A New Prediction Method of the Constitutive Stress–Strain Relationship for Stainless Steel–Clad Bimetallic Steel Bars

Abstract Stainless steel–clad bimetallic steel bars (SCBSBs) as composite structures exhibit complex stress–strain behavior due to dimensional and material heterogeneity, limiting engineering applications. This paper describes a novel approach to effectively predict the stress–strain behavior of SCBSB using machine learning (ML) techniques. A complete data set of 54 tensile test specimens, including 304/316SCBSB, HRB400E steel rebar, 304L, and 316L steel bars, was used to train and test ML models. Six advanced ML models were used to predict the mechanical properties of various steel bars. The GBoost model consistently outperformed other models, predicting stress–strain curve parameters with outstanding accuracy and a high coefficient of determination. Furthermore, the projected stress–strain curves created by GBoost were compared to experimental data, demonstrating its superiority in stress–strain curve points while also emphasizing accuracy and stability. By using the second strain-hardening exponent, the model effectively characterized the nonlinear features of different steel types. Moreover, an importance analysis of the geometric dimensions was carried out, and it revealed that the effect of the cladding ratio on the mechanical parameters becomes more obvious with the increase in diameter.