In process engineering, a fast and efficient fault detection and diagnosis (FDD) system is an essential component to improve both safety and productivity losses under abnormal con-ditions. Over the years, techniques based on models derived from process historical data, specially under a probabilistic framework, have gain a lot of attention. In this paper, probabilistic principal component analysis (PPCA) mixture models are used to cope with the FDD task. A batch-incremental method is proposed for statistical process monitoring, seeking to detect and learn new faulty behaviour, or yet, diagnose an already known fault. The proposed methodology was applied to the Tennessee Eastman Process under a closed-loop control, and it has shown robust and reliable results.