基于MATLAB故障诊断系统设计(9)

沈阳理工大学学士学位论文

monitoring methods. The Tennessee Eastman process simulator has been widely used by the process monitoring community as a source of data for comparing various approaches. The test problem is based on an actual chemical process where the components, kinetics, and operating conditions were modified for proprietary reasons. A diagram of the process is contained in Fig. 1.The simulation code allows 21 preprogrammed major process upsets, as shown in Table1. The plant-wide control structure recommended in Lyman and Georgakis[ was implemented to generate the closed loop simulated process data for each fault. For detailed discussion on the control structures of the Tennessee Eastman process simulator, please refer to Refs.

The training and testing data sets for each fault consisted of n?500 and 960 observations, respectively. Each data set started with no faults, and the faults were introduced 1 and 8 simulation hours into the run, respectively, for the training and testing data sets. All the manipulated and measurement variable except the agitation speed of the reactor’s stirrer for a total of m?52 variables were recorded. The data was sampled every 3 min, and the random seed was changed before the computation of the data set for each fault. Twenty-one testing sets were generated using the preprogrammed faults (Fault 1–21). In addition, one testing set (Fault 0) was generated with no faults.

Fig.2.The (D)PCA multivariate statistic for fault detection for Fault 4

36

沈阳理工大学学士学位论文

Multiple faults occurring within the same time window are likely to hap-pen for many industrial processes. The statistics for detecting a single fault are directly applicable for detecting multiple faults because the thresholds in Eqs. (4) and (6)depend only on the data from the normal operating conditions(Fault 0). Multiple simultaneous faults should be detectable provided that their affects on the measured process variables do not cancel. The task of diagnosing multiple faults is rather challenging and the proficiencies of the fault diagnosis statistics depend on the nature of the combination of the faults.

The minimum missed detection rate achieved for each fault except Faults 3, 9, and 15 is contained in a box in Table 4. Except for the CVA Q-statistic, the statistics which quantify variations in the residual space were usually sensitive to the faults than the statistics quantifying the variations in the score or state space. In other words, the faults usually created new states in the process rather than magnify the states during in-control operations. For example, consider Fault 4, where the missed detection rates for T2and the D (PCA)-based Q statistics are much smaller than T2and the (D.) PCA-based T2statistics (see Table 4).The extent to which the multivariate statistics are sensitive to Fault 4 can be examined in Figs. 2. 4. Conclusions

The Tennessee Eastman process simulator was used to compare PCA, D-PCA, and CVA for detecting faults in terms of sensitivity, promptness, and robustness. This comparison included a proposed residual space CVA statistic (T2) for fault detection. This appears to be the first application of this residual sp-ace CVA statistic for detecting faults.

Except for the CVA-based Q statistic, the statistics quantifying variations in the residual space (CVA T2, PCA Q, and DPCA Q statistics).were more sensitive for most faults than the statistics quantifying the variations in the score or state space (CVA T2, PCA T2and DPCA

T2 statistics.). The statistics exhibiting a small missed detection rate usually exhibited small detection delay and vice versa. The smallest detection delays were usually exhibited by either the CVA-based T2or Q statistics.

Based on the original thresholds, the PCA- and DPCA-based T2statistics gave lower false alarm rates than the DPCA-based Q and the CVA statistics. The lack of robustness of CVA was due to the CVA statistics being overly sensitive to a matrix inversion step. A strength of applying the proposed CVA-based T2statistic is that it is sensitive and prompt;

37

沈阳理工大学学士学位论文

however, its threshold must be adjusted to achieve robustness. With the thresholds defined so that the techniques had similar missed detection rates, DPCA had similar performance as to PCA for most faults.

38

沈阳理工大学学士学位论文

附录B：汉语翻译

基于PCA的工业过程故障诊断

1 概述

大量数据在许多工业生产过程中收集。故障检测任务是使用这数据确定反常行为什么时候发生，是否与设备故障、设备外包或者极端过程故障有关。由于为一个复杂的化学故障创建一个精确的模型需要很高的花费和时间，所以基于基本原理模型的技术都已经出现二十多年了，它在工业实践并没有得到广泛应用。在实践中得到广泛应用并在文献中得到广泛研究的故障检测技术都是利用全部过程数据建立模型的。检测故障的准确性，可以用数据的降维技术来提高，如主成分分析(PCA)动态主元分析(DPCA)和规范变量分析(CVA)。使用低维描述比全维描述更能推断出更新的过程数据。学术界和工业程序控制的工程师从多维化学过程数据[1][2]提取了PCA结构。PCA是根据获取数据的变化度来确定数据的最准确的低维描述法的。低维表示模型已经在诊断失控状态和诊断异常过程操作中应用[3]。PCA在Dupont和其他公司的工业数据中的几种应用已有六年，很多成果已经刊登在各种各样的出版物上。

考虑到数据中的序列相关性，PCA可以用前面的l个观测对每个观测向量进行扩充。我们将借鉴这种做法提出动态 PCA(DPCA)，不考虑滞后时间是怎么选定的(文献[6]就是这种方法的一种应用)。在多元统计分析中CVA是一种涉及选择对变量输入和输出最大化相关统计的降维技术。像DPCA这种方法，在降维过程中就把数据序列相关性的进行了考虑。

PCA 的已应用于从实际化工厂和计算机仿真收集的数据的检测故障(田纳西-伊斯曼过程)。对 DPCA 和 CAV 的应用，无论是在仿真或是化工方面都是有限的。本文的目的就是要评价和比较 PCA、DPCA和CAV在一个现实化工过程中检测故障的表现。在这 …… 此处隐藏：4272字，全部文档内容请下载后查看。喜欢就下载吧 ……