Research On Process Fault Detection Of Process Industry Based On Data-driven

With the development of computer technology and the concept of industrial 4.0 was promoted vigorously,modern industrial processes tend to be more automated,integrated,complex and intelligent.The traditional fault detection method is not suitable for the fault detection in such complex industrial processes.In recent years,the rapid development of sensor technology and real-time data storage technology,a large number of process data stored in the industrial process.In this case,fault detection method based on data-driven is rapidly developed and widely used in modern industrial process monitoring.Fault detection method based on data-driven is a big concept,including a variety of fault detection methods.Multivariate statistical analysis is important branch of fault detection methods based on data driven.At present,its application is also the most widely.This paper mainly focuses on the multivariate statistical analysis.The main contents are as follows:(1)An improved dynamic principal component analysis(DPCA)algorithm is proposed to solve the problem that T2 statistic exceed the control limits but SPE statistic does not in traditional fault detection.This method uses principal component related variable residual(PVR)statistic to replace SPE statistic,which can provide more detailed process variable information related to the principal component,and effectively identify the change of T2 chart.(2)An improved kernel principal component analysis(KPCA)algorithm is proposed to solve the problem of calculating the kernel matrix K of the KPCA and the false alarm rate which are easily caused by some small disturbances in the process of fault detection.Indiscernibility degree and feature vector selection are introduced in this method,which will remove the irrelevant variables or low relevant variables to reduce the false alarm rate by using the indiscernibility and reduce the number of feature samples to shorten the fault detection time and improve the detection efficiency.by using feature vector selection.(3)In order to promote the practical application of data-driven technology in industrial field,with the Distillation Column as one of the controlled objects,the hardware-in-the-loop simulation system was developed.The improved methods are applied to the hardware-in-the-loop simulation platform,and all of them have achieved good detection results.At last,applying the mixed programming of VB and Matlab,combining the fault detection method with the hardware-in-the-loop simulation platform to develop the fault detection monitoring platform of Distillation Column.Aiming to different fault of the Distillation Column,it can selectively use different methods to detect,and improve the efficiency of fault detection.