Geometry parameters estimation in Monte Carlo simulation using machine learning.

This study investigates the application of Artificial Neural Networks (ANNs) to estimate geometry parameters in Monte Carlo simulations of High Purity Germanium (HPGe) detectors. The Comprehensive Nuclear-Test-Ban Treaty Organization utilizes HPGe detectors to measure radioactivity of airborne particles as part of its nuclear test monitoring regime. The International Data Centre employs Monte Carlo simulations to determine Full Energy Peak Efficiencies (FEPEs), but incomplete or imprecise detector geometry information often complicates this process. Currently, parameter estimation relies on iterative trial-and-error simulations, which are time-consuming and labor-intensive. To address this challenge, this study seeks a new methodology to reduce time and effort in the parameter estimation process. Planar and coaxial HPGe detectors were modelled using Geant4 and validated against both experimental measurements and a Monte Carlo N-Particle (MCNP)-based software package. Training datasets were generated by simulating FEPEs at a range of geometry configurations and gamma energies. Separate ANN models were trained for planar and coaxial detectors, with input features representing geometry parameters and output features corresponding to efficiencies. This model architecture was intended to control geometry parameters in ANN applications. Model evaluation using regression metrics demonstrated strong predictive performance. It suggested that ANNs offer a promising alternative for geometry parameter estimation in HPGe detector simulations.