基于主元分析的动态系统故障检测方法研究

发布时间：2018-05-21 15:35

本文选题：基于数据驱动方法 + 主元分析　；参考：《中南大学》2012年硕士论文

【摘要】：动态系统的安全、稳定和高效运行一直是工业界十分关注的问题,尤其是担负有生产任务的动态系统,对其生产过程控制的品质提出了更高的要求,产品的质量,产品的故障率以及生产过程是否满足日益严格的安全和环境要求等问题时刻被关注。开展对动态系统故障诊断研究将是非常有必要而且能为企业运行节约经济成本,同时也能为环境友好工程和可持续发展带来新的方法和研究课题。本文对主元分析故障诊断方法的基本原理进行了介绍,针对传统主元分析在监测动态系统具有多变量耦合、时变、大滞后等特性时容易出现误报和漏报,以及动态主元分析在进行在线监测时反应不够迅速等的问题,研究了基于递归动态主元分析的故障检测方法。同时,在深入分析了旧统计量的基础之上,研究了两个综合统计量,并给出了具体的表达式和阈值计算公式。通过实际密闭鼓风炉数据,建立了递归动态主元分析模型,结合两个新的统计检测量与动态主元分析进行了比较,实验结果证明了递归动态主元分析方法的准确性和可行性,且新提出的监测统计量也能很好的应用在对炉况的在线监测中。在研究了子空间辨识方法的基础上,本文对基于主元分析方法和子空间辨识方法集成的故障检测算法(SIMPCA)进行了研究,通过合理的构造观测数据矩阵,利用SIMPCA方法建立了过程的监测模型,通过理论分析,该算法能消除随机干扰和噪声的影响,使得到的残差仅和故障信号有关。随后,通过CSTH模型仿真研究,验证了SIMPCA故障检测算法进行过程监测的可行性和有效性。通过正常运行过程和异常过程的监测对比分析,SIMPCA故障检测算法能有效区分过程的正常和异常运行情况,准确检测出异常的发生。
[Abstract]:The safety, stability and efficient operation of a dynamic system have always been a matter of great concern in the industry, especially the dynamic system that bears the production task, which puts forward higher requirements for the quality of its production process control, the quality of the product, the failure rate of the product, and whether the production process meets the increasingly stringent safety and environmental requirements. The research on dynamic system fault diagnosis will be very necessary and can save the economic cost for the operation of the enterprise, and also bring new methods and research topics for environmental friendly engineering and sustainable development.
In this paper, the basic principle of the principal component analysis fault diagnosis method is introduced. In view of the traditional principal component analysis, it is easy to misreport and misreport when the dynamic system has the characteristics of multivariable coupling, time-varying, large lag and so on, and the dynamic principal component analysis is not fast enough in the on-line monitoring. At the same time, on the basis of the analysis of the old statistics, two comprehensive statistics are studied, and the concrete expressions and the formula of the threshold calculation are given. The recursive dynamic principal component analysis model is established through the actual closed blast furnace data, and two new statistical detection quantities and dynamic principal component analysis are combined. The experimental results prove the accuracy and feasibility of the recursive dynamic principal component analysis method, and the new monitoring statistics can be well applied to the on-line monitoring of the condition of the furnace.
On the basis of studying the subspace identification method, the fault detection algorithm (SIMPCA) based on the principal component analysis method and the subspace identification method is studied. Through the rational construction of the observation data matrix, the SIMPCA method is used to establish the monitoring model of the process. Through the theoretical analysis, the algorithm can eliminate the random interference and noise. The effect of sound is only related to the fault signal. Then, the feasibility and effectiveness of the process monitoring of the SIMPCA fault detection algorithm are verified by the CSTH model simulation research. The SIMPCA fault detection algorithm can effectively distinguish the normal and abnormal operation of the process through the comparison and analysis of the normal operation process and the abnormal process monitoring. Situation, accurate detection of abnormal occurrence.
【学位授予单位】：中南大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：TH165.3

【引证文献】