基于FRICA算法的非高斯过程故障检测方法

独立元分析(ICA)是一种有效的非高斯过程故障检测方法,但其建模过程仅仅使用正常工况数据,忽视了对先验故障工况数据的利用.针对此问题,提出了一种基于故障相关ICA(FRICA)算法的故障检测方法.该方法使用ICA算法提取正常工况数据中的非高斯特征成分;再将正常工况数据集和先验故障工况数据集融合在一起构成多工况数据集,利用非局部保持投影进行二次特征提取,获得故障判别成分;在两种特征成分的基础上构造新的监控统计量,并利用核密度估计得到相应的置信限,完成对实时数据的监控.连续搅拌反应釜(CSTR)系统的监控仿真结果表明:与基本ICA方法相比,FRICA方法能更有效地检测出过程故障.

Non-Gaussian Process Fault Detection Based on FRICA Algorithm

Independent component analysis (ICA) is an enormously powerful non-Gaussian process fault detection method, but its modeling procedure only uses normal condition data, the utilization of prior fault condition data is frequently neglected. Aiming at this problem, fault related independent component analysis (FRICA) algorithm was presented for monitoring process faults. Firstly, ICA algorithm was applied to extract the non-Gaussian components in the normal condition data. Then the normal condition data set and prior fault condition data sets were integrated together to build multimode data sets. Non-local preservation projection was applied for secondary feature extraction in order to obtain fault discriminant components. New monitoring statistics are constructed on the basis of two kinds of components, and the corresponding confidence limits were obtained by kernel density estimation for the real-time data monitoring. The simulations on continuous stirring tank reactor (CSTR) system showed that FRICA method could be more effective to detect the process faults than that of the basic ICA method.