Distortion Prediction of Wire-Arc Directed Energy Deposition using Physics-Informed Neural Networks
Please login to view abstract download link
Distortion remains a major barrier to the wider application of directed energy deposition (DED), primarily caused by steep thermal gradients and heat accumulation during deposition. To enable efficient and accurate prediction of temperature-induced distortion, a piecewise material deposition strategy is incorporated into the modelling framework to capture the thermal history and the resulting mechanical deformation. A physics-informed neural network (PINN) based method is employed to predict the transient temperature field, which is subsequently coupled with a thermo-mechanical finite element (FE) model to evaluate distortion. Experimental temperature measurements obtained from a wire arc DED thin-wall structure are used to formulate an inverse problem solved using physics-informed neural networks, enabling the identification of key thermal parameters. The identified parameters, including heat convection coefficients and thermal conductivity, are then consistently applied to temperature history prediction and distortion prediction. The close agreement between predicted distortion and experimental measurements validates the accuracy of the identified parameters and demonstrates the effectiveness of PINNs for distortion prediction in DED processes. This framework provides a foundation for real-time thermal and distortion prediction, supporting first-time-right manufacturing.
