转子故障数据CRFS分类方法研究与测试系统开发

发布时间：2018-05-04 03:32

本文选题：特征选择 + 智能诊断　；参考：《兰州理工大学》2013年硕士论文

【摘要】：旋转机械是电力、化工等行业生产中的关键设备,对旋转机械设备进行状态监测和故障诊断以保证设备的安全、可靠运行具有重要的经济价值。随着信息技术、现代控制理论和人工智能的发展,自动化和智能化逐渐成为故障诊断系统的发展方向。然而故障诊断技术发展到现在,仍然面临着支撑理论存在局限性、规则获取困难、诊断模型难以实际应用等各种问题。本研究针对在转子故障诊断中,特征的选择缺乏可靠的依据、多通道信息融合造成的特征冗余问题以及故障识别精度不高的问题,研究了基于K-W检验特征选择与CRFS识别的故障诊断模型在转子故障诊断中的应用,开辟了转子故障诊断的新途径。本文的主要工作和研究结论如下： 1)针对常用的故障特征种类繁多,对其选择缺乏可靠依据和针对性不强的问题,通过对不同故障类型样本同一特征的数据区间进行分析,筛选出了数据区间重叠较小的具备较强表征转子不同运行状态能力的特征。以此为依据,建立了转子原始故障特征数据集。 2)针对多通道监测、数据采集能够更加全面的反映转子运行状态,但对故障特征集的建立造成了维数过高、不同类别子集间可分性差的问题,研究了K-W检验特征选择方法,简化了算法并与主成分分析(PCA)进行对比分析,实验验的证结果表明,K-W检验特征选择的结果空间聚类紧致,降维效果和算法复杂度均优于主成分分析法。 3)将条件随机场模型(CRFS)引入到转子故障诊断中,并将经过特征优化后的数据输入CRFS进行参数学习和样本训练,实验结果表明条件随机场稳定性好且识别准确率高,大量特征叠加对识别精度的影响并不大。通过与HMM模型对比分析表明,在条件复杂、故障种类较多以及特征相似时HMM模型诊断率会明显下降且训练较慢,而CRFS所有特征可以进行全局归一化,能够求得全局的最优解,在多故障诊断中表现出了优良的性能。 4)利用虚拟仪器技术开发了一套集实时状态监测、预警、振动信号分析、振动数据和特征数据存储、报表自动生成及远程网络访问等功能于一体的机械振动自动测试分析信息系统。在实验研究以及工程应用中均取得了较好的效果。研究表明,通过特征降维、特征选择能够获得具备较好的类别可分性的故障特征向量,故如何在智能故障诊断算法研究中获得新突破,及如何将智能诊断算法合理地嵌入于自动化测控系统,将是故障诊断领域开展研究工作的重要方向。
[Abstract]:Rotating machinery is the key equipment in the production of electric power and chemical industry. It is of great economic value to monitor and diagnose the status of rotating machinery and equipment to ensure the safety and reliable operation of the equipment. With the development of information technology, modern control theory and artificial intelligence, automation and intelligence are becoming the developing direction of fault diagnosis system. However, with the development of fault diagnosis technology, it still faces many problems, such as the limitation of supporting theory, the difficulty of obtaining rules, the difficulty of practical application of diagnostic model, and so on. This study aims at the lack of reliable basis for feature selection in rotor fault diagnosis, the problem of feature redundancy caused by multi-channel information fusion and the problem of low fault identification accuracy. The application of fault diagnosis model based on K-W test feature selection and CRFS identification in rotor fault diagnosis is studied, which opens a new way for rotor fault diagnosis. The main work and conclusions of this paper are as follows: 1) aiming at the problems of the variety of common fault features, the lack of reliable basis for their selection and the lack of pertinence, the data interval of the same feature of different fault types is analyzed. The characteristics of the rotor with little overlap in the data interval are selected, which can represent the different running states of the rotor. Based on this, the original fault feature data set of rotor is established. 2) in view of multi-channel monitoring, data acquisition can reflect rotor running state more comprehensively, but the establishment of fault feature set leads to the problems of high dimension and poor separability among different subsets, so the K-W test feature selection method is studied. The algorithm is simplified and compared with the principal component analysis (PCA). The experimental results show that the result space of the feature selection of K-W test is compact, the dimension reduction effect and the complexity of the algorithm are better than that of the principal component analysis. 3) the conditional random field model (CRFs) is introduced into rotor fault diagnosis, and the data after feature optimization is input into CRFS for parameter learning and sample training. The experimental results show that the conditional random field has good stability and high recognition accuracy. The superposition of a large number of features has little effect on the recognition accuracy. Compared with HMM model, it is shown that the diagnostic rate of HMM model decreases obviously and the training is slower when the conditions are complex, there are many kinds of faults and the characteristics are similar, while all the features of CRFS can be normalized globally and the global optimal solution can be obtained. It shows excellent performance in multi-fault diagnosis. 4) A set of real-time state monitoring, early warning, vibration signal analysis, vibration data and feature data storage are developed by using virtual instrument technology. Automatic test and analysis information system for mechanical vibration with automatic report generation and remote network access. Good results have been obtained in both experimental research and engineering application. The research shows that feature selection can obtain fault feature vector with better classification by feature reduction, so how to make a breakthrough in intelligent fault diagnosis algorithm. And how to embed the intelligent diagnosis algorithm into the automatic measurement and control system reasonably will be an important research direction in the field of 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

【参考文献】