Theses and Dissertations

ORCID

https://orcid.org/0009-0003-0943-2803

Advisor

Amirlatifi, Amin

Committee Member

Bakhtiari Ramezani, Somayeh

Committee Member

Heshmati, Mohammad

Committee Member

Elmore, Bill B.

Date of Degree

12-12-2025

Original embargo terms

Visible MSU Only 6 months

Document Type

Dissertation - Campus Access Only

Major

Engineering (Chemical Engineering)

Degree Name

Doctor of Philosophy (Ph.D.)

College

James Worth Bagley College of Engineering

Department

Dave C. Swalm School of Chemical Engineering

Abstract

Accurate estimation of the Maximum Detectable Distance (MDD) is critical for environmental remediation, nuclear incident response, and public safety. Conventional approaches assume stationary detectors and rely on computationally intensive simulations, limiting their applicability for mobile, real-time surveys. Mobile gamma spectrometry introduces additional complexities due to detector motion, variable acquisition times, and heterogeneous backgrounds. This study addresses the central question of how effectively a Physics-Informed Neural Network (PINN) can estimate MDD for mobile gamma spectrometry, while maintaining both accuracy and computational efficiency under dynamically changing environmental conditions. A multi-phase approach was employed, combining field data collection, computational modeling, and operational tool development. A hybrid PINN framework was developed, embedding radiation transport laws directly into the network’s loss function. By enforcing geometric spreading, attenuation, and boundary constraints, the PINN integrates first-principles physics with data-driven learning. The network produces physics-consistent predictions of MDD that account for variable background radiation, source shielding, and detector velocity. It was trained on field gamma counts from a 137Cs-contaminated Superfund site and validated using Monte Carlo N-Particle Transport Code (MCNP) simulations calibrated with measured attenuation coefficients. Field validation shows that the PINN consistently predicts MDD within 0.54 m of observed values, accurately modeling the combined effects of variable background, shielding, and source strength. It reproduces the expected decrease in MDD with increasing detector velocity and accounts for orientation-dependent variations. Sensitivity analyses show that acquisition time and detector speed are the dominant factors influencing detection capability. An MDD Calculator prototype demonstrates that these predictions can be directly applied in operational settings, enabling a real-time decision-support tool for remediation teams. By integrating radiation transport physics, machine learning, and operational constraints, this research establishes a physically interpretable, field-validated framework and represents a significant advancement over conventional simulation-based or purely data-driven approaches to mobile gamma spectrometry.

Download

Share

COinS