A novel model for predicting lower flammability limits using Quantitative Structure Activity Relationship approach

The lower flammability limit (LFL) is the minimum concentration of a combustible gas or vapor that is capable of propagating a flame through a homogeneous mixture with air, so knowledge of the LFL of a combustible gas is indispensable from a safety standpoint. However, it is often time-consuming and costly to measure the LFL of combustible gases by experiment, and is even precarious for toxic or explosive chemicals. Therefore, an effective predictive model is indispensable in this respect. In present work, a QSAR (quantitative structure activity relationships) model of four descriptors is proposed for predicting the LFL of organic compounds containing heteroatoms atoms such as S, N, Si, P and halogen. The proposed model is developed and evaluated based on a set of 458 experimental measurements of LFL. The performance of goodness-of-fit (R2) and predictive capability (Q2) are found to be 0.914 and 0.930, respectively. The average absolute errors (AAE) of training set (366 measurements) and test set (92 measurements) are 0.344 v % and 0.304 v %, respectively. According to the ASTM E 681 method, the experimental reproducibility in measuring LFL is of 0.1 v% for pure hydrocarbons, therefore AAEs of this model indicate that it could predict LFL with reasonable accuracy.