classification effect. 4) based on the improved V-detector algorithm, a rotor fault data classification system based on MATLAB GUI is developed. The system consists of three subsystems. Subsystem 1: noise reduction analysis of vibration signal, spectrum analysis, axis trajectory analysis, etc.; Subsystem 2: normalization of entropy data; Subsystem 3: comparison of the results of the two algorithms. The application effect on the rotor test bench is ideal. The research shows that the immune negative selection algorithm is the core algorithm of artificial immune algorithm, and its application in fault diagnosis and identification has great research space and research value. The idea of this algorithm provides a new idea and method for improving the quality of intelligent fault diagnosis.
【学位授予单位】：兰州理工大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TH165.3;TP311.13

【参考文献】