Redundant sensor calibration monitoring using independent component analysis and principal component analysis

This paper presents a comparison of methods for industrial on-line sensor calibration monitoring for redundant sensors. Principal component analysis (PCA) and independent component analysis (ICA) techniques are developed and compared using both simulated data and data sets from an operating nuclear power plant. The performance is dependent on the types of noise sources; however, under most conditions ICA outperforms PCA, based on the bias and variance of their respective parameter estimates. A case study is included to demonstrate the usefulness of both techniques for the early detection of sensor drift.