A machine learning approach to predict the critical heat flux for ignition of solid fuels

Abstract

The critical heat flux for ignition (CHFI) is an important measure of material flammability, often determined by experimental testing under an applied radiant heat flux. Observations of ignition are often interpreted alongside simplified thermal theories representing ignition as occurring once the surface of a material reaches an assumed critical ignition temperature tabulated for that material. This study proposes an alternative approach, by applying several Machine Learning (ML) algorithms to predict the CHFI based on known material properties and environmental variables. These models are trained using existing ignition data from a wide range of testing apparatuses, ignition scenarios and problem variables, including the Cone Calorimeter (ASTM), LIFT (ASTM E-1321), ISO Ignitability Apparatus, and the Idealized-Firebrand Ignition Test apparatus. The relevance of each variable is analyzed through a Feature Selection algorithm based on Decision Trees, Random Forest and Gradient Boosting. The algorithms are tested by comparing the resulting formulation with experimental results not used during the training of the models, showing good agreement in predicting the CHFI. The ML approach is capable of making predictions under a variety of conditions. It may be a useful tool for quick evaluation of new materials or changes in configurations.