面向多模态TE过程的故障诊断方法研究

发布时间：2018-05-14 03:35

本文选题：故障检测与诊断 + 多模态过程　；参考：《沈阳理工大学》2017年硕士论文

【摘要】：随着科技水平和工业化程度的不断提升,工业过程中的状态监控和故障检测也日趋复杂。这些复杂性主要表现在:多变量、非线性、强耦合等。此外,由于人们需求的日益多样化,这对单一的工业过程模态提出了挑战。于是,对于多模态工业过程领域的故障诊断技术的研究应运而生。这里的多模态是指由于操作条件、外界环境、过程本身固有因素或者特定需求的变化导致产生新的运行模态,使工业生产过程具有了多个稳定工况。多模态工业过程越复杂,其发生故障时造成的经济损失和人员伤亡往往也就越大,后果也就越严重。因此对整个多模态过程进行准确、有效的故障诊断显得尤为必要。本文在单模态TE过程的基础上提出以多模态TE过程为研究对象,展开对面向多模态过程故障诊断方法的研究。同时提出了一种面向多模态TE过程的集合型故障诊断方法,即GFCM-VMD-ICA-KPCA诊断方法,对多模态过程的故障进行了有效的检测和分离。同时,为解决多模态TE过程的大数据量故障诊断,本文又提出了一种基于大数据Hadoop平台以并行计算、分布式处理技术来进行故障诊断和分析的方法。本文提出了面向多模态TE过程的GFCM-VMD-ICA-KPCA集合型故障诊断方法。首先采用全局模糊C均值聚类算法(GFCM)对多模态数据进行聚类分析,以区分样本数据所属的工业模态。同时应用变分模态分解法(VMD)对数据进行预处理,滤除样本数据中的噪声。然后通过独立主元分析(ICA)算法提取主元变量,以降低核主成分分析法(KPCA)对于变量的分析维度,提高诊断效率。最后,利用KPCA的2T、SPE控制图和各变量贡献率图来输出对多模态TE过程的状态监控和故障诊断结果。并且引入一个数值仿真实例来验证该方法的有效性和准确性。本文针对多模态TE过程在海量数据情况下故障诊断算法效率大大降低的弊端,提出应用大数据Hadoop平台进行数据分析、故障诊断的方法。首先在原始样本数据进行模态聚类之后进行数据预处理,根据数值波动范围转化为相关标识字符文件,并通过FTP工具上传入大数据分布式文件系统(HDFS)。然后在MapReduce并行计算框架下编写字符检测程序进行数据分析和故障诊断。最后通过RStudio分析平台对输出相关的故障变量进行可视化展示,以达到故障检测与诊断的目的。为了验证所提的两种方法的有效性,进行了多模态TE过程的实验仿真。首先,将多模态TE过程的GFCM-VMD-ICA-KPCA集合型故障诊断方法与传统方法KPCA方法相比较,验证该方法的有效性和准确性。其次,将基于大数据Hadoop平台的故障检测方法付诸实验,验证该方法的有效性。实验仿真结果表明本文所提出的GFCM-VMD-ICA-KPCA集合型故障诊断方法和基于大数据Hadoop平台的故障数据诊断方法能够有效检测出故障,准确性和快速性优于传统方法。
[Abstract]:With the development of science and technology and industrialization, condition monitoring and fault detection in industrial process are becoming more and more complicated. These complexities are mainly manifested in: multivariable, nonlinear, strong coupling and so on. In addition, due to the increasing diversity of human needs, this poses a challenge to a single industrial process mode. Therefore, the research of fault diagnosis technology in the field of multimodal industrial process came into being. The multi-mode operation means that the operation conditions, the external environment, the inherent factors of the process itself or the change of the specific demand lead to the new mode of operation, which makes the industrial production process have more than one stable working condition. The more complex the multimodal industrial process is, the greater the economic loss and casualties will be, and the more serious the consequences will be. Therefore, it is necessary to make accurate and effective fault diagnosis for the whole multimodal process. On the basis of single mode te process, this paper presents a new method of fault diagnosis for multimodal te process based on multimodal te process. At the same time, a set fault diagnosis method for multimodal te process, called GFCM-VMD-ICA-KPCA diagnosis method, is proposed, which can effectively detect and separate the faults of multimodal process. At the same time, in order to solve the problem of mass data fault diagnosis in multimodal te process, this paper presents a method of fault diagnosis and analysis based on big data Hadoop platform, which is based on parallel computing and distributed processing technology. In this paper, a GFCM-VMD-ICA-KPCA set fault diagnosis method for multimodal te process is proposed. Firstly, the global fuzzy C-means clustering algorithm (GFCM) is used to analyze the multi-modal data in order to distinguish the industrial modes to which the sample data belong. At the same time, the variational mode decomposition (VMD) method is used to preprocess the data to filter the noise in the sample data. Then the independent principal component analysis (ICA) algorithm is used to extract the principal component variables in order to reduce the analysis dimension of KPCAs and improve the diagnostic efficiency. Finally, the state monitoring and fault diagnosis results of multimodal te process are outputted by using the control chart of 2T KPCA and the contribution rate diagram of each variable. A numerical simulation example is introduced to verify the validity and accuracy of the method. In this paper, a method of data analysis and fault diagnosis based on big data Hadoop platform is proposed to solve the problem that the efficiency of fault diagnosis algorithm of multimodal te process is greatly reduced under the condition of mass data. First, the data is preprocessed after modal clustering of the original sample data, and then converted into the relevant identification character files according to the range of numerical fluctuations, and then passed to the big data distributed file system (HDFS) via the FTP tool. Then the character detection program is written under the MapReduce parallel computing framework for data analysis and fault diagnosis. Finally, the output related fault variables are visualized through RStudio analysis platform to achieve the purpose of fault detection and diagnosis. In order to verify the effectiveness of the proposed two methods, the experimental simulation of multimodal te process is carried out. Firstly, the GFCM-VMD-ICA-KPCA set fault diagnosis method for multimodal te process is compared with the traditional KPCA method to verify the validity and accuracy of the method. Secondly, the fault detection method based on big data Hadoop platform is tested to verify the effectiveness of the method. The experimental results show that the proposed GFCM-VMD-ICA-KPCA set fault diagnosis method and the fault data diagnosis method based on big data Hadoop platform can effectively detect the fault, and the accuracy and rapidity are superior to the traditional methods.
【学位授予单位】：沈阳理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP277

【参考文献】