Knowledge of flammability limits values, including both the lower flammability limit (LFL) and upper flammability limit (UFL), is essential to maximise safety in process design and operational procedures. This study aims to develop theoretical models for predicting both lower and upper flammability limits for fuel mixtures from molecular structures with the same molecular descriptors. All samples were collected from a single reference and the “compounds out” strategy was employed to divide the training and the test set. Descriptors for pure compounds were calculated by Gaussian 16 and Multiwfn package. The developed LFL and UFL models were validated with rigorous internal and external validations. The results indicated the robustness, validity, and satisfactory predictivity of the established models. The applicability domain (AD) of the model was also defined. The established LFL and UFL models demonstrated that the higher dipole moment and van der Waals volume are, the higher LFL and lower UFL values are. The proposed models can be expected to predict the LFL and UFL values for other fuel mixtures for which experiment values are unknown, only based on molecular structures.

了解可燃性限值，包括可燃性下限 (LFL) 和可燃性上限 (UFL)，对于最大限度提高工艺设计和操作程序的安全性至关重要。本研究旨在开发理论模型，用于从具有相同分子描述符的分子结构预测燃料混合物的燃烧下限和上限。所有样本均从单个参考中收集，并采用“化合物输出”策略来划分训练集和测试集。纯化合物的描述符由 Gaussian 16 和 Multiwfn 包计算。开发的 LFL 和 UFL 模型经过严格的内部和外部验证。结果表明所建立模型的稳健性、有效性和令人满意的预测性。还定义了模型的适用性域 (AD)。建立的 LFL 和 UFL 模型表明，偶极矩和范德华体积越大，LFL 值越高，UFL 值越低。可以预期所提出的模型仅基于分子结构来预测实验值未知的其他燃料混合物的 LFL 和 UFL 值。