Neural Operators for Spark Plasma Sintering Simulation

Abstract

Over the last two decades, Spark Plasma Sintering (SPS) has become a major technique for manufacturing advanced materials. But mastering SPS process is complex and, for a better understanding, requires a computational calculation support. SPS simulations are mostly conducted using Finite Element Method (FEM). However, FEM approaches are well-known for being time and cost-consuming due to high computational complexity. On the other hand, the emerging approaches such as Deep Learning (DL) could be considered as a viable alternative to traditional modelling and has proven their effectiveness in many fields. But classic DL methods are not well suited for approximate data based on Partial Differential Equations (PDE). Thereby, the interesting properties of Neural Operator (NO) lies in the ability of the network to implicitly learn the underlying PDE operators hidden in the data. In this study, we investigated the potential of emerging NO approach as a faster alternative to SPS FEM models.