Failure mode and capacity prediction for bolted T-stub connections using ensemble learning

Abstract

The present study utilizes machine learning (ML) algorithms, namely decision trees, and ensemble learning methods including bagging, random forests, gradient boosting, and extreme gradient boosting to identify the failure mode and predict the corresponding capacity of preloaded bolted T-stub steel connections subjected to tension. A nonlinear finite element model (FEM) was first developed and validated against experimental results from literature. Subsequently, a matrix of 3447 FEMs, covering a wide range of design parameters, were generated to train and test the ML models. In addition, to evaluate the significance of each input parameter in predicting the connections' capacity, different embedded feature selection techniques were employed. Furthermore, different interpretability methods were utilized to uncover the interrelationships between the input features and the modeled outputs to further interpret the developed Blackbox models. The results highlight the effectiveness of ML techniques in predicting the failure modes and corresponding capacity of bolted T-stub connections, supporting their potential as robust and rapid design tools which ensure adequate designs for new connections or effective assessments for existing ones.