Predicting part quality in injection molding using artificial neural networks

This paper presents a method for predicting the quality of an injection molded part, defined in this study as its weight, using on-line measurements of key processing variables. By focusing on the behavior of the polymer melt in the nozzle by means of its pressure and temperature, it is possible to accurately predict the weight of the final molding. The approach used analyses complete cycle profiles of the two variables by training a neural network to recognize differences in their data patterns. Two experiments were conducted to collect process data that was used to test the ability of artificial neural networks (ANNs) to model the relationship between process changes and part weight. The results presented show that the networks were able to learn the relationships captured within the cycle profiles and succeed in predicting part weight accurately. Such a strategy has a number of potential advantages in the on-line prediction and assurance of part quality in injection molding.