Time-of-flight based one-dimensional position estimation of radioactive sources using artificial neural network model

Abstract

This study presents a novel approach for one-dimensional gamma ray source position estimation by integrating plastic scintillating fiber technology, time-of-flight (ToF) measurements, and artificial neural network (ANN) techniques. The methodology employs a systematic signal processing framework consisting of constant fraction discrimination (CFD) for precise timing extraction, amplitude-based filtering for noise reduction, and statistical analysis of ToF data to enhance measurement consistency. A two-stage ANN architecture was developed incorporating dual hidden layers with ReLU activation functions and weighted correction factors to optimize spatial localization performance. The system was experimentally validated using a Cs-137 radiation source across a 10-m measurement range with data collected at both regular intervals and random positions to assess interpolation capabilities. Comparative analysis between the ANN-based approach and theoretical calculations demonstrated a 90.17 % enhancement in position estimation precision, achieving an average error of 0.0225 m compared to 0.2289 m with conventional methods. Standard deviations in position estimates remained consistently below 0.1 m across the operational range, indicating robust performance stability. These results substantiate that combining sophisticated timing measurements with machine learning strategies advances radiation detection systems applicable to environmental monitoring, nuclear safety protocols, and emergency response scenarios.