Prediction of the axial compression capacity of ECC-CES columns using adaptive sampling and machine learning techniques.

An innovative form of concrete-encased steel (CES) composite columns incorporating engineered cementitious composites (ECC) confinement (ECC-CES) has recently been introduced, displaying superior performance in failure behavior, ductility, and toughness compared to traditional CES columns. This study presents an innovative approach to predicting the axial capacity of ECC-CES columns using adaptive sampling and machine learning (ML) techniques. This study initially introduces a finite element (FE) modeling for ECC-CES columns, integrating material and geometric nonlinearities to accurately capture the inelastic behavior of ECC and steel through appropriate constitutive material laws. The FE model was validated against experimental data and demonstrated strong predictive accuracy. An adaptive sampling process is employed for efficient exploration of the design space to generate a database of 840 FE models. Subsequently, seven ML models are utilized to predict the axial compression capacity based on the FE database. These models were comprehensively evaluated, displaying a superior prediction performance compared to design standards such as EC4 and AISC360. From evolution metrics, the Gaussian process regression, CatBoost (CATB), and LightGBM (LGBM) models emerged as the most accurate and reliable model, with nearly more than 97% of FE samples within the 10% error range. Despite the robust performance of the ML models, their black-box nature limits practical applicability in design contexts. To address this, the study proposes a symbolic regression-derived design that offers interpretable, explicit design equations with competitive performance metrics.