Neural network implementation for the prediction of secondary phase precipitation and mechanical feature in a duplex stainless steel

Duplex stainless steels are extremely valuable materials in the manufacturing environment, featuring remarkable mechanical and physical characteristics. Anyway, the exploitation of this material often requires the creation of welded joints; this is a critical process for the duplex steel, entailing the precipitation of secondary phases. These precipitates undermine the peculiar features of the duplex steels and particularly toughness and corrosion resistance. For the design of welding processes or thermal cycles in general, literature presents several models aimed at the prediction of the sigma-phase precipitation furtherly to the precipitation diagram. In this paper, the presence of secondary phases within a duplex stainless steel 2205 microstructure thermally treated was evaluated with several techniques. At a later stage, an indentation test with a flat-ended cylinder was carried out, obtaining load-indentation depth curves that allow the evaluation of the yield stress. The data acquired during the experimental activities, which highlighted a correlation between secondary phases amount and yield stress, were used for the training of two artificial neural networks aimed at secondary phase amount and indentation curve prediction. The networks implemented are connected in series. The first network predicts the secondary phases' amount with an error of the magnitude of 1% and can be used as starting point for the second network, while the accuracy in the indentation curve prediction allows a precise evaluation of the yield stress.