論文ID: 24107
Silicon-based ceramic cores with aluminum silicate particles and fibers as mineralizers were prepared separately using the injection molding method. A comprehensive property evaluation model was proposed based on the physical property parameters of the ceramic cores, which were optimized using weight coefficients. A prediction model based on a backpropagation neural network was constructed to predict the effects of aluminum silicate fibers and particles as mineralizers on the properties of ceramic cores, and a properties optimization study on single materials were conducted. Additionally, a blending model was established to achieve cross-material properties predictions. The results indicate that the optimization predictions for single materials were more precise than those obtained through traditional experimental methods. The cross-material blending model identified an abnormal increase in the mean squared error at an addition amount of 1 wt.% aluminum silicate fibers, suggesting the presence of complex material behavior at this point. The full combination prediction results demonstrate that the neural network model performs well overall in predicting cross-material properties. The results of cross-material prediction show that the prediction results are in good agreement with the experimental measurement results, which indicate that the cross-material property parameter mapping and prediction of the neural network has good accuracy.
