Scholarlink

<p>Accurate prediction of the Maximum Detectable Distance (MDD) in mobile gamma spectrometry is essential for efficient surveying and reliable source detection. This study integrates a Physics-Informed Neural Network (PINN) with Monte Carlo N-Particle (MCNP) simulations to characterize how environmental and operational factors influence MDD. Local (OAT) and global (Sobol’) sensitivity analyses reveal nonlinear&#xa0;and asymmetric effects driven primarily by soil attenuation and detector velocity, while acquisition time exerts a dominant&#xa0;proportional influence (S<sub>1</sub> = 0.55). Uncertainty quantification, combining operational variability with stochastic Monte Carlo (MC) Dropout, shows that predicted MDD varies by ±5–10% under typical field conditions and up to ±15% in heterogeneous or shielded environments. By identifying the&#xa0;parameters that govern detection performance&#xa0;most strongly, this work provides a basis for developing adaptive survey strategies and improving decision-making in environmental radiation&#xa0;monitoring.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Sensitivity of input parameters in hybrid physics-informed neural networks for estimating maximum detectable distance in mobile gamma spectrometry

  • 期刊论文

  • 发表日期：

    2025年12月9日

查看全文
  • Nancy A. Ibrahim,
  • Amin Amirlatifi,
  • Somayeh Bakhtiari Ramezani

摘要

Accurate prediction of the Maximum Detectable Distance (MDD) in mobile gamma spectrometry is essential for efficient surveying and reliable source detection. This study integrates a Physics-Informed Neural Network (PINN) with Monte Carlo N-Particle (MCNP) simulations to characterize how environmental and operational factors influence MDD. Local (OAT) and global (Sobol’) sensitivity analyses reveal nonlinear and asymmetric effects driven primarily by soil attenuation and detector velocity, while acquisition time exerts a dominant proportional influence (S1 = 0.55). Uncertainty quantification, combining operational variability with stochastic Monte Carlo (MC) Dropout, shows that predicted MDD varies by ±5–10% under typical field conditions and up to ±15% in heterogeneous or shielded environments. By identifying the parameters that govern detection performance most strongly, this work provides a basis for developing adaptive survey strategies and improving decision-making in environmental radiation monitoring.