Ensemble Learning for Predicting the Tensile Strength of Alloy Steels from Chemical Composition and Processing Parameters

The production of steels with certain mechanical properties is essential for industry and materials science. These properties depend on a number of indicators of the production process - chemical composition, heat treating, and other processing parameters. This paper applies the Arcing ensemble algorithm to study and predict the tensile strength (yield) of 591 types of AISI alloy steels. The aim is to determine the dependence of tensile strength on the chemical composition, two main processing temperatures, the type of treatment and the diameter of the specimen. Modeling was carried out in two stages, dividing the studied data into three groups for training, validation and testing in the ratio of 80:10:10 and 40:10:50. The built models predict the yield strength test data from the first stage with R2=92.4
^%
and RMSE=71.9 mPa, and for the second stage with R2=86.1% and RMSE=87.9 mPa. The achieved difference in R2 of 6.3% and in prediction error of 22.2% suggests that the implementation of the arcing ensemble algorithm does not depend significantly on the testing scheme.