【作者】 钟鑫；

【导师】 高金吉；

【作者基本信息】 北京化工大学 ， 计算机应用技术， 2010， 硕士

【摘要】 近年来,随着现代制造工业的发展,各种机械设备越来越广泛的应用于石化、电力等行业,而且它们本身还不断向大型化、智能化、高速化和复杂化等方面发展,这些发展极大的推动了社会生产力的发展。然而,一旦这类设备发生严重故障,不仅会给企业带来经济损失,严重时甚至还会造成人员伤亡。因此,研究设备故障诊断技术,确保它们的安全稳定运行,降低生产成本,已成为现代企业管理的重要目标之一。本文在国家“十一五”科技支撑计划：“大型高参数高危险性成套装置长周期运行安全保障关键技术研究及工程示范”(项目编号：2006BAK02B02)的资助下开展研究,主要研究工作体现在以下几个方面：(1)介绍了课题研究的背景、目的和意义,阐述了设备性能退化评估和设备故障模式识别的发展现状,明确了需要解决的问题。(2)研究了信号处理、特征选择与特征提取技术,着重针对本研究中所涉及到的两种信号处理方法FFT和小波包分解,以及基于故障特征频率的特征选择方法和基于主成分分析(PCA)的特征抽取方法进行了研究。(3)详细研究了本文研究中的重要模型,即逻辑回归和高斯混合模型。针对逻辑回归,对其进行了理论概述以及基于极大似然估计法的参数获取研究；针对高斯混合模型,介绍了模型理论以及EM参数估计方法,另外,对于使用EM算法进行参数估计时所遇到的初值设定和高斯混合数的确定问题进行了一定的探索研究。(4)将逻辑回归和高斯混合模型引入到旋转机械设备的性能退化评估中。使用逻辑回归建立了滚动轴承运行性能退化评估模型,并用本实验室获取的实验数据和网上公开的CWRU数据进行了验证；使用高斯混合模型建立了离心压缩机运行性能退化状态评估模型,并用中国石油某炼化企业现场实际监测数据进行验证,取得良好效果。(5)将高斯混合模型引入旋转机械设备的故障模式识别研究中。首先使用传统的基于贝叶斯极大似然分类器的模式分类方法,并在此基础上提出一种改进的基于特征空间重合度计算的模式分类方法,最后利用CWRU滚动轴承数据进行验证,取得良好效果。更多还原

【Abstract】 In recent years, as modern manufacturing industry develops, various kinds of mechanical equipments have been widely utilized in petro-chemical, electricity industries and so on, while the equipments concerned have the trend towards large scale, intelligentization, high speed and integration, which greatly enhances the progress of the social productivity. However, if faults occur in these equipments, it will not only cause economic losses to the enterprises, but also will result in severe human casualties. As a result, to research equipment fault diagnosis technology and guarantee the safe and stable operation of the equipments has been of extreme significance to modern enterprises.This paper is supported by National Key Technologies R& D Program of China:(2006BAK02B02), and the main jobs of this study are as follows:(1) Introduced the background, goal and significance of this research, illustrated the current developing status of performance degradation assessment and fault pattern recognition method for equipments, and ascertained the problem to be solved.(2) Researched signal processing, feature selection and feature extraction technology. Signal processing methods including FFT, Wavelet Transform and Wavelet Packet Decomposition are paid special attention to. Meanwhile, feature selection method based on fault characteristic frequency and feature extraction method based on principal component analysis are both thoroughly investigated.(3) Investigated logistic regression and Gaussian mixture model that are theoretically essential in this paper. As for LR, its theoretical illustration and parameter estimation based on MLE (maximum likelihood estimation) are introduced; as for GMM, its theory and parameter estimation based on EM algorithm are introduced. In addition, this paper explored the problem of parameter initialization and optimal mixture number for GMM.(4) Introduced LR and GMM into the performance degradation assessment research of rotating machinery. LR is used to assess the performance of bearing, and the method is validated by both experimental data and CWRU data; GMM is utilized to assess the performance of centrifugal compressor, which is validated by data acquired in the field of a chemical refinery enterprise of Petro China. Both the two methods have gained favorable effect.(5) GMM is also introduced in the fault pattern recognition for rotating machinery. Firstly, traditional Bayesian classifier-based method is adopted, and then an improved method based on the overlap between GMMs is proposed, which have been validated by CWRU bearing data.更多还原