Sensor Fault Detection and Isolation With Interval Principal Component Analysis: Application to a Heat Exchanger System

Multivariate statistical process monitoring (MSPM) methods, particularly principal component analy- sis (PCA), have been widely applied for sensor Fault Detection and Isolation (FDI) in industrial systems. However, conventional PCA-based approaches typically rely on deterministic models derived from normal operating data and often fail to consider uncertainties in model parameters, which compromises their reliability in complex environments. To address this limitation, this work proposes an interval PCA-based FDI framework that explicitly incorporates uncertainties into the monitoring model. We introduce a new squared prediction error (SPE) indicator, specially adap- ted for interval PCA models, to improve the reliability of fault detection under uncertain conditions. In addition, we propose a fault isolation method based on a reconstruc- tion strategy. The proposed approach is validated through simulations on a heat exchanger system. Experimental results demonstrate improved fault detection accuracy and effective fault isolation compared to conventional PCA-based methods. The integration of interval modeling enhances the reliability of the monitoring system and significantly reduces the number of required reconstructions from 1893 to just 57.