【作者】 刘斌；

【导师】 肖聚亮； 黄国勇；

【作者基本信息】 天津大学 ， 机械工程（专业学位）， 2016， 硕士

【摘要】 离心输油泵是原油管道输送系统提供动力支持的关键设备,其工作环境恶劣,系统结构复杂,调节操作频繁,具有较高的突发故障率。输油泵一旦发生故障而未被及时发现,将可能导致恶性停机事件。管线的停输可能引发管道泄漏及火灾爆炸等连锁安全事故,造成人员伤亡、环境破坏以及重大经济损失。因此,定期对离心输油泵开展切实有效的故障诊断研究,及时发现早期故障,避免发生恶性停机事件对保障整个原油运输系统的安全平稳运行具有十分重要的意义。本文首先对离心输油泵的基本结构和工作原理进行研究,根据文献调研分析输油泵常见的故障类型,并对输油站场离心泵故障进行调研,重点研究重大故障案例,寻找事故多发部位进行重点研究,然后对基于振动检测的故障诊断方法的基本原理进行了阐述,该方法的重点在于故障特征的识别,因此对输油泵常见故障的时频域特征进行分析归类,并用MATLAB仿真故障信号,得到故障时频域特征图。同时对输油站场根据经验得出的离心输油泵故障诊断要点进行汇总。寻找出现行的常用振动检测故障诊断的不足,提出相应的改进。改进主要针对振动信号降噪、故障特征提取和模式识别进行。1、阐述了奇异值分解降噪的基本原理,对降噪过程中有效秩阶次的选择和Hankel矩阵结构的确定进行了重点研究,将算法应用于转子和轴承故障振动信号降噪,得到信噪比明显的振动信号。2、提出了一种基于小波包能量熵的离心输油泵故障特征提取方法,对降噪后的振动信号进行时频分析,得到转子和轴承不同设定状态下的振动信号特征向量。3、利用人工神经网络对离心输油泵的常见转子/轴承故障进行故障模式识别,用振动信号的小波包能量熵特征参数作为网络输入,经过神经网络的训练和学习,实现了对离心输油泵的常见转子/轴承故障的诊断。更多还原

【Abstract】 Centrifugal pump is the key equipment of oil pipeline transportation system to provide power support.Due to harsh working environment and complicated structure of the centrifugal pumpand frequent operation of the transportation system,the sudden failure rate of the centrifugal pump is higher than average.Once the fault of oil pump was not found in time,it may cause malignantshutdown events.The shutdown of the system may cause chained safety accidents,such as pipeline leakage,fire,explosions and so on,which may cause casualties,major economic losses and environmental damage.Therefore,for the safety of the whole system,it is very important to carry out effective research on fault diagnosis of centrifugal oil pump regularly,to detect early faults in time and to avoid malignant shutdown events.On the one hand,the structure and principle of centrifugal oil pump are discussed.Then the forms and mechanism of common failures of pump are analyzed in detail.Moreover,investigation and case study about centrifugal pump failures in the oil station and are done to discover the most frequent failure parts.Then research is focus on these parts.On the other hand,the principle of vibration detetion based fault diagnosis method is studied.The method is focused on fault feature recognition.Therefore,time and frequency domin characteristics of common faults are analyzed and simulated.Then study on the key points of centrifugal pump fault disgnosis which is summarized according to the experience of pump stations to find out the weaknesses of the commonly used method.And in order to improve the method,particularly pay attention to de-noising method,feature extraction and pattern recognition.First,as the basic principle of noise reduction by singular value decomposition has been expounded,the determination of effective order and the structure of reconstruction matrix are researched significantly.After that,the algorithm is applied to noise reduction of vibration signals for rotor and bearing to obtain denoised signals which have high signal-to-noise ratio.Second,a method for feature extraction from denoised signal based on wavelet energy entropy has been presented.Vibration signals of Rotor and bearing obtained under different conditions are analyzed based on time-frequency analysis method.As a result,the feature vectors of denoised vibration signals are obtained.Third,patterns of common faultsof centrifugal pump rotor and bearing have been recognized by using artificial neural network.With obtained feature vectors inputted as the grid,the diagnosis of common faults of centrifugal pump rotor and bear has been realized through the training and learning abilities of back propagation neural network.更多还原