Prediction of lower flammability limits of blended gases based on quantitative structure–property relationship

This work focuses on developing predictive quantitative structure−property relationship (QSPR) models for lower flammability limits (LFLs) of gas mixtures. Experimental LFLs of 86 blended gases were extracted from a single reference, and their mixture descriptors were calculated solely from individual molecular structure by Gaussian 09. Multiple linear regression (MLR) analysis was employed to develop the models, and three different external validation methods were carried out to check the predictive capabilities of the models. The validations have shown that these models possess great predictive power with excellent goodness of fit and internal robustness; hence, they are deemed to be qualified to predict LFLs for other blended gases with no experimental LFL data available. The applicability domains (AD) of the models were defined as well, and all the points were within the AD area. The main advantages of the established models are their simplicity and possibility of extending them for the determination of LFLs of other gas mixtures, while the LFLs of the individuals do not need to be provided.