CFAST simulations and application of a machine learning algorithm for the fire safety of a switchgear room

A cabinet fire in the Motor Control Center (MCC) of a switchgear room in a nuclear power plant can cause severe equipment damage, potentially leading to critical system failures and posing significant safety risks. This study explores the use of neural networks and Consolidated Fire and Smoke Transport (CFAST) simulations as alternatives to traditional physics-based modeling tools. The proposed neural network is designed to rapidly predict cable and cabinet damage times without relying on the time-intensive CFAST simulations. The CFAST model simulates fire scenarios in a closed switchgear room, using key input parameters such as room dimensions (width, depth, height for rooms 1 and 2), air exchange rate, and ambient temperature. Target outputs include the surface temperature and heat flux of cables and cabinets, which serve as training data for a machine learning model to predict damage times. Results indicate that room 1's height has the greatest impact on cable and cabinet damage, with taller rooms experiencing faster deterioration. The neural network’s predictions closely align with CFAST simulation results, demonstrating both accuracy and efficiency in fire damage assessment.