Axial Compressive Capacity of Circular Steel Tube Recycled Concrete Short Columns via Ensemble Learning

Literature Overview and Research Context

This study addresses the axial compressive bearing capacity of short columns composed of circular steel tubes filled with recycled aggregate concrete (RAC), employing ensemble learning methods to develop predictive models. As a steel pipe engineering professional, I recognize that recycled aggregate concrete filled steel tube (CFST) columns represent an important direction in sustainable structural engineering, particularly where demolition waste needs to be repurposed while maintaining structural integrity. The circular steel tube provides lateral confinement to the concrete core, while the recycled aggregate introduces inherent voids and weaker interfacial transition zones that reduce the concrete's compressive strength compared to natural aggregate concrete. The ensemble learning approach combines multiple weak learners—such as random forests, gradient boosting machines, and support vector machines—into a robust predictive model that can capture the complex nonlinear relationships between input parameters and structural response.

Core Technical Parameters and Material Behavior

The key technical parameters governing the axial compressive capacity of circular steel tube recycled concrete short columns include the steel tube diameter-to-thickness ratio (D/t), the concrete core compressive strength, the recycled aggregate replacement ratio, the column slenderness ratio, and the steel yield strength. The following table summarizes typical parameter ranges and their influence on structural performance.

The confinement mechanism in circular steel tube columns operates through the hoop stress developed in the steel tube wall as the concrete core dilates under axial loading. For recycled aggregate concrete, the confinement effect is somewhat diminished because the porous recycled aggregate particles absorb part of the volumetric strain, reducing the lateral expansion demand on the steel tube. However, the steel tube still prevents spalling and provides a ductile failure mode. The ultimate axial load can be expressed through modified design equations that account for the confinement enhancement factor η, which depends on the concrete strength, the D/t ratio, and the steel yield strength.

Ensemble Learning Approach and Model Performance

The ensemble learning methodology employed in this research combines multiple base classifiers to produce a final prediction with improved accuracy and generalization. Typical ensemble methods used include AdaBoost, Bagging, Stacking, and Voting classifiers. The input feature set usually includes geometric parameters (D, t, L), material properties (f_c, f_y, replacement ratio), and derived indices (D/t, D²t, confinement index). The model is trained on experimental data from published literature and validated against holdout datasets.

The performance metrics typically reported include the coefficient of determination (R²), mean absolute error (MAE), root mean square error (RMSE), and mean absolute percentage error (MAPE). A well-calibrated ensemble model for CFST columns can achieve R² values above 0.90 and MAPE below 5%, significantly outperforming single-model approaches. The advantage of ensemble learning lies in its ability to reduce overfitting and capture interactions between parameters that simple regression cannot address. From a practical engineering standpoint, such models serve as rapid screening tools during preliminary design, though they must be validated against code-based design equations before use in final structural calculations.

Engineering Practice Implications

In practical steel tube fabrication and structural engineering applications, the use of recycled aggregate concrete in CFST columns presents both opportunities and challenges. The steel tube itself must be manufactured to appropriate standards such as GB/T 8163 for structural purposes or EN 10216-1 for hot-finished welded tubes. The welding of tube segments, if required, must follow qualified welding procedures with full penetration butt welds and appropriate heat-affected zone control. The recycled concrete must be properly mixed and pumped to avoid segregation and void formation within the confined core. Quality control during fabrication should include dimensional inspection of the steel tube, hydrostatic testing, and verification of concrete fill density through ultrasonic testing.

A critical observation from this research is that the ensemble learning model can identify optimal parameter combinations that maximize axial capacity while minimizing material usage. For instance, the model may reveal that a 50% recycled aggregate replacement ratio combined with a D/t ratio of 30–40 provides the best balance between sustainability and structural performance. Engineers should use such insights during material selection and cross-sectional design, while always applying appropriate safety factors in accordance with the relevant design codes such as GB 50017 or Eurocode 4.

Summary and Reflection

This research demonstrates that ensemble learning methods provide a powerful tool for predicting the axial compressive capacity of circular steel tube recycled concrete short columns. The combination of material science understanding of recycled aggregate behavior with data analysis modeling creates a practical framework for sustainable structural design. Engineers working with steel tube structures should view such predictive models as supplementary tools that complement, rather than replace, code-based design procedures and experimental validation. The key takeaway is that recycled aggregate concrete can be effectively used in CFST columns when the confinement effect of the circular steel tube is properly leveraged and the design parameters are optimized through integrated computational and experimental approaches.