Work

Investigation of transfer learning with changing machine, mold and material combinations in injection molding

Year: 2024

Type: article

Abstract: One disadvantage of machine learning for quality prediction in injection molding is that when the machine, mold or material used changes, new data must be collected to build accurate prediction models. One solution is to use transfer learning, where pre-trained models are reused and adapted to new problems. The objective of this study was to evaluate the benefits of transfer learning. Therefore, two injection molding machines, two molds and two materials were used. Five transfer learning methods were evaluated, resulting in 35 models. It could be shown that the application of the transfers led to an overall higher prediction performance.

Source: Procedia CIRP

DOI: 10.1016/j.procir.2024.08.295

Authors: Dimitri Kvaktun, Felix Müller, Reinhard Schiffers

Institution: University of Duisburg-Essen

Language: English

Topic: Injection Molding Process and Properties

Subfield: Mechanical Engineering

Field: Engineering

Domain: Physical Sciences

SDG: Industry, innovation and infrastructure

Open Access status: diamond

Funders: Bundesministerium für Wirtschaft und Klimaschutz, Bundesministerium für Wirtschaft und Energie