Machine learning for the prediction of duplex stainless steel mechanical properties : Hardness evolution under low temperature aging

Sammanfattning: Duplex stainless steels, DSS are stainless steels that consist of the two phases austenite and ferrite.  The DSS have superb properties and are widely used in industries such as nuclear power and in pressure vessels, pipes and in pipelines.  The use of DSS are limited due to embrittlement which occurs at temperatures from 250 to 550 oC. This imposes a general limited service temperature of 250 oC. The mechanism mainly responsible for the embrittlement is a phase separation occurring in the ferrite phase. Furthermore, there is a direct link between the phase separation and the mechanical properties:  the ferrite hardness increases whereas the toughness decreases under low temperature aging.  In this thesis, the low-temperature embrittlement of duplex stainless  steels  was  studied  through  machine learning  modelling  and  experimental hardness-  and  microscopy measurements.   The  resulting  model  describes  the  data with an accuracy, R-squared = 0.94.  In combination with the experimental results, nickel  was identified  as  an  important  parameter  for  the  hardness  evolution.   This work aims to provide a fundamental study for understanding the importance of alloying elements on the phase separation in DSS, and provides a new methodology via a combination of machine learning and key experiments for the material design.