Greater situational awareness of plant conditions is necessary to move the current fleet of nuclear power facilities away from costly periodic maintenance activities. Sensed data provide the indicators of plant and equipment condition; however, these instrumentation and transmitters are themselves subject to aging and degradation over time. Online monitoring methods have long been proposed to assess the calibration status of sensors based on the data collected during normal plant operation. Auto-associative kernel regression models (AAKR) are commonly applied to predict the “expected” sensor value, and statistical hypothesis tests or thresholding algorithms are used to determine if the measured value agrees with the expectation. AAKR models work well for stationary operation of systems, but these models may not be as well suited for systems that undergo normal operational transients, as we expect to see in small modular reactors, advanced reactors, and many fuel cycle facilities. This paper presents an alternative approach to detection and diagnostics of sensor degradation and anomalies based on generalized singular value decomposition (GSVD) in computational linear algebra. The proposed method is demonstrated on experimental data collected on a two loop forced-flow water loop, but the approach is expected to be more generally applicable to a variety of nuclear facilities and to equipment and components beyond sensor suites.