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

Literature Overview and Research Context

This study investigates the axial compressive bearing capacity of circular steel tube columns infilled with recycled concrete, employing ensemble learning methods to develop predictive models. The use of recycled aggregates in structural concrete is a growing trend driven by sustainability objectives, but the mechanical behavior of recycled concrete differs from that of natural aggregate concrete due to the presence of adhered mortar on recycled aggregate surfaces, higher porosity, and lower stiffness. The ensemble learning approach combines multiple base learners to improve prediction accuracy and generalization capability, addressing the nonlinear and complex relationships between input parameters and structural response.

Core Technical Content and Key Findings

The research likely involves a database of experimental results from circular steel tube recycled concrete (CFST-RA) short columns, with input parameters including steel tube diameter, wall thickness, yield strength, recycled aggregate replacement ratio, recycled concrete compressive strength, and column length. The output variable is the axial compressive bearing capacity. Ensemble learning methods such as Random Forest, Gradient Boosting, or Stacking may be employed to train predictive models and evaluate their performance against analytical solutions and experimental data.

Interpretation of Technical Points

The mechanical behavior of CFST-RA columns under axial compression is governed by the interaction between the steel tube and the infill concrete. Under increasing axial load, the concrete tends to expand laterally, and the steel tube resists this expansion by providing confining pressure. This confinement effect enhances the triaxial compressive strength of the concrete, leading to a higher ultimate load capacity compared to plain concrete cylinders of the same cross-section. The degree of confinement is characterized by the D/t ratio and the steel yield strength.

The introduction of recycled aggregates modifies the concrete's mechanical properties in several ways. Recycled aggregates typically exhibit lower density, higher water absorption, and reduced stiffness compared to natural aggregates. These properties affect the concrete's compressive strength, elastic modulus, and post-peak ductility. The ensemble learning model must capture these nonlinear relationships to provide accurate predictions across the full range of input parameters.

The ensemble learning approach offers several advantages over single-model predictions. By combining multiple diverse learners, the ensemble reduces variance, improves bias-variance tradeoff, and provides more robust predictions for out-of-sample data. Feature importance analysis from tree-based ensemble methods can reveal which input parameters have the greatest influence on bearing capacity, guiding experimental design and engineering optimization.

Integration with Engineering Practice

In practical engineering applications, the predictive models developed through ensemble learning can serve as preliminary design tools, enabling engineers to estimate bearing capacity during the conceptual design phase before detailed analysis and testing. However, these models should be validated against experimental data from specific projects, as the performance of recycled concrete is highly dependent on the source and processing of recycled aggregates.

Quality control considerations include:

• Verification of recycled aggregate properties through laboratory testing (density, water absorption, crushing value).
• Confirmation of concrete mix design through trial batches and cube testing.
• Inspection of steel tube dimensional tolerances and material certificates.
• Post-construction load testing or non-destructive evaluation to confirm in-situ performance.

The PDCA cycle is particularly relevant in the context of recycled concrete production, where batch-to-batch variability is greater than with natural aggregates. Continuous monitoring and feedback of test results to the mix design process ensures consistent quality.

Key Questions and Reflections

A fundamental question in this research domain is the generalizability of ensemble learning models trained on one dataset to other datasets with different material sources, geographic locations, or construction practices. The diversity and representativeness of the training database are critical factors that determine the applicability of the predictive models. Engineers must exercise caution when applying models developed from one context to another without adequate validation.

Another reflection pertains to the physical interpretability of ensemble learning models. While these models often achieve high predictive accuracy, their internal decision-making processes can be opaque, making it difficult to extract engineering insights about the underlying mechanical behavior. The balance between predictive accuracy and physical transparency is an ongoing challenge in the application of data analysis to structural engineering.

Study Insights and Implications

This research demonstrates the potential of ensemble learning methods for predicting the structural performance of recycled concrete-filled steel tube columns, contributing to the advancement of sustainable construction technologies. The integration of data-driven approaches with traditional structural engineering knowledge can accelerate the adoption of recycled materials in load-bearing applications. However, the ultimate reliability of these predictions depends on the quality and representativeness of the underlying experimental data, as well as the careful validation of models against independent test results before implementation in design practice.