基于PCA的流程工业故障诊断方法研究

发布时间：2018-01-16 20:42

  本文关键词：基于PCA的流程工业故障诊断方法研究　出处：《西南科技大学》2016年硕士论文　论文类型：学位论文

  更多相关文章： 故障检测 主元分析 局部保持投影 多模态 局部近邻标准化

【摘要】：故障诊断对保证流程工业生产安全、提高产品质量有着重要作用。本文针对实际流程工业过程中存在的数据信息提取、多模态、数据分布复杂等问题,对流程工业故障诊断方法进行研究。具体包括:(1)针对传统主元分析方法在数据降维中仅考虑数据全局结构的问题,采用一种局部整体结构保持投影(LGSPP)算法,使投影后的低维空间不仅和原始空间有相似的整体结构,而且保留了相似的局部结构。将高维数据降维到低维空间后,构造统计量和贝叶斯分类器对故障进行检测和辨识;考虑到数据中存在的动态问题,构造含有前h个观测的增广矩阵,将其应用到LGSPP算法中,改善了动态工业过程的故障检测效果。TE过程的仿真结果表明了算法的有效性。(2)从数据预处理的角度改善工业过程数据的多模态问题,在对数据进行标准化处理时引入局部近邻标准化策略,得到LNS-LGSPP算法。采用LNS对数据进行标准化处理,去除不同变量量纲的影响以及多模态数据的多分布特征,将处理后的数据应用到LGSPP算法中进行故障检测。数值仿真和TE过程的仿真结果表明了该算法在多模态过程监控中的有效性。
[Abstract]:Fault diagnosis plays an important role in ensuring the production safety of process industry and improving product quality. This paper aims at the problems of data information extraction, multi-modal and complex data distribution in the actual process industry. This paper studies the fault diagnosis method of process industry, including: 1) the traditional principal component analysis method only considers the global structure of data in data dimensionality reduction. A local global structure preserving projection (LGSPP) algorithm is used to make the projected low-dimensional space not only have a global structure similar to the original space. After reducing the dimension of high-dimensional data to low-dimensional space, we construct statistics and Bayesian classifier to detect and identify faults. Considering the dynamic problem in the data, the augmented matrix with the first h observations is constructed and applied to the LGSPP algorithm. The simulation results show that the algorithm is effective and can improve the multi-modal problem of industrial process data from the point of view of data preprocessing. When the data is standardized, the local nearest neighbor standardization strategy is introduced, and the LNS-LGSPP algorithm is obtained. The LNS is used to standardize the data. The influence of different variables dimension and the multi-distribution characteristics of multi-modal data are removed. The processed data are applied to the LGSPP algorithm for fault detection. The numerical simulation and the simulation results of te process show that the algorithm is effective in multi-modal process monitoring.
【学位授予单位】：西南科技大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP277
，

本文编号：1434771