多变量经验模态分解在化工过程故障诊断中的应用研究

发布时间：2018-04-23 16:29

本文选题：经验模态分解 + 可见图　；参考：《北京化工大学》2015年硕士论文

【摘要】：现今化工过程生产工艺日趋复杂、规模日渐庞大,加之化工数据本身维度高、数量大、相关性强且充斥噪声等特点,无不令安全生产和产品质量面临更大挑战和更高要求。为此,本文利用经验模态分解(EMD)按照频率尺度自适应处理信号提取故障特征的能力,以及多变量经验模态分解(MEMD)处理关联信息的优势,对其在化工过程中的故障诊断应用探索研究,同时结合改进的动态可见图算法,提出一种数据驱动的故障诊断方法,应用于田纳西-伊士曼(TE)过程的实时监测和在线诊断,仿真结果验证了提出方法的有效性与优越性。主要研究内容涵盖如下：首先,利用残差对故障的敏感性,提出基于总体平均EMD(EEMD)残差的故障诊断方法。根据历史数据的66控制图,确定残差的故障诊断控制限；利用在线实时数据采用贝叶斯信息准则在线确定EEMD的移动窗口；通过移动窗口的采样数据,在线获得EEMD残差最大值的变化,结合相应的故障诊断控制限在线诊断故障并确定故障发生时间及原因。其次,为提高过程监测的效率和精度,克服单一变量监测的局限性,提出一种基于改进动态可见图(MDVG)算法的多变量过程在线故障监测方法。通过数据归一化和引入时间间隔常数,改进动态可见图(DVG),使得DVG所关注特性的众数出现次数的均值最小,以细分不同时序数据的网络特性。最后,结合MEMD和MDVG,提出MEMD-MDVG故障诊断方法。利用MEMD残差确定监测变量,将各个监测变量的历史数据利用M MDVG确定监测指标及阈值并实施在线监测,异常情况下再借助MEMD残差进行相关性分析以确定故障原因。
[Abstract]:Nowadays the production process of chemical process is becoming more and more complex and the scale is increasing. In addition the chemical data itself has the characteristics of high dimension large quantity strong correlation and full of noise. All of them make the safety production and product quality face more challenges and higher requirements. Therefore, this paper makes use of the ability of EMD) to extract fault features according to frequency scale adaptive signal processing, and the advantage of multivariable empirical mode decomposition (MEMD) in processing associated information. A data-driven fault diagnosis method is proposed, which is applied to real-time monitoring and on-line diagnosis of Tennessee Eastman (TET) process. Simulation results verify the effectiveness and superiority of the proposed method. The main contents are as follows: firstly, using the sensitivity of residuals to faults, a fault diagnosis method based on the total average EMDEEMD residuals is proposed. According to the 66 control chart of historical data, the fault diagnosis control limit of residuals is determined; the moving window of EEMD is determined online by using online real-time data using Bayesian information criterion; the sampling data of moving window is sampled through the moving window. The variation of the maximum residual error of EEMD is obtained online, and the time and reason of fault occurrence are determined by combining with the corresponding fault diagnosis control limit. Secondly, in order to improve the efficiency and precision of process monitoring and overcome the limitation of single variable monitoring, a method of on-line fault monitoring for multivariable process based on improved dynamic visibility map (MDVG) algorithm is proposed. By means of data normalization and the introduction of time interval constant, the dynamic visibility map (DVG) is improved to minimize the average number of modes of appearance of the characteristics concerned by DVG, so as to subdivide the network characteristics of different time series data. Finally, combined with MEMD and MDVG, the method of MEMD-MDVG fault diagnosis is proposed. The monitoring variables are determined by MEMD residuals, and the monitoring indexes and thresholds are determined by M MDVG with the historical data of each monitoring variable, and on-line monitoring is carried out. In abnormal cases, the correlation analysis of MEMD residuals is used to determine the causes of faults.
【学位授予单位】：北京化工大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TQ02

【参考文献】