Fatigue life evaluation model for various austenitic stainless steels at elevated temperatures via alloy features-based machine learning approach

An alloy features-based and chemical compositions-based machine learning method was used to examine the low cycle fatigue life of austenitic stainless steels at different elevated temperatures employing one model. Furthermore, eight algorithms were used to examine the impact of algorithms on the precision of constructed models. As input, physicochemical features of elements were transformed from chemical compositions. After being conducted by the feature screening process, electronegativity deviation (E2.sd), ionization energy deviation (E6.sd), testing conditions, and tensile strength were chosen as input. The results show that algorithms affect accuracy and the models with the highest accuracy are SVR and ANN for alloy features and chemical compositions-based method, respectively. Chemical composites-based model demonstrates relatively lower precision than the alloy feature model. Almost all testing data distribute within two-factor band lines predicted by alloying feature-based model. The validation testing results indicate that 83% data plots distribute within two-factor band lines.