往复式高压隔膜泵单向阀状态监测及故障诊断研究
发布时间：2018-07-21 12:40
【摘要】：大型往复式高压隔膜泵是长距离、高扬程、高浓度矿浆管道输送的核心动力设备,它的工作状态直接影响企业生产效率。单向阀作为泵的核心机械零部件之一,需具有良好的快开、快关、密封性及承压性,比泵的其他部件更容易出现故障。此外,单向阀的运行状态与输送矿物的粒径级配、浆体流变特性、输送压力、泵的固有材质属性及安装等密切相关,致使单向阀的故障具有突发性、并发性、多源性、非平稳性和非线性等特点,大大增加了单向阀状态监测和故障诊断的难度。因此,从单向阀振动信号分析入手,选取有效的特征提取及故障诊断方法是单向阀运行状态监测及故障诊断研究的核心内容,具有重要理论研究价值及经济意义。本文围绕单向阀状态监测及故障诊断开展了如下研究工作：(1)提出了一种基于局部均值分解(Local Mean Decomposition, LMD)和包络解调的单向阀故障检测方法。单向阀故障振动信号通常表现为复杂的调幅调频信号,使得利用包络解调方法提取单向阀故障特征频率成为可能。但是,单向阀受环境噪声、耦合工况及其他激励源干扰等影响,其振动信号表现出明显的非线性,直接对其进行包络解调无法获得理想的效果。因此,提出基于LMD和包络解调的单向阀故障检测方法,先利用LMD将信号分解为一系列纯调幅调频信号——乘积函数(Production Function, PF);进而对PF分量进行包络解调以完成单向阀故障检测。(2)提出了一种基于多域混合特征极限学习机(extreme learning machine, ELM)的单向阀故障诊断方法。针对单一域特征无法完全描述单向阀运行状态、支持向量机(Support Vector Machine, SVM)和BP神经网络等模型优化参数多、速度慢等问题,结合多域混合特征和ELM的优势,提出基于多域混合特征ELM的单向阀故障诊断方法。提取单向阀振动信号时域、频域、小波域、TK (Teager Kaiser)域特征构建多域混合特征集,并引入核主元分析(Kernel Principal Component Analysis, KPCA)方法进行多域混合特征集的二次特征提取,消除特征冗余。最后基于二次特征提取后的多域混合特征集建立单向阀ELM故障诊断模型,完成单向阀故障诊断。(3)提出了一种基于小波包能量熵和模糊核极限学习机(fuzzy kernel extreme learning machine, F-KELM)的单向阀故障诊断方法。在讨论复杂非线性振动信号、样本分布不均衡及ELM隐含层神经元个数对ELM分类性能影响的基础上,引入小波包能量熵、核函数、模糊隶属函数建立小波包能量熵和模糊核极限学习机的故障诊断模型。通过滚动轴承和单向阀的实验对比分析,证实了方法能有效解决上述难题,提高了模型分类性能及泛化能力。(4)提出了一种基于多核代价敏感极限学习机(multi-kernel cost sensitive extreme learning machine, MKL-CS-ELM)的单向阀故障诊断方法。针对单一核函数分类器无法完全诠释分类决策函数、分类代价均等的不合理假设及样本分布不均衡对分类器影响严重等问题,引入多核函数和代价敏感学习机制,建立基于多核代价敏感极限学习机的故障诊断模型(MKL-CS-ELM)。并通过滚动轴承和单向阀二分类和多分类故障诊断的对比实验分析,方法取得与多核代价敏感支持向量机(multi-kernel cost sensitive support vector machine, MKL-CS-SVM)相当的处理效果,并继承了ELM时间消耗少的优点,提高了方法的实用性。同时,方法引入鲁棒性指标对代价敏感处理方法的效果进行评判,为代价敏感处理方法的选取提供了依据。(5)完成单向阀状态监测及故障系统的研发及测试。基于C#和Matlab混合编程模式,完成了单向阀状态监测及故障诊断系统开发。选取云南大红山铁精矿管道输送高压隔膜泵作为测试对象,采集单向阀整个生命周期的振动信号,完成单向阀运行状态监测及故障诊断系统测试。本文以矿浆管道输送大型往复式高压隔膜泵单向阀为研究对象,完成其状态监测与故障诊断方法的探索研究及系统开发,丰富了往复式机械设备的故障诊断研究理论,推动了往复式机械设备的故障诊断技术的应用及发展。
[Abstract]:The large reciprocating high pressure diaphragm pump is the core power equipment of long distance, high lift and high concentration slurry pipeline. Its working condition directly affects the production efficiency of the enterprise. As one of the core mechanical parts of the pump, the one-way valve needs to have good fast opening, fast closing, sealing and pressure bearing, which is more prone to failure than the other parts of the pump. In addition, the operation state of the one-way valve is closely related to the grain size distribution of the conveying minerals, the rheological characteristics of the slurry, the conveying pressure, the inherent material properties of the pump and the installation of the pump, which makes the failure of the one-way valve have the characteristics of sudden, concurrency, multi source, non-stationary and nonlinear, which greatly increases the difficulty of the state monitoring and fault diagnosis of the one-way valve. Therefore, starting from the analysis of the vibration signal analysis of one way valve, selecting effective feature extraction and fault diagnosis method is the core content of the monitoring and fault diagnosis of one-way valve operation and fault diagnosis. It has important theoretical research value and economic significance. The following research work is carried out on the state monitoring and fault diagnosis of one-way valve. (1) a new research work is put forward. A one-way valve fault detection method based on Local Mean Decomposition (LMD) and envelope demodulation is used. The one-way valve fault vibration signal is usually expressed as a complex amplitude modulation and frequency modulation signal, making use of the envelope demodulation method to extract the frequency of the one-way valve fault, but the one-way valve is subjected to environmental noise, coupling conditions and Other excitation sources, such as interference, the vibration signal shows obvious nonlinearity, and its envelope demodulation can not achieve the ideal effect. Therefore, a one-way valve fault detection method based on LMD and envelope demodulation is proposed. First, LMD is used to decompose the signal into a series of pure amplitude modulation signals, product function (Production Function,) PF); then the PF component is enveloped and