Physics-Informed Machine Learning of Argon Gas-Driven Melt Pool Dynamics

However, despite its potential, metal AM has not yet reached its expected level of usage in industries, in part due to a lack of accurate prediction of the properties of printed components. For example, in laser powder bed fusion (LPBF), the layer of metal powder is scanned by a laser heat source which converts the metal powder to liquid, which eventually solidifies and converts to the final product. Accurate thermal history prediction is crucial for LPBF, as all other phenomena, including thermal residual stress and microstructure, depend on it. The melt pool dynamics play a very important role in the development of the thermal map for LPBF. Many factors influence the melt pool dynamics in LPBF such as the unique thermal cycle of rapid heating and solidification, steep temperature gradient and high cooling rate, evaporation, surface tension, natural convection, Marangoni convection, vapor recoil pressure, and Argon flow over the melt pool. Several researchers have developed computational models to better understand melt pool dynamics, incorporating these complex phenomena [1-5]. Physics-based simulation such as computational fluid dynamics (CFD) is the key method to model melt pool dynamics (Figure 1). Li et al. [6] utilized a 2D model to examine the melting and