基于数据驱动的转子故障特征信息建模方法研究

发布时间：2019-06-07 14:57

【摘要】：随着信息科学技术的快速发展,计算机技术与各种智能仪表在机械装备的监测中获得了广泛应用,反映系统运行状态的海量过程数据被采集并存储下来,但是存在着“数据丰富,但信息匮乏”的缺陷。利用这些离、在线运行数据,采用智能数据分析算法,建立起一种能够科学描述机械装备运行状态的量化特征模式,实现对机械装备信息化技术发展中机器故障的智能自动辨识,具有非常积极地奠定作用。为此,本文从反映机械装备信息特点的量化特征模式构造出发,基于数据挖掘的知识发现原理,围绕着利用智能数据分析工具,开展基于数据驱动的故障监测诊断方法的研究工作。通过采用数据驱动故障诊断的常用算法,重点对反映机械装备信息特点的量化特征模式构造及不均衡故障数据集的分类方法进行探讨,为量化描述机组运行状况,实现对故障模式的在线诊断提供依据。本文主要研究工作包括以下几个方面的内容： (1)介绍多域特征的提取方法,对KPCA-SVM的故障数据分类方法进行研究,并对该方法在转子系统故障诊断中的应用情况进行探讨。 (2)针对故障信息的量化特征描述问题,提出了一种特征选择与特征信息融合的加权KPCA方法。首先对单个通道的振动信号提取时域、频域和时频域的多域特征参量,通过特征选择方法筛选出利于故障模式辨识的敏感特征；其次融合多通道的敏感特征,得到融合特征向量；然后采用加权KPCA方法提取出融合特征向量的核主成分。通过SVM分类器的实验验证情况表明,该算法可有效辨识出不同故障类型。 (3)针对不均衡故障数据分类精度低,辨识效率不高的问题,提出了一种基于滑动窗口相似性因子分析方法。该方法引入滑动窗口技术,通过分析目标数据与历史数据的PCA相似性因子,从旧的过程数据中筛选出与诊断目标相似的数据,构成待选数据池；然后采用距离相似性因子,从待选数据池中选择出与目标数据最相似的数据用于辅助训练。将该方法用于转子故障的不均衡数据分类中,在不同偏斜率下采用KPCA-SVM方法进行故障分类。结果表明,该方法可有效地改善分类决策边界,降低由样本不均衡而引起的误诊断率。 (4)基于测试测量仪器发展中“虚拟化”、“控件化”的发展方向,对智能控件化虚拟仪器仪表的系统软件构架及其应用开展研究。在C#软件平台下,提出了转子系统故障数据采集的实验系统方案,尝试对以下模块开展设计：波形显示模块、电机控制模块、数据存储模块。
[Abstract]:With the rapid development of information science and technology, computer technology and various intelligent instruments have been widely used in the monitoring of mechanical equipment. The massive process data reflecting the running state of the system has been collected and stored. However, there are the defects of "rich data, but lack of information". Based on these off-line operation data and intelligent data analysis algorithm, a quantitative feature model which can scientifically describe the running state of mechanical equipment is established. The realization of intelligent automatic identification of machine faults in the development of mechanical equipment information technology plays a very active role. Therefore, starting from the construction of quantitative feature pattern reflecting the characteristics of mechanical equipment information, and based on the principle of knowledge discovery in data mining, this paper revolves around the use of intelligent data analysis tools. The research work of fault monitoring and diagnosis method based on data drive is carried out. By using the common algorithm of data-driven fault diagnosis, the construction of quantitative feature mode reflecting the characteristics of mechanical equipment information and the classification method of unbalanced fault data set are discussed in order to quantitatively describe the operation status of the unit. The online diagnosis of fault mode is realized. The main research work of this paper includes the following aspects: (1) the extraction method of multi-domain features is introduced, and the fault data classification method of KPCA-SVM is studied. The application of this method in rotor system fault diagnosis is also discussed. (2) aiming at the problem of quantitative feature description of fault information, a weighted KPCA method for feature selection and feature information fusion is proposed. Firstly, the multi-domain feature parameters of time domain, frequency domain and time-frequency domain are extracted from the vibration signal of a single channel, and the sensitive features conducive to fault pattern identification are screened out by feature selection method. Secondly, the fusion feature vector is obtained by fusion of the sensitive features of multi-channel, and then the kernel principal components of the fusion feature vector are extracted by weighted KPCA method. The experimental results of SVM classifiers show that the algorithm can effectively identify different fault types. (3) in order to solve the problems of low classification accuracy and low identification efficiency of unbalanced fault data, a similarity factor analysis method based on sliding window is proposed. In this method, the sliding window technology is introduced, and the PCA similarity factor between the target data and the historical data is analyzed, and the data similar to the diagnostic target is screened out from the old process data to form the data pool to be selected. Then the distance similarity factor is used to select the data most similar to the target data from the data pool to be selected for auxiliary training. This method is applied to the unbalanced data classification of rotor faults, and the KPCA-SVM method is used for fault classification under different slope. The results show that this method can effectively improve the boundary of classification decision and reduce the misdiagnosis rate caused by the imbalance of samples. (4) based on the development direction of "virtualization" and "control" in the development of test and measurement instruments, the system software architecture and application of intelligent control virtual instruments are studied. Under the platform of C # software, the experimental system scheme of rotor system fault data acquisition is put forward, and the following modules are designed: waveform display module, motor control module, data storage module.
【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】