Chemical Separation Process Monitoring And Diagnosis Using Multivariate Statistical Process Control

For the chemical separation process, a large number of variables are usuallymeasured and stored in computer database during process operation. These variablesare usually highly correlated and the real dimensionality of the monitored process isconsiderably less than that represented by the number of process variables collectedand the accurate formulation of the process is hard to acquire. So we adopt the methodcalled multivariate statistical process control (MSPC) to monitor the process.Firstly we use principal component analysis (PCA) to reduce the dimensionalityof the process by creating a new set of variables, principal components, which canreflect the true underlying system dimension. Process performance can then bemonitored in a low dimensional component space by SPE and T2 plots. Secondly we combine some developments up to date in this field, divide the rawdata into two subspaces and use two new statistics to take place the SPE statistic inthe conventional PCA. The improved PCA can efficiently detect weak changes andbetter explain the process behavior. Thirdly a novel non-linear PCA method is adopted to monitor the process basedupon the Input-Training neural network. Adopting this method, both the linear andnon-linear information of the process can be captured in the final non-linear principalcomponent scores. Then we realize on-line process monitoring based on above theories, makingstatistical process control more useful to practical manufacture. At last, a software is designed using the VC++ language and its interface withMATLAB language. The software is consisted of principal component analysismodule, data reconstruction module, process monitoring module, fault diagnosisfunctions and so on. It gives users a convenient process monitoring tool and helppeople who don't master the SPC theory can also analyze the process easily andcorrectly.