Knowledge-based bidirectional thermal variable modelling for directed energy deposition additive manufacturing

Abstract

Directed energy deposition additive manufacturing (DED-AM) has gained significant interest in producing large-scale metallic structural components. In this paper, a knowledge-based machine learning (ML) approach, combining both physics-based simulation and data-driven modelling, is proposed for a study on thermal variables of DED-AM. This approach enables both forward and backward predictions, which breaks down the barriers between the basic process parameters and key process attributes. Process knowledge plays a critical role to enable the prediction and enhance the accuracy in both prediction directions. The proposed ML approach successfully predicted the thermal variables of wire arc based DED-AM for forward modelling and the process parameters for backward modelling, typically within 7% errors. This approach can be further generalised as a powerful modelling tool for design, control, and evaluation of DED-AM processes regarding build geometry and properties, as well as an essential constituent element in a digital twin of a DED-AM system.

Keywords

• Directed energy deposition additive manufacturing (DED-AM)
• digital manufacturing
• knowledge-based data-driven modelling
• machine learning (ML)
• finite element analysis (FEA)