Generic flame extension model development based on machine learning for NPPs fire hazard analysis (FHA)

Abstract

In enclosed fire situations, the flame interacts with ceiling and its length extension take place, altering the fire severity significantly. Fire safety analysis demand a generic flame extension model. A generic reliable model is not available as the construct of flame itself had wide variation in literature. The present paper aims to develop a Machine Learning (ML) based generic model of non-dimensional flame extension as a function of non-dimensional Heat Release Rate (HRR). The non-dimensional scaling reduces the number of parameter and also provide a generic nature. Literature review was utilized to collect the data from various open literature sources. This eliminates the limitations of individual correlations and gives a best optimized model which is valid for a wide range of flow regimes and conditions as compared to a specific correlation. Various simple ML models are compared for their performance against test data and a MARS based model was finally recommended. The MARS model was tested against the data which was not used in training and also against the other reported correlation. The developed model has performed well against the test data and marked improvement over other reported correlation as a better optimized performance over an extended range of non-dimensional range. The results of the model were also conservative as compared to another model in the most of the practical requirement of NPPs. A Large Eddy Simulation (LES) based CFD code FDS was also used to generate the flame extension data for demonstrating conservative nature of the ML model.