Study on cross-material property prediction of mineralizer fiber ceramic core materials based on deep learning

抄録

In this study, silicon-based ceramic cores were prepared using aluminum silicate fiber and mullite fiber as mineralizing agents through injection molding. A comprehensive properties evaluation model was proposed, optimizing the weight coefficient. The back-propagation (BP) neural networks were employed to predict the effects of different mineralizers on ceramic core properties, establishing a cross-material property mapping and prediction model. The results showed an abnormal increase in mean square error with the addition of 1 wt.% aluminum silicate fiber, indicating complex material behavior. Blending model predictions demonstrated the neural network’s strong capability in cross-material property prediction. The predictions closely matched experimental results, confirming the model’s accuracy.