Deep Learning for Neutron Transport

Overview

This analysis looks at the use of neural networks in solving the neutron transport equation in a one-dimensional slab geometry. The neural net solution is compared to traditional finite volume and Monte Carlo approaches. The PyTorch framework is used for neural net implementation.

Usage

Environment

You must first have conda installed before the environment can be built. To install the environment, run

make env

This command will create a conda environment named neutron_net (specified in the file environment.yml) with all Python packages required to run the analyses.

Running the analyses

All code can be ran and figures saved by running

make analysis

Alternatively, one can run all notebooks individually and explore intermediate results along the way.

ANNSlabSolver

The ANN code is implemented in a Python class called ANNSlabSolver, which is found in the file ANNSlabSolver.py. The user defines the problem parameters when instantiating an ANNSlabSolver object (see the class docstrings for all arguments).

An example usage is:

# Import the class
from ann.ANNSlabSolver import ANNSlabSolver

# Instantiate the class using default parameters
solver = ANNSlabSolver()

# Train the ANN
solver.train()

# Get an estimate of the scalar flux
flux = solver.predict()