demodulated to complete the one-way valve fault detection. (2) a one-way valve fault diagnosis method based on the multi domain hybrid feature limit learning machine (extreme learning machine, ELM) is proposed. The single domain characteristics can not fully describe the one-way valve movement state, the support vector machine (Support Vector Machine, SVM) and B are used. P neural network model has many optimization parameters, slow speed and so on. Combining the multi domain mixed feature and the advantage of ELM, a multi domain hybrid feature ELM based one-way valve fault diagnosis method is proposed. The multi domain mixed feature set is extracted from the time-domain, frequency domain, wavelet domain, and TK (Teager Kaiser) domain characteristics of the vibration signal of one-way valve, and the kernel principal component analysis (Kernel) is introduced. Principal Component Analysis, KPCA) method is used to extract the two characteristics of multi domain mixed feature set and eliminate feature redundancy. Finally, based on the multi domain mixed feature set after two feature extraction, a one-way valve ELM fault diagnosis model is established, and the one-way valve fault diagnosis is completed. (3) a kind of learning based on wavelet packet energy entropy and fuzzy kernel limit learning is proposed. Fuzzy kernel extreme learning machine (F-KELM) fault diagnosis method of one way valve. Based on the discussion of the complex nonlinear vibration signal, the disequilibrium of sample distribution and the influence of the number of neurons in the ELM hidden layer on the ELM classification performance, the wavelet packet energy entropy, the kernel function and the fuzzy membership function are introduced to establish the wavelet packet energy entropy and the fuzzy kernel. The fault diagnosis model of the limited learning machine. Through the comparison and analysis of the experiment of rolling bearing and one-way valve, it is proved that the method can solve the above problems effectively and improve the classification performance and generalization ability of the model. (4) a kind of multi-kernel cost sensitive extreme learning machine, MKL-CS-ELM is proposed. The single kernel function classifier can not fully interpret the classification decision function, the unreasonable assumption of the classification of the cost equality and the serious influence of the disequilibrium of the sample distribution on the classifier, and introduces the multi kernel function and the cost sensitive learning mechanism, and establishes the fault diagnosis model based on the multi-core cost sensitive limit learning machine. Type (MKL-CS-ELM). Through comparative experiment analysis of two classification and multi classification fault diagnosis of rolling bearing and one-way valve, the method has obtained the equivalent processing effect with multi-kernel cost sensitive support vector machine, MKL-CS-SVM, and inherits the advantages of low consumption of ELM time, and improves the method's reality. At the same time, the method introduces the robustness index to judge the effect of the cost sensitive processing method, and provides the basis for the selection of the cost sensitive processing method. (5) complete the state monitoring of the one-way valve and the development and test of the fault system. Based on the C# and Matlab hybrid programming model, the state monitoring and fault diagnosis system of the one-way valve is completed. The high pressure diaphragm pump of Yunnan Da Hongshan iron concentrate is selected as the test object, and the vibration signals of the whole life cycle of the one-way valve are collected to complete the monitoring of the operation state of the one-way valve and the test of the fault diagnosis system. The paper takes the large reciprocating high-pressure diaphragm pump as the research object, and completes its state monitoring and failure. The research and system development of the diagnostic method enrich the theory of fault diagnosis for reciprocating machinery and promote the application and development of the fault diagnosis technology of the reciprocating machinery.
