Quality-related Fault Detection Of Chemical Process Based On Multivariate Statistics

With the development of science and technology,the features include complexity,enormousness all belongs to the modern industrial production machines.However,they still need to face the issues of coupling and nonlinearity.Although the traditional methods have the accurate physical modes for management and monitoring,they still could not deal with the problems proposed above.A fault detection method based multivariate statistics is focused on as its merits which do not need a priori knowledge on the system.The well fault detection results,if there is fault or not,could be obtained only relying on analyzing result for the collection data.If the appropriate method can be employed in a system to establish quality related variables model,then the false alarm,such as the outliers and irrelevant abnormal data,would be avoided.The idea has been widely concerned in the fault detection research field as for its better accuracy,production efficiency and lower production cost.So,based above discussion,the Principle Component Regression(PCR)and Improved principal Component Regression(IPCR)are employed in the multivariate statistical method to be the research basis and improves it in the thesis.The Tennessee Eastman Process(TEP)is employed also to be the research object for faults monitoring.The specific research contents are as follows.(1)The theory of three kinds of algorithm and its corresponding faults detection scheme are described in detail,which paves the way for the introduction and analysis of the algorithm in the future.(2)In order to solve the problem of false alarm and missing alarm caused by the fixed threshold value of T~2 statistic in the faults detection of TEP by IPCR,an adjustable threshold IPCR algorithm is proposed.Firstly,the normal data is used to establish the IPCR model,and the fixed threshold of traditional T~2 is obtained;secondly,in the online detection process,the new threshold is obtained by combining the fixed threshold and the exponentially weighted moving average in statistics,and the new threshold is used for faults detection.Finally,the simulation results in TEP show that this method can effectively reduce the false alarm rate and improve the faults detection ability of TE system.(3)There are feedback faults in the TEP system,which can be restored to normal through self-regulation in the system.However,the faults algorithm such as PCA and IPCR algorithm are always alarmed in monitoring.This problem is caused by the interference of unrelated variables,so an improvement standard principal component regression(ISPCR)method is proposed.This method first uses the mutual information to screen out the variables which are more related to the quality and eliminate the interference of the unrelated variables.Secondly,the mean value of two quality variables is used to judge whether the system has faults.Finally,in order to verify the effectiveness of ISPCR method,the experiment is compared with IPCR and PCA.Simulation results show that ISPCR method not only has outstanding detection effect in feedback faults,but also has good effect in other faults detection.