Bayesian Calibration and Sensitivity Analysis of Rayleigh Scattering Fiber Optic Distributed Temperature Sensing in Water Flow Loop

This study presents a methodology for the calibration and sensitivity analysis of temperature measurements with Rayleigh scattering distributed fiber optic sensors (DFOSs) using Bayesian inference and sensitivity analysis techniques. Traditionally, thermocouples are used to measure temperature within reactors. While thermocouples are well suited for highly radioactive environments, a penetration location across the vessel is required for each measurement location, increasing risk with the reactor. Also, thermocouples have often experimentally shown to experience drift, requiring additional calibration past the initialization. The approach uses the Delayed Rejection Adaptive Metropolis algorithm, a Markov Chain Monte Carlo sampling technique, to calibrate key physical model parameters based on reference thermocouple measurements. Compared to prior model predictions, Bayesian calibration achieved a substantial reduction in root-mean-square error from an uncalibrated average of 1.83 deg to a posterior average of 0.92 deg.

In addition to calibration, the study performs a global sensitivity analysis using Sobol’, Morris, and correlation coefficients to determine which of the measuring locations are most important for calibration. It is ideal to reduce the number of co-located thermocouples for future validation study of the DFOS system used as the ground truth. Although all sensitivity measurements were consistent, Sobol’ indices provided the clearest hierarchical ranking and should be prioritized in future analysis.

This work demonstrates the potential of Bayesian calibration to improve DFOS performance in complex thermal-fluid environments, with direct relevance to instrumentation strategies for advanced nuclear systems. The presented framework supports higher-fidelity temperature estimation while offering uncertainty-aware tools for sensor validation in experimental test loops.