Abstract: A SRF flame retardant performance prediction method based on deep neural network model (DNN) is proposed. The SRF samples were prepared and the initial data set was obtained by experimental characterization test. The DNN prediction model was optimized and compared with the other four basic models. The results showed that the limiting oxygen index increased to 28.6%, and the peak heat release rate and total heat release decreased by 24.83% and 24.7%, respectively, when the hydrogen-containing silicone oil and platinum catalyst were increased to 7.24% and 1.16%, respectively. The coefficient of determination of the best DNN model is 0.925. The validation set is set to analyze the prediction accuracy of the DNN model, and the relative error does not exceed 10.8 %. Finally, the interpretability between DNN prediction results and input variables is analyzed by partial dependence graph. The experimental parameters of flame retardant properties such as limiting oxygen index and heat release rate of SRF can be predicted by the model, which can effectively guide the rapid optimization of the properties of silicone rubber foam composites.

Key words: silicone rubber foam, machine learning, deep neural network, oxygen index, heat release