一种基于相对主元分析的故障检测方法

通过主元分析方法进行主元提取时协方差矩阵特征值有时会出现变化“均匀”缺陷，并且该方法运用于故障检测时，通常使用单一的性能指标T2指标或SPE指标作为检测判据，容易遗漏一些重要信息，降低故障的检测效果。针对这一状况，本文研究了相对主元分析方法，并且将T2指标和SPE指标有机融合成综合指标，结合TE过程进行故障检测。通过单一指标SPE—时间图和综合指标—时间图的对比，发现综合指标图比SPE图报警时间早、误报少，说明了将综合指标运用于相对主元分析方法进行故障检测的优越性及有效性。

A fault detection approach using relative principal component analysis

The eigenvalues of the covariance matrix are almost the same in principal component analysis (PCA), and the single index T2 or squared prediction error (SPE) has often been utilized when using relative principal component analysis (RPCA) in fault detection. However, some important information in the fault-detection process is omitted when using this single index, and incorrect detection results can be obtained. To solve this problem, a comprehensive index has been introduced, by combining the index T2 and SPE in an effective way. Then, the combined index can be used in fault detection by relative principal component analysis in a typical Tennessee Eastman (TE) process. By comparing the simulation results in terms of the single index SPE and the combined index, it can be seen that the warning time was earlier and number of false alarms was less when using the combined index in the process of fault detection. The outcomes of this study demonstrate the effectiveness and feasibility of the method proposed.