Prediction of Injection Molding Parts Weight Basing on Electrical Time Series Signals Combining With Process Parameters

ABSTRACT Predicting the weight of injection‐molded products is important for systematically improving production efficiency and process stability. However, the data acquisition methods currently used for product weight prediction often affect product quality and reduce mold lifespan. In this study, electrical signals collected during the molding process were used as input data for product weight prediction. Long Short‐Term Memory (LSTM) networks were employed to improve the prediction accuracy of long‐term time‐series signals generated during the injection molding process. Meanwhile, a Transformer model was introduced to preprocess the input data for the LSTM network and was further combined with hard sample mining and weighted loss fine‐tuning to optimize the prediction model. The results demonstrate that accurate product weight prediction can be achieved by incorporating low‐level process parameters, using 12‐dimensional input features, and applying appropriate model architectures and training strategies. The final model achieved RMSE and MAE values of 0.0333 and 0.0261, respectively. Compared with the 26‐dimensional input configuration, the RMSE and MAE were reduced by about 32% and 28%, while the prediction accuracy reached 97%. The proposed method provides a foundation for quality prediction based on injection molding signals for digital transformation and benefit for future process monitoring and production stability in injection molding.