A Hybrid Methodology for Predicting the Load-Bearing Capacity of Rectangular Concrete-Filled Steel Tube Columns Under Eccentric Compression

Abstract

Objective. To develop a hybrid methodology for predicting the load-bearing capacity of rectangular concrete-filled steel tube (CFST) columns under eccentric compression, based on the integration of analytical calculation models, nonlinear finite-element modeling, and machine learning techniques.

Method. The study is based on a combined computational approach that includes a preliminary analytical assessment using the limit equilibrium method and a refining nonlinear finite-element analysis employing Geniev's theory of concrete plasticity and the Newton–Raphson iterative scheme. A synthetic dataset was generated from the obtained results and used to train a CatBoost machine learning model, with performance evaluated using the Root Mean Squared Logarithmic Error (RMSLE) metric.

Result. An algorithm for the generation and validation of a computational database was developed, ensuring the physical plausibility and representativeness of the training sample. A structured dataset was formed, covering a wide range of geometric and strength parameters of the columns. The resulting machine learning model demonstrates high accuracy in predicting the ultimate load-bearing capacity under eccentric compression.

Conclusion. The developed hybrid methodology enables high-accuracy prediction of the load-bearing capacity of CFST columns under eccentric compression by combining analytical models, finite-element analysis, and machine learning. The obtained results can be used to create digital tools for structural analysis and optimization, as well as to improve existing engineering design practices.