Transfer-Learning: Bridging the Gap between Real and Simulation Data for Machine Learning in Injection Molding

Abstract In the field of manufacturing process planning and initial operation of machines, machine parameters are often provided from few either expensive and time-consuming experiments or faster but less accurate numerical simulations. Another option is to use machine learning to predict process qualities based on machine parameters. Thereby, transfer learning can overcome the gap between real and simulation data. We evaluated two different approaches based on artificial neural networks, namely soft-start and random initialization, in a real injection molding process. The results show better learning rates and predictions that are more accurate while using fewer experimental data.