Experimental analysis of combustion characteristics of corn starch dust clouds under the action of unilateral obstacles and machine learning modeling based on PSO-XGBoost

Abstract

Corn starch powder is highly flammable and explosive, presenting significant safety hazards of dust explosions when encountering obstacles during its production and processing. This study indicate that with an increase in the number of obstacles, obstacle blockage ratio, and dust concentration, both the average flame spread velocity (AFSV) and the maximum flame spread velocity (MFSV) initially rise and then decline. However, the presence of obstacles significantly enhances both MFSV and AFSV compared to the absence of obstacles. Additionally, Using the Extreme Gradient Boosting (XGBoost) algorithm, predictive models for the MFSV and AFSV of corn starch dust were developed. By employing the Particle Swarm Optimization (PSO) algorithm for hyperparameter tuning, the model achieved an coefficient of determination (R²) of 0.9821 for MFSV and 0.9687 for AFSV, enabling highly accurate flame spread velocity (FSV) predictions. Random Forest importance analysis revealed that obstacle characteristics exert a more pronounced impact on FSV.