ANN-MoC Method for Inverse Transient Transport Problems in One-Dimensional Geometry

TL;DR

This paper introduces the ANN-MoC method, combining neural networks and a characteristic solver, to accurately estimate absorption coefficients in inverse transient particle transport problems, with promising results in homogeneous and heterogeneous media.

Contribution

The paper presents a novel approach integrating ANN with a Method of Characteristics solver for inverse transport problems, demonstrating high accuracy in both homogeneous and heterogeneous cases.

Findings

ANN-MoC achieves high accuracy in estimating absorption coefficients.

The method's success depends on the direct solver's solution quality.

Potential applicability to more complex inverse transport problems.

Abstract

The inverse problems of particle neutral transport models have many important engineering and medical applications. Safety protocols, quality control procedures, and optical medical solutions can be developed based on inverse transport solutions. In this work, we propose the ANN-MoC method to solve the inverse transient transport problem of estimating the absorption coefficient from measurements of the scalar flux at the boundaries of the model domain. The main idea is to train an Artificial Neural Network (ANN) from data generated by direct solutions computed by a Method of Characteristics (MoC) solver. The direct solver is tested on a problem with a manufactured solution. And, the proposed ANN-MoC method is tested on two inverse problems. In the first, the medium is homogeneous and has a constant absorption coefficient. In the second, a heterogeneous medium is considered, with the absorption coefficient constant by parts. Very accurate ANN estimations have been achieved for these two problems, indicating that the quality of the results relies on the accuracy of the direct solver solutions. The results show the potential of the proposed approach to be applied to more realistic inverse transport problems.