Weighted principal component analysis applied to continuous stirred tank reactor system with time-varying

Fault detection approach based on principal component analysis (PCA) has been applied widely and effectively to chemical processes monitoring system. A novel PCA approach called weighted PCA (WPCA) for continuous stirred tank reactor system (CSTR) with time-varying is addressed in the paper. Time-varying can cause unfavorable influence on feature extraction, but weighted PCA approach can obtain slow features information of observed data in CSTR with time-varying. The monitoring statistical indices are based on WPCA approach and their confidence limits are computed by kernel density estimation (KDE). A simulation illustrated that the proposed method achieves better dynamical performance from the perspective of fault detection rate and fault detection time than PCA approach.