Predicting the Load-Bearing Capacity of Square-Section Pipe-Concrete Columns Using Machine Learning Methods

Introduction. In this paper, we consider the problem of predicting the strength of square-section centrally compressed short concrete-filled tubular columns using machine learning methods. Traditional methods, such as the finite element method and the theoretical-experimental approach involving selection of empirical formulas require significant computational resources and time. At the same time, these methods are not always capable of accounting for complex nonlinear dependencies between the parameters. The key objective is to develop a high-precision model capable of predicting the load-bearing capacity of columns using the major parameters. Materials and Methods. For the current study, a database was generated containing the results of numerical experiments on calculating the load-bearing capacity of square-section concrete-filled tubular columns in a physically nonlinear formulation. As part of the study, models based on machine learning methods were designed and implemented using the Jupyter Notebook interactive computing platform. The main method is the CatBoost mechanism (Gradient Boosting Regressor). The resulting models were trained by means of nonlinear optimization methods. Results. The article evaluates the degree of impact of each of the input parameters on the final predictions of the model. The results on the degree of impact for the CatBoost and Random Forrest Regressor (RFR) models are obtained. The quality of the resulting models evaluated using the R 2 value was 98% for CatBoost and 94% for RFR. Discussion and Conclusions. The resulting approach has proved to be highly efficient in predicting the load-bearing capacity of concrete-filled tubular columns, providing a balance between the accuracy of the results and computational complexity.