Uncertainty Quantification Techniques for Sensor Calibration Monitoring in Nuclear Power Plants

Abstract

Executive SummarySafe, efficient, and economic operation of nuclear systems (nuclear power plants, fuel fabrication and storage, used fuel processing, etc.) relies on accurate and reliable measurements. Newer types of sensors, and sensors to monitor non-traditional parameters, are expected in next-generation nuclear power plant (NPP) and fuel-cycle environments. A number of factors (besides changes in the monitored variable) affect the measured signals, resulting in effects such as signal drift and response time changes, requiring techniques to distinguish between signal changes from plant or subsystem performance deviations and those from sensor or instrumentation issues. Advanced algorithms that continuously monitor sensor responses can address this issue and facilitate automated monitoring and control of plant and subsystem performance.Currently, periodic sensor recalibration is performed to avoid problems with signal drift and sensor performance degradation. Periodic sensor calibration involves (1) isolating the sensor from the system, (2) applying an artificial load and recording the result, and (3) comparing this "As Found" result with the recorded "As Left" condition from the previous recalibration to evaluate the drift at several input values in the range of the sensor. If the sensor output is found to have drifted from the previous condition, then the sensor is adjusted to meet the prescribed "As Left" tolerances. However, this approach is expensive and time-consuming, and unnecessary maintenance actions can potentially damage sensors and sensing lines. Online monitoring (OLM) can help mitigate many of these issues, while providing a more frequent assessment of calibration and signal validation. However, widespread utilization of traditional OLM approaches is lacking with the need to better quantify OLM uncertainty a key factor in this.Sources of uncertainty in OLM can be roughly categorized as (1) process noise, (2) measurement uncertainty, (3) electronic noise, and (4) modeling uncertainty. Approaches to uncertainty quantification (UQ) that are data-driven may be capable of providing estimates of uncertainty that are time-varying as the quantities being measured vary with time. Such a capability provides the option of adjusting acceptance criteria and, potentially, setpoints in a time-varying fashion to meet the needs of the nuclear power system. A Gaussian Process (GP) model is proposed in this study for addressing the UQ issue. The advantage of this approach is the ability to account for spatial and temporal correlations among the sensor measurements that are used in OLM. The GP model, as proposed, may be considered an extension of a commonly used OLM model and, therefore, the hypothesis is that the UQ methodology may be readily extended to accommodate commonly used OLM models.Two approaches were taken for generating the data sets needed for evaluating the proposed model. Experimental data was acquired using an instrumented flow loop, with varying test conditions. In addition, a simulation model of ...

Quantification of Signal Reconstruction Uncertainty in Fault Detection Systems

“…The reconstructions provided by an empirical model, e.g., AAKR (Chevalier et al, 2009;Baraldi et al, 2010;Baraldi et al, 2012), are subject to the following 4 sources of uncertainty (Lin & Stadtherr, 2008;Ramuhalli, Lin, Crawford, Konomi, Braatz, Coble, Shumaker & Hashemian, 2013) , and will be used only when strictly required.…”