Predicting critical flame quenching thickness using machine learning approach with ResNet and ANN

In the study of flame quenching, quenching thickness is one of the important parameters to determine the design of a flame arrester, and often determines the flame quenching performance of the arrester. In the study, residual network (ResNet) and artificial neural network (ANN) are used to predict the critical quench thickness of combustible gas in pipelines. The critical quench thickness is influenced by fuel concentration and density, pipeline size, inert gas type and concentration, porous media porosity, and thermal conductivity. The influence of different combinations of hyper-parameters on the prediction performance of the two models is explored. The results show that the prediction performance of both models reaches the best after hyper-parameter optimization. Compared with ANN, the ResNet model shows more stable and better prediction ability, and its optimal evaluation parameters are: MAE is 1.4679, MSE is 91.7431, R² is 0.9216. The prediction errors of the two models on the same dataset are subjected to analysis, and the impact of the use of normalized data on the performance of the two models is compared. It is determined that the ResNet model demonstrated superior robustness and generalization ability in predicting the critical quenching thickness of combustible gases. The study is helpful for the safety protection of combustible gas and the safety design of pipeline arresters.