Markov Chain Monte Carlo Bayesian Unfolding Algorithm for Stilbene-Based Fast Neutron Spectrometry

Enabling the safe and secure deployment of advanced nuclear fuel cycles requires new methodologies for material control and accounting (MC&A). Neutron-based techniques enable non destructive characterization and assay of special nuclear material (SNM) in bulk samples. However, current safeguards and MC&A instruments do not exploit neutron spectroscopic information due to practical implementation challenges. We have developed a hierarchical Bayesian model coupled with a Markov Chain Monte Carlo (MCMC) sampling algorithm to solve the neutron spectral unfolding problem. The advantages of this approach include the use of an absolute detector response matrix to deconvolve the neutron fluence at the detector surface and the sampling of the a-posteriori unfolded spectrum, inherently providing associated confidence intervals (CIs) reflective of the uncertainty of the unfolded spectrum. To demonstrate the effectiveness of the Bayesian MCMC unfolding algorithm, the spectra from two PuBe sources (238PuBe, 0.4 Ci, and 239PuBe, 1 Ci) were simulated and unfolded. The neutron emission spectra from the PuBe sources were generated using SOURCES-4C, and the resultant light output was simulated using MCNPX-PoliMi. It was found that the unfolding algorithm successfully unfolded and resolved slight differences in the light output spectra, with spectral angles of 7.7° and 10.0° for the 238PuBe and 239PuBe spectra, respectively. These low spectral angles indicate that the unfolding algorithm recovered the major features of the spectra but was unable to resolve sharp turns and finer details. Still, we believe that with slight adjustments, the unfolding algorithm can be tuned to recover finer spectral details, enabling high unfolding accuracy and discrimination between SNMs with different compositions.