Prediction of the quality of plastic molded parts based on process parameters is gaining a lot of attention. The complexity of the injection molding process lies in the high number of parameters that intervene throughout the process. From the large set of process parameters, the key issue is to find the relevant ones that should be used to correctly classify the produced parts. In this work we compared algorithms from the three main families of feature selection methods: filter, wrapper, and embedded. Additionally, a hybrid approach was also evaluated that takes into account not only the supervised contribution but also an unsupervised method in an effort to evaluate each feature without the influence of the target label. Validation involves three different machines from three different brands working with three different processes and materials. A novelty in this paper is the inclusion of two injection molding processes: traditional injection molding and stretch blow injection molding. All datasets were provided by a plastic injection company in Portugal. The results show that despite the variability in materials and machines, there is a group of variables that should be monitored in all of these processes (including the two types of molding techniques). In addition to the process parameters, adding the ambiance temperature around the machine improves the representation of the process through the data.

Trabajo presentado en la 9th International Conference on Computational Intelligence and Virtual Environments for Measurement Systems and Applications (CIVEMSA), celebrada en Chemnitz (Alemania), del 15 al 17 de junio de 2022

Peer reviewed