부산대학교 도서관

Machine learning using quality indices does not require a large amount of data for model training to obtain accurate prediction results, which is conducive to the practical application of quality prediction in injection molding. However, previous approaches to model training were limited to case-by-case processing. When a new case emerges, data needs to be re-collected for training, which consumes a lot of time and cost. This study applies a transfer learning model to generate a quality prediction model based on the sensed in-mold pressure profile and features extracted from product quality. Later, in a similar situation, keeping the original model parameters, only a small amount of data is needed to update the model to perform the new goal. This study uses the weight prediction of two injection molded optical components of similar dimensions as a case study. Experimental results show that transfer learning saves model training time and demand for training data while achieving the desired goal.