Neural network-based prediction of plastic strain-hardening in metal additive manufacturing

Additive manufacturing (AM) process parameters significantly affect the mechanical behavior of printed materials under uniaxial tension. Therefore, establishing the relationship between these parameters and material behavior is essential to enhance AM performance. This study focuses on predicting the strain hardening behavior of AM materials using the different plastic hardening models, including: Hollomon, Ludwick, Voce, Swift, Ramberg-Osgood, and Swift-Voce to describe the plastic zone during tensile testing. Since artificial intelligence excels at capturing complex interactions by applying algorithms like artificial neural networks, we aim to uncover the relation between key process parameters (such as layer thickness, scanning speed, building orientation, printing power, hatch spacing, and laser spot size) and chemical composition/material properties (such as density, thermal conductivity, melting temperature, coefficient of thermal expansion and specific heat) on plastic hardening zone of uniaxial stress-strain curve. Experimental validation and comparative analysis of the developed models underscore their effectiveness and highlight the potential of intelligent algorithms in advancing AM research. Among the models evaluated within the study, the Ludwick model demonstrated the best fit and training.