基于故障相关慢特征分析的过程监测方法

Process monitoring algorithm based on fault-related slow feature analysis

针对慢特征分析(SFA)算法在特征选择时没有利用过程在线故障信息的问题,提出基于选择在线故障相关特征的慢特征分析(FRSFA)故障检测算法。首先采用SFA算法提取过程动态特征,采用核密度估计方法估计特征平均阈值,作为在线特征选择的基准。然后记录超过平均阈值的在线慢特征为故障相关特征,在正常数据的动态特征中挑选当前故障相关特征,估计当前监测样本的控制限。最后,将提出的算法应用于田纳西-伊斯曼过程,结果表明相比于主成分分析和SFA算法,FRSFA算法充分利用动态过程在线故障信息,增强模型的有效性。

Because the slow feature analysis(SFA)algorithm does not utilize online fault information in feature selection,a fault detection algorithm based on the selection of online fault-related feature(FRSFA)was proposed.Firstly,the SFA algorithm was used to extract the dynamic features of the process,and the kernel density estimation method was used to estimate the average threshold of the features as the reference for online feature selection.Then the online slow features exceeding the average threshold were recorded as the fault-related features.The current fault-related features were selected from the dynamic features of normal data,and the control limit of the current monitoring sample was estimated.Finally,the proposed algorithm was applied to Tennessee Eastman process.The results show that compared with principal component analysis and SFA algorithms,FRSFA can fully use the online fault information of dynamic process and enhance the effectiveness of the model.