Development of Neural-Network-Based Digital Pulse Processing for Photoneutron Detection

Abstract

High-energy photon interrogation is a non-destructive technique that is used to detect hidden special nuclear materials (SNMs) and characterize nuclear waste. The development of such systems is complex and requires Monte Carlo simulations to optimize system performance. Monte Carlo simulations rely on various scattering, absorption, and photonuclear cross-section data. The scattering and absorption cross-section data for neutrons and photons has been extensively studied and validated with experiments because of their importance in nuclear reactor and radiation shielding simulations. However, photonuclear cross-sections are not extensively studied and lack the desired validation with measured results. The under-prediction by Monte Carlo codes can range from 20-30%, and therefore there is a need to validate photonuclear cross-section data with precise new measurements. The present Ph.D. research provides new measured results for photoneutron count rates from various high-Z targets. The measured results were compared with the simulated results obtained by Monte Carlo codes. The simulations were performed using the MCNPX-PoliMi transport code with the most updated photonuclear cross-section data. For measurements of photoneutrons, several high-Z targets were interrogated with bremsstrahlung photons from a 9-MV electron linac, and fast neutrons were detected with four trans-stilbene organic scintillators. The comparative study between measurement and simulation provided a quantitative assessment of the under-prediction in the photoneutron count rates by MCNPX-PoliMi. During interrogation of targets, the intense bremsstrahlung photons from the linac creates significant pulse pile-up in trans-stilbene. The pulse pile-up effect was mitigated by developing an artificial neural network (ANN) system for digital processing of scintillation pulses. The developed ANN system outperformed traditional pulse shape discrimination methods during an active 252Cf measurement. In this measurement, prompt fission neutrons were measured from a 252Cf spontaneous fission source in the presence of the intense photon flux from the linac, imitating a challenging radiation environment for the measurement of fast neutrons. Photoneutrons were measured from a SNM surrogate, such as depleted uranium (DU), and a non-SNM target, such as lead. The results obtained from the developed ANN system showed a 5x increase in the photoneutron count rate when lead target was replaced with the DU target. Additionally, the light output distribution for lead photoneutrons was softer than the light output distribution for DU photoneutrons. This difference in the light output distributions is a new result because there exists no prior work in the literature that detected photoneutrons without time-of-flight and coincidence counting. The DU target was further interrogated in various iron and polyethylene shielded configurations. The measured photoneutron count rates decreased with an increase in shield wall thickness. This decreasing trend is due to the moderation of photoneutrons by polyethylene, and the attenuation of bremsstrahlung radiation by iron. The interrogation measurements of DU in bare and shielded configurations were simulated using the developed MCNPX-PoliMi framework. For a light output window of 0.28 – 2.67 MeVee (1.66 – 6.85 MeV proton recoil energy), the simulated photoneutron count rate under-predicted the measured rate by 32.8 ± 3.2%. The findings from this work provide new measured results that can help improve photonuclear cross-section data for uranium, which in turn will enhance simulation capabilities with existing Monte Carlo codes.