基于特征值变化的工业过程实时故障检测

REAL-TIME FAULT DETECTION OF INDUSTRIAL PROCESS BASED ON VARIATION OF EIGENVALUE

针对核主元分析(Kernel Principal Component Analysis,KPCA)模型所产生的对角矩阵包含原始数据信息,且能够反映数据的特征,提出一种基于特征值变化的工业过程实时故障检测方法。因滑动窗口在收集数据建模时会出现故障数据被正常数据覆盖的现象,故采用实时数据和正常数据相结合的组合滑动窗口策略收集数据建立KPCA模型。通过KPCA模型所产生特征值的信息变化构造新的监测统计量,即数据发生故障时,变量值超出原来的范围,特征值会变大,利用这一变化规律构造监控统计量。将统计数据与置信限进行比较,从而对样本状态进行监视。通过数值例子和田纳西-伊斯曼过程的仿真实验,结果表明与其他的方法进行比较,该方法在过程故障检测中提高故障检测率,降低误报率。

Aimed at the fact that the diagonal matrix generated by kernel principal component analysis(KPCA)model contains the original data information and can reflect the characteristics of the data,this paper proposes a real-time fault detection method for industrial process based on the variation of eigenvalue.Because the fault data would be covered by the normal data when the sliding window collected data for modeling,the KPCA model was established by the sliding window strategy with combination of real-time data and normal data.A new monitoring statistic was constructed by the information change of the eigenvalue generated by KPCA model,that was,when the data failed,the variable value exceeded the original range,the eigenvalue would increase,and the monitoring statistic was constructed by using this change rule.The statistical data was compared with the confidence limit to monitor the sample status.Through simulation experiments of a numerical example and the Tennessee Eastman process,the results show that compared with other methods,this method improves the fault detection rate and reduces the false alarm rate in process fault detection.