高压隔膜泵的早期微弱故障特征提取及诊断方法研究

High-pressure diaphragm pump is a key power equipment in the slurry transportation pipeline system, the working environment of the solid-liquid two-phase medium and high pressure always leads to faults such as severe aging of its valve body; the breakdown of the valve chest and such. . Aiming to solve the problems including the weak response of early faults caused by the complex working condition of diaphragm pump, and the serious interference by the noise from the low frequency components. This research will focus on studying the theory and technology of fault diagnosis of low-speed and high-pressure diaphragm pump. Firstly, the multi-sensor information fusion technology based on optimization discount coefficient D-S evidence theory will be applied attempting to resolve the problem of large quantity of testing node; the faulty features will be extracted from signals by using multi-feature modeling method in the time-frequency domain, so that the fault diagnosis model can be established by training variable prediction models. Meanwhile, the method of extracting features from weak signals in frequency domain will be studied based on singular value decomposition (SVD) and Hilbert-Huang transform. The method of extracting features from weak signals under the condition of intensive noise will then be discussed by using the translation invariant of Dual-Tree Complex Wavelet Transform (DT-CWT) and the nonlinear dimension reduction of manifold learning. This project will provide a new thinking and theoretical foundation for the further study of fault diagnosis of weak signal from large-scale machinery with reciprocating and fixed axis motion under complex working environment.

高压隔膜泵是矿物管道输送系统的核心动力设备，特殊的固液两相流工作介质、高压的工作环境经常导致阀体严重磨损和阀室击穿等故障。针对高压隔膜泵复杂工况导致的早期故障信号微弱、特征频率成份受噪声污染严重等问题，本项目将深入研究高压隔膜泵早期微弱故障诊断的理论与方法。首先针对故障敏感测点较多的问题，综合考虑聚焦程度和冲突程度，研究基于优化折扣系数D-S证据理论的多传感器信息融合技术；研究采用时、频域联合多特征建模方法提取故障特征信号，通过改进变量预测模型建立故障诊断模型。同时以奇异值分解和Hilbert-Huang变换为基础，深入研究频域中弱信号的特征提取方法 ；利用对偶树复小波变换的平移不变性和流形学习的非线性降维思想，研究时域中强噪声背景下弱信号的特征提取方法。本项目的研究成果将为以固液两相流为工作介质，作往复、定轴运动大型机构的早期微弱故障诊断提供新思路。

深入分析了高压隔膜泵工作原理及单向阀的结构组成，对其结构进行数学建模，并分析了关键参数对运行性能及故障的影响。研究了高压隔膜泵单向阀破坏机理，获取了单向阀故障类型。. 针对高压隔膜泵早期微弱故障的振动信号故障特征不明显及难以提取的问题展开研究工作。提出一种基于微分经验模态分解（DEMD）和K-L散度的单向阀早期故障检测方法，提高高频成分的振幅比，使微弱高频成分在后续分解中更易提取；鉴于单纯的时频域方法在检测出早期故障时有一定的局限性，提出一种优化随机共振（SR）和DEMD的单向阀早期故障检测方法；针对单向阀微弱故障阶段冲击特征较微弱，且故障振动信号含有多分调幅调频信号的问题，提出了基于VMD和Wigner-Ville的单向阀微弱故障诊断方法；考虑到单向阀的故障特征信息分布在多尺度上，单一尺度的改进排列熵难以全面提取的问题，提出了一种基于改进多尺度排列熵（IMPE）的单向阀故障诊断方法。最终实现了一系列针对不同场景的微弱故障特征提取方法。. 针对特征提取后的故障分类问题，开展分类模型的研究。变分模态分解（VMD）方法在降噪同时也削弱了有用信号和变量预测模型（VPMCD）分类器只能选择单一模型导致分类结果片面的问题，提出基于优化随机共振VMD和多模型变量预测模型（MFVPMCD）的单向阀微弱故障诊断方法；针对深度置信网络(DBN)分类识别之前需要将特征矩阵重新排列成向量，会丢失时频图像二维平面上的特征，且DBN参数多、计算量大的问题，结合卷积神经网络(CNN)的优点，提出基于广义S变换和卷积深度置信网络(CDBN)的单向阀故障诊断模型。. 采用现场采集数据对上述方法进行验证，结果显示对故障的识别率都超过90%，表明能诊断单向阀的微弱故障。以此为基础构建了故障诊断系统，并用于工业现场的测试。