Neural Network–Based Prediction of Mold Temperature Distribution Using a Cooling Layer During Cavity Heating in Injection Molding

Abstract

In the injection molding of thin-walled plastic parts, premature solidification of the polymer melt upon contact with the relatively cold mold surface reduces cavity filling capability and adversely affects product quality. Therefore, proper control and distribution of mold cavity temperature during the heating stage play a crucial role in improving melt flow behavior without significantly extending the injection molding cycle. However, studies focusing on layered heating channels for mold cavities and the application of artificial intelligence methods for predicting temperature distribution remain limited. This study investigates the feasibility of using artificial neural networks (ANN) to predict the temperature distribution of an injection mold cavity equipped with a layered heating channel. Temperature data were collected during the mold heating process and used to develop and train an ANN model, which was further compared with a random forest model. When the number of neurons in the hidden layer was increased from 7 to 150, MSE decreased from 15.2575 to 0.6670, while the R_all increased from 0.9579 to 0.9981. Meanwhile, the Random Forest (RF) model also achieved a low prediction error, with MSE values ranging from 0.24 to 0.64 and R_all of 0.9991. The results indicate that the artificial neural network achieves high prediction accuracy and effectively captures the nonlinear relationships governing heat transfer, demonstrating strong potential for application in mold temperature optimization in injection molding processes.