Multivariate statistical process monitoring based on blind source analysis

In this paper a new multivariate statistical process control (MSPC) method is presented based upon blind source analysis and wavelet transform. Blind source analysis based on ICA (independent component analysis) is used to compress the information in the data into low-dimensional spaces. Wavelet transform is employed to de-noise measured signals and extracted blind signals to remove the process noise. Later a MSPC based on de-noised data are developed to monitor process. The Q statistic and Hotelling T-2 statistic are used to calculate the confidence bounds. A double-effect evaporator is monitored and diagnosed by the presented method. The simulation results show that the method can detect fault more quickly, and so it improves monitoring performance of the process than conventional MSPC.