Metal Additive Manufacturing Conference - MAMC 2022

Most of the research on the numerical simulations of Additive Manufacturing (AM) processes have focused on the detailed Finite Element (FE) simulations to optimise these manufacturing processes and to calibrate their process parameters. Although, many details of the dynamic and transient multi-physical processes, like AM processes, are required to be simulated to obtain accurate results, the amount of computing time and efforts are still challenging even for the today’s highly parallelized computing schemes. Meanwhile, with the manufacturing digitalisation transformation and its real-time modelling requirements, there are prerequisites for fast and reliable predictive and corrective models to handle real-time optimisations. The hybrid and reduced physical-data driven modelling schemes have already been employed in some digital twining technologies to reduce the simulations’ times towards the real-time scale. This research work presents the overview of the hybrid and Reduced Order Modelling (ROM) schemes for AM process modelling where quick and agile models can be created to handle the multi-physical and dynamic nature of these processes. Additionally, the ways to combine the Machine Learning (ML) technologies with ROM scheme to estimate the optimized process parameters and improve the product qualities are elaborated and pilot case studies are presented to show the applicability and accuracy of these hybrid techniques.

Keywords: Reduced order models, machine learning, additive manufacturing, hybrid modeling, digital twin, data driven models