Data-driven Process Monitoring Method Based on Dynamic Component Analysis

A novel data-driven process monitoring method based on dynamic independent component analysis-principle component analysis (DICA-DPCA) is proposed to compensate for shortcomings in the conventional component analysis based monitoring methods. The primary idea is to first augment the measured data matrix to take the process dynamic into account. Then perform independent component analysis (ICA) and principle component analysis (PCA) on the augmented data to capture both the non-Gaussian and Gaussian process information. Finally, a combined monitoring statistic is proposed by support vector data description (SVDD) with its control limit being determined by bootstrap quantile estimation method to lessen monitoring workload. The Tennessee Eastman process is used to demonstrate the improved monitoring performance of the proposed mechanism in comparison with existing component analysis based monitoring methods, including PCA, ICA, ICA-PCA, dynamic PCA, and dynamic ICA.