【学位授予单位】：昆明理工大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：TH323
，
本文编号：2135556
[Abstract]:The large reciprocating high pressure diaphragm pump is the core power equipment of long distance, high lift and high concentration slurry pipeline. Its working condition directly affects the production efficiency of the enterprise. As one of the core mechanical parts of the pump, the one-way valve needs to have good fast opening, fast closing, sealing and pressure bearing, which is more prone to failure than the other parts of the pump. In addition, the operation state of the one-way valve is closely related to the grain size distribution of the conveying minerals, the rheological characteristics of the slurry, the conveying pressure, the inherent material properties of the pump and the installation of the pump, which makes the failure of the one-way valve have the characteristics of sudden, concurrency, multi source, non-stationary and nonlinear, which greatly increases the difficulty of the state monitoring and fault diagnosis of the one-way valve. Therefore, starting from the analysis of the vibration signal analysis of one way valve, selecting effective feature extraction and fault diagnosis method is the core content of the monitoring and fault diagnosis of one-way valve operation and fault diagnosis. It has important theoretical research value and economic significance. The following research work is carried out on the state monitoring and fault diagnosis of one-way valve. (1) a new research work is put forward. A one-way valve fault detection method based on Local Mean Decomposition (LMD) and envelope demodulation is used. The one-way valve fault vibration signal is usually expressed as a complex amplitude modulation and frequency modulation signal, making use of the envelope demodulation method to extract the frequency of the one-way valve fault, but the one-way valve is subjected to environmental noise, coupling conditions and Other excitation sources, such as interference, the vibration signal shows obvious nonlinearity, and its envelope demodulation can not achieve the ideal effect. Therefore, a one-way valve fault detection method based on LMD and envelope demodulation is proposed. First, LMD is used to decompose the signal into a series of pure amplitude modulation signals, product function (Production Function,) PF); then the PF component is enveloped and demodulated to complete the one-way valve fault detection. (2) a one-way valve fault diagnosis method based on the multi domain hybrid feature limit learning machine (extreme learning machine, ELM) is proposed. The single domain characteristics can not fully describe the one-way valve movement state, the support vector machine (Support Vector Machine, SVM) and B are used. P neural network model has many optimization parameters, slow speed and so on. Combining the multi domain mixed feature and the advantage of ELM, a multi domain hybrid feature ELM based one-way valve fault diagnosis method is proposed. The multi domain mixed feature set is extracted from the time-domain, frequency domain, wavelet domain, and TK (Teager Kaiser) domain characteristics of the vibration signal of one-way valve, and the kernel principal component analysis (Kernel) is introduced. Principal Component Analysis, KPCA) method is used to extract the two characteristics of multi domain mixed feature set and eliminate feature redundancy. Finally, based on the multi domain mixed feature set after two feature extraction, a one-way valve ELM fault diagnosis model is established, and the one-way valve fault diagnosis is completed. (3) a kind of learning based on wavelet packet energy entropy and fuzzy kernel limit learning is proposed. Fuzzy kernel extreme learning machine (F-KELM) fault diagnosis method of one way valve. Based on the discussion of the complex nonlinear vibration signal, the disequilibrium of sample distribution and the influence of the number of neurons in the ELM hidden layer on the ELM classification performance, the wavelet packet energy entropy, the kernel function and the fuzzy membership function are introduced to establish the wavelet packet energy entropy and the fuzzy kernel. The fault diagnosis model of the limited learning machine. Through the comparison and analysis of the experiment of rolling bearing and one-way valve, it is proved that the method can solve the above problems effectively and improve the classification performance and generalization ability of the model. (4) a kind of multi-kernel cost sensitive extreme learning machine, MKL-CS-ELM is proposed. The single kernel function classifier can not fully interpret the classification decision function, the unreasonable assumption of the classification of the cost equality and the serious influence of the disequilibrium of the sample distribution on the classifier, and introduces the multi kernel function and the cost sensitive learning mechanism, and establishes the fault diagnosis model based on the multi-core cost sensitive limit learning machine. Type (MKL-CS-ELM). Through comparative experiment analysis of two classification and multi classification fault diagnosis of rolling bearing and one-way valve, the method has obtained the equivalent processing effect with multi-kernel cost sensitive support vector machine, MKL-CS-SVM, and inherits the advantages of low consumption of ELM time, and improves the method's reality. At the same time, the method introduces the robustness index to judge the effect of the cost sensitive processing method, and provides the basis for the selection of the cost sensitive processing method. (5) complete the state monitoring of the one-way valve and the development and test of the fault system. Based on the C# and Matlab hybrid programming model, the state monitoring and fault diagnosis system of the one-way valve is completed. The high pressure diaphragm pump of Yunnan Da Hongshan iron concentrate is selected as the test object, and the vibration signals of the whole life cycle of the one-way valve are collected to complete the monitoring of the operation state of the one-way valve and the test of the fault diagnosis system. The paper takes the large reciprocating high-pressure diaphragm pump as the research object, and completes its state monitoring and failure. The research and system development of the diagnostic method enrich the theory of fault diagnosis for reciprocating machinery and promote the application and development of the fault diagnosis technology of the reciprocating machinery.
【学位授予单位】：昆明理工大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：TH323
，
本文编号：2135556
本文链接：https://www.wllwen.com/jixiegongchenglunwen/2135556.html
