INVERSE PROBLEMS FOR STOCHASTIC NEUTRONICS

Corentin Houpert, Josselin Garnier, Philippe Humbert

Abstract:

Fissile matter detection and characterisation are crucial issues; especially in nuclear
safety, safeguards, matter comptability, reactivity measurements. In this context, we want to
identify a source of fissile matter knowing external measures such as instants of detection of
neutrons during an interval of measure. Thus we observe the neutrons detection times emitted by
the fissile matter and going through the detector, then we compute the moments of the empirical
distribution of the number of neutrons detected during a time gate T. In order to identify the
source we have to get the following parameters: the multiplication factor k of the system, the
intensity of the source S, the fission efficiency εF .
Given the parameters of the source there are some models that allow us to predict the moments of counted number of neutrons during a time gate T. We consider a point model stating monokinetic neutrons are moving in an infinite, isotropic and homogeneous medium. The
method makes it possible to compute the first moments of the count number distribution.
Then, given the moments of counted number of neutrons during a time gate T we want to
get the parameters of the fissile source. In order to achieve this goal, we will use a Bayesian approach in order the get the distribution of parameters. The a posteriori distribution is non-trivial, samples can be achieved with Markov Chain Monte-Carlo methods
with covariance matrix adaptation (MCMC with CMA).