Predictive model of the microstructure of the arc-based directed energy deposited AWS ER-2594 stainless steel using artificial neural networks and mathematical modelling

In super duplex stainless steels, microstructural alterations, such as improper ferrite-to-austenite phase ratios and the formation of intermetallic phases, can significantly impact the material properties during service. Therefore, predicting the resulting microstructure (i.e., phase fractions) as a function of input parameters (e.g., wire feed speed, travel speed, shielding gas) is crucial when planning the additive manufacturing of components intended for field applications. In this context, an alternative approach is the development of predictive tools for phase balance using artificial neural networks and mathematical modeling, aiming to map a range of process parameters and their influence on the final microstructure of the printed part. This study employed neural network training (Backpropagation, Patternet, and Neuro-Fuzzy), and mathematical modeling to predict the austenite content in ER 2594 stainless steel produced by additive manufacturing. The effectiveness of the developed model, with R and R² values ranging from 0.97 to 0.99, demonstrates the potential of applying machine learning tools to microstructure prediction, which in turn directly affects key properties such as corrosion resistance and mechanical performance.