ADAPTIVE CHART BASED ON INDEPENDENT COMPONENT ANALYSIS FOR MULTIVARIATE STATISTICAL PROCESS MONITORING

Early detection of process faults is an important issue for ensuring plant safety and retaining high yield of final product in many industries, especially for process industries. The Independent Component Analysis (ICA) has been successfully applied in non-Gaussian multivariate statistical process monitoring (MSPM) recently. However, the conventional ICA-based monitoring method is not suitable for detecting small shifts of process since the monitoring statistic of ICA considers only the magnitudes of the most up-to-date samples but ignores the direction of process mean shifts. To overcome the drawback, this study aims to develop an adaptive chart based on ICA to enhance the fault detectability. The proposed method utilizes the Exponential Weighted Moving Average (EWMA) to predict the patterns of process mean shift and then constructs the adaptive monitoring statistic by combining the process mean shift and the ICA-extracted components. The proposed method is implemented by using two simulation studies to demonstrate the faults detection of process mean shifts and the small changes of system parameters. Furthermore, a real system, the Tennessee Eastman process, is conducted to evaluate the efficiency of the proposed method. The results show that the proposed method possesses superior performance when compared with various monitoring schemes.