基于变量预测模型模式识别的旋转机械故障诊断研究

发布时间：2018-03-01 15:42

本文关键词： 基于变量预测模型的模式识别 LMD能量矩多尺度高阶奇异谱分析模型融合特征选择新异类检测旋转机械故障诊断　出处：《湖南大学》2015年博士论文　论文类型：学位论文

【摘要】：随着科学技术的发展,故障诊断技术逐渐成为了保障旋转机械设备安全可靠运行的核心支持技术之一。对旋转机械故障诊断新技术、新方法的研究具有重要的理论和实际意义。旋转机械故障诊断技术的实质是模式识别的问题。模式识别方法的选择与运用对提高故障诊断的精度和稳定性具有十分重要的作用。在旋转机械故障诊断领域,广泛使用的模式识别方法有神经网络、支持向量机等,但这些方法都存在着各自的局限性,且没能充分利用特征变量之间的相互内在关系。实际上,通过现代信号处理方法提取的特征值之间往往存在一定的相互内在关系,不同的系统或者同一系统不同的状态,相互内在关系的数学表达式存在明显差异。基于变量预测模型模式识别(Variable Predictive Model Based Class Discriminate,VPMCD)方法是一种新的模式识别方法。VPMCD方法能充分利用各个特征值之间的相互内在关系建立变量预测模型(Variable Predictive Model,VPM)的数学表达式,从而进行分类识别。为了将VPMCD方法应用于小样本多分类的旋转机械故障诊断,本文在国家自然科学基金项目的资助下(编号:51175158),对VPMCD方法的关键理论及其在小样本多分类的旋转机械故障诊断中的应用进行了深入而系统地研究。本文主要的研究内容和创新点如下:(1)研究了VPMCD方法的基本原理和具体算法,总结了VPMCD方法的特点,将VPMCD方法与神经网络、支持向量机等方法进行了对比研究,分析结果表明:VPMCD方法在分类性能、运算速度等诸多方面具有明显的优势。(2)针对原VPMCD方法中模型参数估计方法存在的不足,提出了采用加权最小二乘参数估计来代替最小二乘参数估计,从而改进VPMCD方法,仿真分析结果证表明,改进后的VPMCD方法在更少的训练样本下,可以取得了更高的模型拟合精度。(3)针对具体的旋转机械故障诊断问题,结合最新的现代信号处理技术,提出了多种特征提取方法:LMD(Local Mean Decomposition,LMD)能量矩的特征提取方法,改进ITD(Intrinsic Time-scale Decomposition,ITD)特征提取方法,LCD(Local Characteristic-scale Decomposition,LCD)和模糊熵相结合的特征提取方法,LCD和SVD(Singular Value Decomposition,SVD)相结合的LCD-SVD特征提取方法,以及多尺度高阶奇异谱特征提取方法。结合以上特征提取方法,提出了各种基于VPMCD的故障诊断模型,并通过应用实例验证了论文提出的各种模型均能有效性地应用于旋转机械故障诊断领域。(4)针对原VPMCD方法的模型选择单一、信息利用不充分的问题,结合遗传算法(Genetic algorithm,GA),提出了GA-VPMCD分类识别方法。首先采用回代(Re-substitution,RS)验证或者交叉验证方法,结合模型检验,选取验证精度最高,且模型拟合优度最高的模型作为弱VPM;然后,采用模型融合的思想,利用遗传算法融合各个弱VPM的预测值得到最佳预测值;最后,依据误差平方和最小来实现分类识别。结合阶次包络分析技术,将GA-VPMCD方法应用于变速滚动轴承故障诊断;结合多尺度高阶奇异谱分析,将GA-VPMCD方法应用于转子故障诊断;实验结果表明,GA-VPMCD方法有效提高了故障诊断精度和稳定性。(5)针对特征选择问题,将VPMCD方法与ANN、平均影响值(Mean Impact Value,MIV)相结合,提出了ANN-MIV-VPMCD分类识别方法,并进一步提出了基于LCD-SVD和ANN-MIV-VPMCD的滚动轴承故障诊断模型。实验结果验证了ANN-MIV-VPMCD方法的有效性和优越性。(6)在多数情况下,旋转机械故障诊断面临只有正常样本,或者故障模式不完备、典型故障样本缺乏。针对这个问题,提出了OC-VPMCD新异类检测方法,并应用于旋转机械新异类检测。实验分析结果表明,OC-VPMCD方法能有效地应用于旋转机械新异类检测。
[Abstract]:With the development of science and technology, fault diagnosis technology has gradually become the core guarantee safe and reliable operation of rotating machinery is one of the support technology. The new technology of the fault diagnosis of rotating machinery, has important theoretical and practical significance to study the new method. The essence of technology of rotating machinery fault diagnosis is the problem of pattern recognition and pattern recognition. The choice is very important to improve the fault diagnosis accuracy and stability. In the field of rotating machinery fault diagnosis, pattern recognition methods are widely used neural network, support vector machine and so on, but these methods have their own limitations, and can not make full use of the mutual relationship between the variables. In fact, there are often some inherent relation between the values through the modern signal processing feature extraction method, different systems or different systems of the same There are obvious differences between the state, the mathematical expression of internal relations. Pattern recognition based on variable prediction model (Variable Predictive Model Based Class Discriminate, VPMCD) is a new pattern recognition method.VPMCD method can make full use of each feature value between the intrinsic relationship between variables to establish prediction model (Variable Predictive Model, VPM) mathematical expressions thus, the classification and recognition. In order to fault diagnosis of rotating machinery VPMCD method applied in small sample classification, based on the National Natural Science Foundation of China (No. 51175158), the application of rotating machinery fault diagnosis key theory of VPMCD method and its