JavaScript is disabled for your browser. Some features of this site may not work without it.

Hybrid Monte Carlo/Deterministic Methods for Radiation Shielding Problems.

Becker, Troy L.

Becker, Troy L.

2009

Abstract: For the past few decades, the most common type of deep-penetration (shielding)
problem simulated using Monte Carlo methods has been the source-detector problem,
in which a response is calculated at a single location in space. Traditionally,
the nonanalog Monte Carlo methods used to solve these problems have required
significant user input to generate and sufficiently optimize the biasing parameters
necessary to obtain a statistically reliable solution. It has been demonstrated that
this laborious task can be replaced by automated processes that rely on a deterministic
adjoint solution to set the biasing parameters – the so-called hybrid methods.
The increase in computational power over recent years has also led to interest in
obtaining the solution in a region of space much larger than a point detector. In this
thesis, we propose two methods for solving problems ranging from source-detector
problems to more global calculations – weight windows and the Transform approach.
These techniques employ some of the same biasing elements that have been used
previously; however, the fundamental difference is that here the biasing techniques
are used as elements of a comprehensive tool set to distribute Monte Carlo particles
in a user-specified way. The weight window achieves the user-specified Monte Carlo
particle distribution by imposing a particular weight window on the system, without
altering the particle physics. The Transform approach introduces a transform
into the neutron transport equation, which results in a complete modification of the
particle physics to produce the user-specified Monte Carlo distribution.
These methods are tested in a three-dimensional multigroup Monte Carlo code.
For a basic shielding problem and a more realistic one, these methods adequately
solved source-detector problems and more global calculations. Furthermore, they
confirmed that theoretical Monte Carlo particle distributions correspond to the simulated
ones, implying that these methods can be used to achieve user-specified Monte
Carlo distributions. Overall, the Transform approach performed more efficiently
than the weight window methods, but it performed much more efficiently for source-detector
problems than for global problems.