Multivariable Statistical Process Monitoring

The traditional statistical process monitoring is based on the single variable statistical control method. The single variable statistical control method only monitors the change of single variable, and won't provide the effective information of correlation of more variables. In additional, single variable statistical control isn't adaptive in batch process.Multivariable statistical process monitoring can find the error of the process by effective monitoring and improvement the safety of process.MSPCA was improved in this paper, and was applied in continuous process and batch process. The work had finished was:1 Pre-processing (Filtering) the measurements from industryOn the basis of the wavelet analysis, via the wavelet toolbox of matlab language, a method of denosing via wavelet thresholding was introduced and applied. Then, on-line filtering was introduced and applied to the emulational signal form matlab.2 Application and compare of MSPCA and Improvement of MSPCA MSPCA method has wide application on the process monitoring. On the basis of studying MSPCA, this paper introduced an improvement method that will threshold the wavelet coefficients when the data is decomposed, and it could combine the wavelet rectification with the MSPCA, and the squared prediction error (SPE) of the statistical control graph was used to detect the process variants which induce the change or fault of the process. On the principle of invariability of the complexity, it could eliminate the contamination of the data, and decrease the false alarm of the fault diagnosis.3 Application of MPCA in batch processMPCA methods has extensive application in batch process, it predigests analysis of the multidimensional data and resolves the special problem of batch statistical. MSPCA and MPCA was combined and applied in batch process in this paper.4 Application of Multi-PCA model in batch processWhen the process has some different stages and every variable's correlation is different in different stages, the control limit of multivariable statistical wills not the same in every stage. This would lead to the data with error was considered to be the normal data and improved the false alarm rate of the fault diagnosis. So we should make models relation to different stages. Firstly, the K-means method was used to analyze the sample according to the different stages, and then the clustering sample was analyzed by PCA and compared with the PCA which was not use the K-means.