【作者】 由大伟；

【导师】 段玉波；

【作者基本信息】 大庆石油学院 ， 电力电子与电力传动， 2004， 硕士

【摘要】 市场的迫切需求促进了机械故障珍断技术的迅猛发展，故障诊断技术发展至今，已经提出了大量的诊断方法，但是其实际应用成果显得非常不足，能在工业装置上实际应用的还不多。在故障诊断领域，还有许多问题亟待解决。目前，往复泵泵阀故障诊断需要解决的两个关键问题是有效提取往复泵工作时非平稳时变信号中的故障特征和将故障特征准确分类。 故障诊断常用的方法是以泵缸体上的振动信号作为系统特征信号来提取故障特征向量。这种振动诊断技术虽然取得了一定的成果，但是在多个泵阀同时发生故障的场合，这种方法遇到了无法解决的难题，使之不得不求助于粗集理论、遗传算法等数据处理方法来分辩故障类型和判断故障具体发生在哪一个泵阀上。 为此，本文创造性地提出以常见的压力信号(阀箱内的压力)作为系统特征信号来提取故障特征向量的方法。这种方法信号测取简单、处理方便，有着振动信号方法无法比拟的优点。文中利用时域中的相关分析和频域中的功率谱分析、小波包分析技术提取了故障特征向量，且各故障之间的特征区分明显，充分验证了此方法的有效性。此方法的优点在于特征信号取自于阀箱内的压力，不易受到外部环境的干扰，适用于多个泵阀同时发生故障的情形。同时，文中还构造了三层的前向神经网络，以小波包分析提取的故障特征向量作为网络的训练样本数据，采用加惯性因子、共轭梯度法和迭代过程中改变学习率的反向传播算法来对神经网络进行训练，并采用试验的方法调整神经网络的初始值。在确定了神经网络的结构和参数后，经检验数据验证训练后的神经网络所得的网络结构和参数是合理的。 本文以故障诊断系统为核心开发了往复泵泵阀故障的智能诊断系统。数据采集系统用VC++进行开发，其主要功能是进行现场数据的采集和数据库管理。故障诊断系统以Matlab作为软件平台，用小波包对数据进行处理并提取故障特征向量，并利用神经网络技术实现泵阀的故障诊断功能。更多还原

【Abstract】 Urgent requirement of market has promoted the improvement of the technology of mechanical fault diagnosis (FD). So far, many methods have been proposed in the field of fault diagnosis. But practical application is insufficient, and less in the application of industrial devices. Two key issues of fault diagnosis for the pump valves of reciprocating pump are extracting the fault feature information of stationary process efficiently from system feature signals and identifying the specific faults correctly with analysis of causes.At present, usual method of FD is measuring vibration signal as system feature signals for fault characteristic eigenvectors pickups. Although the technology of vibration diagnosis has acquired some achievement, on the occasion of having several faults , the method has some trouble that cannot be solved . To solve this problem, some methods i.e. rough-set theory and genetic algorithm have to be used to identify specific fault and decide which pump valve has fault.The paper creatively proposes the idea ofusing ordinary pressure signals;(pressure in valve boxes) as system feature signal to pickup fault characteristic eigenvectors and verifies the correction of the method through self-correlation analysis in the time-zone and power spectral and wavelet packet analysis in the frequency-zone. What’s more, it is not easy to be disturbed by outside environment for requiring the signals from the inside of reciprocating pump cylinder. The method has special obvious advantages to diagnosis the faults when several faults exist simultaneously. The paper constructs a three-layer forward neural network to diagnosis the fault and trains the network with characteristic eigenvectors extracted through the wavelet packet analysis , when training the net using the method of adding inertia item and the BP algorithm of conjugate gradient method, at the same time adapt the initial data of the network through the test. After confirm the construction and parameters of the net, the result is valid and justified through the verification using the test sample data.The paper develops the intelligent fault diagnosis system regarded/the fault diagnosis factor as the center of the whole system. The main function of the data collection system developed by VC++ program is collecting field data and managing the data. And the fault diagnosis system on the basis of Matlab software platform pick up the fault characteristic eigenvectors through the method of wavelet packet analysis and realize the fault diagnosis using the technology of neural network.更多还原