(203d) Minimum Ignition Energy (MIE) prediction from QSPR using machine learning

Dust and gas explosions pose a serious hazard to process industries, resulting in loss of life, property and resources. The Minimum Ignition Energy (MIE), defined as the smallest amount of energy required to ignite a flammable compound at a given temperature and pressure, is a critical safety parameter of significant importance. However, often the high value materials of concern are only available in small quantities, making destructive testing impractical during early stages of development. Because of this prediction methods should be explored and advanced.

In this study, the MIE prediction of 60 flammable hydrocarbon compounds has been conducted using a Quantitative Structure-Property Relationship (QSPR) and machine learning techniques. The prediction models were developed using Random Forests (RF), Decision Trees (DT) and Artificial Neural Networks (ANN) resulting in promising (> 0.70) R² values for the test sets. Decision trees were used to identify the 10 most important molecular descriptors influencing the MIE prediction model accuracy. In addition, a Genetic Function Approximation (GFA) algorithm in Materials Studio was used to develop a 10 parameter MIE prediction equation resulting in significant R² value. The GFA, RF and DT algorithms resulted in a more robust MIE prediction model as compared to ANN algorithm.