Plutonium mass estimation can be performed using neutron coincidence counting and the neutron point model. Usage of this model requires input from either known values or values acquired from simulations. Item simulations of MCNP use known mass and isotopics but unknown doubles rate, leakage multiplication value, and (alpha, n) rate. Each of these parameters requires a separate MCNP run. This work trains a random forest regression model to predict the leakage multiplication value of a sample using geometric variables, obviating the need for an additional MCNP run. The resulting leakage multiplication prediction from the random forest model better estimates the expected doubles rate from a sample of known effective mass, radius, and height than the neutron point model and weighted neutron point model. To train the random forest regression model MCNP runs for 2533 cylinders (20 radii, six enrichments, and up to 25 heights) were done to determine the shape of the leakage multiplication curve within each sample. To do this, the sample was broken into ten annuli, with a source thickness of R/10 moving outward. A second-order polynomial was fit to each sample. The coefficients of this polynomial were used to determine the function of best fit for each coefficient as a function of radius-to-height ratio. The roots of the polynomial, and the integral and derivative leakage multiplication values were also calculated. A random forest model was trained to predict the roots of the polynomial from the radius, height, volume, and radius-to-height ratio. The ability to predict the expected doubles rate from a sample could aid safeguards verification. The declared mass of an object can be used to predict the expected doubles rate, which can then be compared to the measured doubles rate. Deviations between the predicted and compared rates can be used to verify the declarations for the item.
Year
2020
Abstract