Linking Stress-Rupture Properties to Processing Parameters of HAYNES^® 718 Nickel Superalloy Using Machine Learning

Requirements of stress-rupture life and elongation of nickel alloy 718 are often prescribed by specification AMS5596™ or AMS5662™, which broadly state that the stress-rupture life and elongation must exceed 23 h and 4% at 649 ºC (1200 ºF), respectively. Variability in product stress-rupture life can range from less than 2 h to more than 1000 h depending on test load, which can cause significant delays for testing, shipping, and delivery of a product. In this work, we predict the stress-rupture life and elongation of HAYNES® 718 sheet product utilizing machine learning models. The models were trained on data from 448 lots of material and inputs including composition, room temperature mechanical property data, processing data such as finish gauge, total reduction, final reduction, rule of mixture averaged properties, and environmental factors. Different sets of input features were chosen from the highest absolute Pearson correlation values, one-way ANOVA analysis, random forest (RF) model analysis methods, and generated compound features, and two separate RF models were trained using an 80–20% split between training and testing data. The resulting mean squared errors of best performing models of stress-rupture life and elongations were 102 h and 7.2%, respectively. Input features of the highest importance were observed to be room temperature tensile properties, finish gauge, and tramp elements such as Co, P, and Si. These models can be utilized to accelerate acceptance testing of 718 products by selecting the highest testing load that will still guarantee passing test life and elongation results.