In metal additive manufacturing processes such as laser powder-bed fusion (LPBF) a high-intensity laser heat source is used to fuse metal powder and produce parts with high geometrical complexity in a layer by layer fashion. During this process, the alloys experience high cooling rates and large thermal gradients that result in the development of residual stresses and a unique microstructure, which in turn influence the mechanical properties of the products. Therefore, studying the thermal history is an essential step in fully understanding the LPBF build process and optimizing the performance of the parts. However, conventional analysis techniques such as the finite element method for thermal analysis of LPBF are faced with expensive computational costs that limit the scope of their application.   As a powerful alternative, this study evaluates the applicability of physics-informed neural networks (PINNs) for such analyses. PINNs are a class of deep neural networks that can be trained to compute the behaviour of physical systems governed by partial differential equations (PDEs) via automatic differentiation and an unsupervised or semi-supervised learning strategy. This study shows that a well-trained PINN can be an accurate and computationally efficient replacement for FE thermal analysis of the LPBF process.

Download the slides for this talk.Download ( PDF, 2895.02 MB)