Machine Learning-Driven Uncertainty Quantification and Parameter Analysis in Fire Risk Assessment for Nuclear Power Plants

Fire poses a critical risk to the safe operation of nuclear power plants (NPPs), with electrical cabinet and liquid spill fires being among the most frequent and challenging scenarios to address. Traditional fire risk assessment models often lack precision due to complex physics and inherent uncertainties, especially in predicting the heat release rate (HRR) — a key parameter for assessing fire severity. This dissertation presents an innovative framework that integrates machine learning (ML) models, particularly neural networks and tree-based algorithms, with uncertainty quantification (UQ) techniques to enhance fire modeling and risk assessment in NPPs. The framework is applied to electrical enclosure cabinets and spill fires that represent about 50% of challenging fire scenarios in NPPs. By leveraging extensive experimental datasets, this study develops ML models that capture the influence of critical fire parameters on HRR, enabling more accurate predictions of fire behavior. Key features are evaluated to establish their influence on peak HRR. Advanced UQ tools, including Monte Carlo sampling and sensitivity analysis are applied to quantify uncertainties and identify parameters with the greatest impact on model output variability. The resulting ML-driven insights allow for a refined understanding of fire dynamics, guiding experimental planning and uncertainty reduction efforts. For electrical enclosure fires, the models highlight the importance of cable surface area, heat release rate per unit area of the cable, ignition source heat release rate, ventilation area, and cabinet volume in determining peak HRR. Sensitivity analysis revealed that HRRPUA is the most significant parameter. For spill fires, the models underscore the significance of substrate thermal conductivity and slope, ignition delay time, and fuel properties, showing that fuel amount and properties are key in fixed quantity spills, while fuel discharge rate and properties are most influential in continuous spills.

Name thesis:degree_name

Doctor of Philosophy

Level thesis:degree_level

doctoral

Discipline thesis:degree_discipline

Nuclear Engineering

Department dc:contributor.department

Mechanical Engineering

Grantor dc:publisher

Virginia Tech

Year dc:date.issued

2025

Author dc:creator

• Sahin, Elvan

Chair dc:contributor.committeechair

• Pacheco Duarte, Juliana

Committee members dc:contributor.committeemember

• Liu, Yang
• Lattimer, Brian Y.
• Wu, Zeyun

• Fire
• Machine Learning
• Uncertainty Quantification
• Sensitivity Analysis
• Probabilistic Risk Assessment
• Monte Carlo

Statement dc:rights

• In Copyright

Licence dc:rights.uri

• http://rightsstatements.org/vocab/InC/1.0/

Language dc:language.iso

en

Dc Identifier Other

vt_gsexam:42490

OAI identifier oai:identifier

oai:vtechworks.lib.vt.edu:10919/124406