基于时变奇异谱的往复压缩机故障特征提取方法研究

发布时间：2018-07-20 10:30

【摘要】：在石油化工领域中,往复压缩机主要负责油气炼化和天然气、乙烯等不稳定气体的运输工作,其零部件易发生腐蚀破坏。除此之外,往复压缩机的构造复杂,并经常处于连续工作状态,因此许多易损零部件易发生疲劳破坏,寿命很短,导致故障发生频率高,故障类型复杂多样。为降低事故发生率,确保工作人员人身安全,降低设备的购买、维修成本,往复压缩机故障检测与诊断技术成为了人们研究的热点之一。针对往复压缩机振动加速度信号的非线性、非平稳等特性,将时间信息引入多重分形理论,提出了一种基于时变奇异谱的往复压缩机故障特征提取方法,用以描述振动信号的整体和细节信息,并利用支持向量机(SVM)进行模式识别和分类,结果表明该方法能够更加详细精准地表达故障信息,有利于提高故障诊断的精确度。首先,对往复压缩机故障诊断技术进行概括、研究和对比,提出本文故障特征提取方法的研究思路;对智能模式识别方法研究现状进行概括、研究和对比,为验证特征提取方法的效果提供了方法和依据。然后,对往复压缩机的基本结构、主要零部件、工作原理和工作循环过程进行分析,总结故障形式,并研究其故障机理,并根据工作过程中关键零部件的受力状态建立力学模型;利用相空间重构理论,计算出能够定量表征混沌性的关联维数,来说明其振动信号具有混沌性特征。接着,介绍分形及多重分形理论和算法,阐述多重分形谱中重要谱参数在往复压缩机故障信号分析中代表含义;将时间信息引入多重分形理论,建立时变奇异谱理论模型,并提出根据往复压缩机的工作过程来实现时变奇异谱的计算方法;对支持向量机分类理论进行研究,根据所提取故障特征向量的特点,选择“一对多”分类法为基础建立SVM模型,并对其中参数进行优选和设置。最后,将基于时变奇异谱的故障特征提取方法应用于D122往复压缩机的故障诊断中。设定往复压缩机的诊断流程,包括振动加速度信号采集,小波分解结合LMD分解的信号降噪处理,时变奇异谱提取故障特征向量和利用SVM分类器进行故障类型分类。分类结果表明,气阀故障分类准确率达到100%,轴承故障分类准确率达到93%,验证了基于时变奇异谱方法提取故障特征的有效性,说明采用该方法提取的特征向量能够准确区分往复压缩机的主要故障类型。
[Abstract]:In the field of petrochemical industry, reciprocating compressors are mainly responsible for oil and gas refining and transportation of unstable gases such as natural gas and ethylene. In addition, the reciprocating compressor is complex in structure and often in continuous working state. Therefore, many vulnerable parts are prone to fatigue damage, life is very short, resulting in high frequency of fault occurrence and complex fault types. In order to reduce the incidence of accidents, ensure the personal safety of staff, reduce the purchase of equipment, maintenance costs, reciprocating compressor fault detection and diagnosis technology has become one of the hot spots. In view of the nonlinear and non-stationary characteristics of vibration acceleration signal of reciprocating compressor, a method of fault feature extraction based on time-varying singular spectrum is proposed by introducing time information into multifractal theory. It is used to describe the whole and detailed information of vibration signal, and the support vector machine (SVM) is used for pattern recognition and classification. The results show that the proposed method can express the fault information in more detail and accurately, which is helpful to improve the accuracy of fault diagnosis. First of all, the fault diagnosis technology of reciprocating compressor is summarized, studied and compared, and the research train of thought of fault feature extraction method in this paper is put forward, and the research status of intelligent pattern recognition method is summarized, studied and compared. It provides the method and basis for verifying the effect of feature extraction method. Then, the basic structure, main parts, working principle and working cycle process of reciprocating compressor are analyzed, the fault forms are summarized, and the fault mechanism is studied. The mechanical model is established according to the stress state of the key parts in the working process, and the correlation dimension which can represent chaos quantitatively is calculated by using the theory of phase space reconstruction, which shows that the vibration signal is chaotic. Then, the fractal and multifractal theory and algorithm are introduced, the meaning of important spectral parameters in multifractal spectrum is expounded in the fault signal analysis of reciprocating compressor, the time information is introduced into multifractal theory, and the time-varying singular spectrum theory model is established. According to the working process of reciprocating compressor, the calculation method of time-varying singular spectrum is put forward, and the classification theory of support vector machine is studied. The SVM model is built on the basis of "one to many" classification, and the parameters are optimized and set. Finally, fault feature extraction method based on time-varying singular spectrum is applied to fault diagnosis of D122 reciprocating compressor. The diagnosis flow of reciprocating compressor is set up, including vibration acceleration signal acquisition, wavelet decomposition combined with LMD decomposition signal de-noising, time-varying singular spectrum extraction of fault feature vector and classification of fault type using SVM classifier. The classification results show that the accuracy of gas valve fault classification is 100 and bearing fault classification accuracy is 933. The validity of fault feature extraction based on time-varying singular spectrum method is verified. It shows that the feature vectors extracted by this method can accurately distinguish the main fault types of reciprocating compressors.
【学位授予单位】：东北石油大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TH45

【参考文献】