Rigid polyurethane foam (RPUF) is extensively used in insulation and structural applications. However, its high flammability, with a limiting oxygen index (LOI) of 18 %, presents significant safety concerns. Traditional flame-retardant optimization methods largely depend on trial and error, making them time-consuming and costly. In this study, a machine learning approach was developed to optimize flame-retardant design for RPUF. A curated database of 435 RPUF formulations containing reactive flame retardants was constructed using data from 89 published sources. This database consists of 806 input features, including 800 molecular descriptors generated using alvaDesc 3.0.6, which encompass topological, physicochemical, and electronic features, as well as six formulation-related variables such as additive content and curing conditions. A three-step feature selection strategy reduced dimensionality by 75.2 %, retaining 200 key descriptors. An Extreme Gradient Boosting (XGBoost) model was trained to predict LOI, achieving strong performance with a mean squared error (MSE) of 1.79 and coefficient of determination (R²) of 0.84. Feature importance analysis identified key predictive descriptors, while boxplot-based thresholding defined optimal descriptor ranges. Based on these insights, a phosphorus–boron-based reactive flame retardant (THPO-B) was synthesized and combined with a nitrogen-containing polyol (FRPN) at a 1:1 ratio (total loading: 26.0 wt.%). Experimental validation produced an LOI of 26.8 ± 0.3 %, closely matching the predicted value of 26.2 ± 1.34 %, with a relative error of 2.2 %, and representing a 43.2 % improvement over unmodified RPUF. This study demonstrates that machine learning can effectively guide the design of high-performance flame-retardant systems, offering a faster and more efficient alternative to conventional methods.
扫码关注我们
求助内容:
应助结果提醒方式:
