I. Introduction
Fault detection and process monitoring play an indispensable role in the safe, reliable, and efficient operation of industrial processes [1], [2], [3]. Thanks to the advanced sensor technology and industrial Internet of Things (IoT), massive process data are available for machine learning and analysis [4], [5], [6]. Data-driven process monitoring based on multivariate statistical analysis has been extensively studied in recent years [7], [8]. Since industrial process data are often cross-correlated or collinear, latent variable methods that extract principal variations from data [9], [10] are widely adopted as effective tools to establish models for process monitoring.