Physics-informed neural network approach to speed up Laser-DED modelling
Sebastian Hartmann, Oihane Murua, Jon Iñaki Arrizubieta, Aitzol Lamíkiz, Peter Mayr
Abstract
The extreme physical conditions during laser-based Directed Energy Deposition (DED-LB) hinder the quick adoption of novel process strategies. Numerical techniques, such as the Finite Element Method (FEM), have proven to accelerate this adoption process by enabling the prediction of the process conditions before the printing process. While showing high accuracy, these techniques also imply intensive computation and long processing times. The present work investigates the integration of physical laws into the loss function of a neural network and evaluates different loss function designs. The resulting physics-informed neural network (PINN) can predict the temperature field during the DED-LB process and is tested with a validated FEM simulation. The novelty of this approach lies in applying the PINN to complement the FEM simulation for predicting the temperature field of additional time steps. The PINN closely resembles the simulation results, yielding an average error of 2.5 Kelvin for every point in the simulation domain.