Rossi-Alpha distribution analysis of DDSI data for spent nuclear fuel investigation

The differential die-away self-interrogation (DDSI) prototype instrument is a neutron coincidence instrument developed to investigate nuclear safeguards-relevant properties of spent nuclear fuel (SNF). The distribution of the time intervals between neutrons detected in coincidence creates the Rossi-Alpha distribution (RAD). The RAD depends on properties of the SNF such as the net multiplication. In this work, a new technique to analyze the RAD and infer SNF parameters is investigated. In previous analysis, the simulated RAD was fitted with a single exponential between a time window of 4 and 52 µs to determine the so-called early die-away time (tearly). The tearly has been found to be correlated to the net multiplication. However, tearly is sensitive to the choice of the time window, and an analysis of measurement data collected from SNF at the Central Interim Storage Facility for Spent Nuclear Fuel (Clab) in Sweden revealed that the optimal time window identified by simulations could not be used. In addition, tearly is not very sensitive to burnup (BU) and initial enrichment (IE). Therefore, an analysis technique that is not dependent on fixed time windows and is more sensitive to BU and IE is explored. In this work, a function consisting of two exponential functions is fitted to the whole RAD. The method is applied to both simulated and experimental data. Results from experiments and simulations are compared, and significant discrepancies in the RAD-parameters are found in line with previous work using a single exponential. The four parameters determined by the fit of the double exponential (two amplitudes and two time constant) are used as input parameters in a machine learning algorithm employing a k-nearest neighbours algorithm to infer BU and IE. In addition, a method to correct for discrepancies between predicted parameters and operator data was developed. It is found that BU and IE can be predicted with a RMSE of 2.69 GWd/tHM and 0.104 % on the simulated dataset.