Loading capacity prediction and optimization of cold-formed steel built-up section columns based on machine learning methods

This study aims to improve the loading capacities of cold-formed steel (CFS) built-up section columns by optimizing cross-sectional geometric dimensions with constant material consumption. An optimization framework was proposed, in which well-trained machine learning (ML) models were employed to predict the compressive strengths of columns, and a genetic algorithm was utilized to generate solutions and search for optimal results. To enhance the training efficiency of ML models, a two-step model construction scheme was adopted with empirical variables and radii of gyration involved. A dataset of 246 test and finite element results covering two cross-section prototypes, namely one closed (CW) section and one open (OI) section, was collected from the literature. Two powerful ML algorithms including neural networks and the Extreme Gradient Boosting (XGBoost) were introduced to predict column loading capacities. Forty optimized results incorporating five groups of benchmark built-up sections and four member lengths for each cross-section prototype were obtained. To examine the applicability of the proposed optimization framework, the forty optimized results were verified by the validated finite element models. The analysis results showed that the constructed two-step ML models in this study performed well in predicting the loading capacities of the CFS built-up section columns. More precisely, the values of coefficient of determination (R-2) for strength predictions of the ML models based on the testing set are all not less than 0.966. With identical material consumption, the proposed optimization framework considerably improved the compressive strengths of the CFS built-up section columns. The strength enhancements of the optimized CW-and OI-sections under compression were up to 21% and 91%, respectively, which manifested that the proposed optimization framework could be applied in the optimum design of the CFS built-up section columns.

Automated summary

This study aims to improve the loading capacities of cold-formed steel built-up section columns by optimizing cross-sectional geometric dimensions with constant material consumption. A machine learning model and a genetic algorithm were used to predict compressive strengths and search for optimal results. The proposed optimization framework significantly improved the compressive strengths of the columns and can be applied in the optimum design process.