classification in the small sample in depth and systematically the research. The main research content and innovation are as follows: (1) the basic principle of VPMCD method and specific algorithm, summed up the VPMC The characteristics of the D method, VPMCD method and neural network, support vector machine method are studied. The analysis results show that the classification performance of the VPMCD method, has obvious advantages in computing speed and other aspects. (2) aiming at the shortage of the original VPMCD method in the estimation of model parameters in the proposed method, using weighted least squares parameter estimation instead of least squares parameter estimation, improved VPMCD method, the simulation results show that the VPMCD card, the improved method in less training samples, the model can achieve higher precision. (3) according to the fault diagnosis of rotating machinery in detail, combined with modern signal processing technology, is put forward extraction of various features: LMD (Local Mean Decomposition, LMD) feature extraction method of energy moment, improved ITD (Intrinsic Time-scale Decomposition ITD (LCD) feature extraction method, Local Chara Cteristic-scale Decomposition, LCD) feature extraction method and combining fuzzy entropy, LCD and SVD (Singular Value Decomposition, SVD) LCD-SVD feature extraction method combining, and multi-scale singular spectra feature extraction method. The extraction method and combining with the above characteristics, put forward various fault diagnosis model based on VPMCD, and through the application instance the validation of the models proposed in this paper can be effectively applied in the field of fault diagnosis of rotating machinery. (4) according to the original VPMCD model to choose a single, the problem of insufficient information utilization, combined with genetic algorithm (Genetic algorithm GA), proposed a GA-VPMCD classification method. Firstly, using back substitution (Re-substitution, RS) validation or cross validation method, combined with the model test, select the highest accuracy and model validation, the highest goodness of fit as a weak VPM; then, using the model of integration of thinking To predict, using genetic algorithm to fuse the various weak VPM is the best predictor; finally, based on the minimum error sum of squares to achieve classification and recognition. Combining the order envelope analysis technique, the GA-VPMCD method is applied to fault diagnosis of rolling bearing transmission; combined with multiscale high-order singular spectrum analysis, the GA-VPMCD method was applied to the fault diagnosis of rotor GA-VPMCD; experimental results show that the method improves the fault diagnosis accuracy and stability. (5) according to the feature selection problem, and the ANN VPMCD method, the average value of influence (Mean Impact Value, MIV) combined with the proposed ANN-MIV-VPMCD classification method, and further puts forward the rolling bearing fault diagnosis model based on ANN-MIV-VPMCD and LCD-SVD. The experimental results verify the validity and superiority of the ANN-MIV-VPMCD method (6). In most cases, the fault diagnosis of rotating machinery facing only normal or fault samples. Incomplete models and lack of typical fault samples. Aiming at this problem, a new OC-VPMCD heterogeneous detection method is proposed and applied to the new heterogeneous detection of rotating machinery. Experimental results show that OC-VPMCD method can be applied to the new heterogeneous detection of rotating machinery effectively.

【学位授予单位】：湖南大学
【学位级别】：博士
【学位授予年份】：2015
【分类号】：TH17

【相似文献】