LEAP: LEArning to Print – towards data-driven real-time predictions for additive manufacturing

by Nathanael Perrraudin

Wednesday 13 Dec 2023, 11:30 → 13:00 Europe/Zurich

OHSA/E13

ZOOM ID: https://psich.zoom.us/j/66861527191

Abstract:

Laser Powder Bed Fusion (LPBF) stands out as the predominant additive manufacturing process for producing high-performance metal components. This method constructs 3D parts by iteratively spreading and selectively melting a thin layer of powder. Within each layer, the 2D cross-section of the geometry undergoes melting through a moving laser spot, subsequently solidifying and bonding with the underlying layer. Determining the optimal scan pattern for a given part geometry remains an open question, often addressed empirically in practice.

In the LEAP project, we address this challenge by employing simulations to predict the geometry of the printed part based on the specified printing parameters. Notably, these simulations incur significant computational costs, rendering current simulators too slow for effective scanning strategy optimization. Consequently, within LEAP, we are actively developing machine learning (ML) surrogate models for the simulators using Physically Inspired Neural Networks (PINNs). These models enable us to efficiently navigate the parameter space and identify optimal scan strategies.