ANN Prediction-Assisted Control of Compensating Shrinkage in Concrete-Filled Steel Tubes and Mechanism Analysis

Overview of the Study

This research employs artificial neural network (ANN) prediction models to optimize the compensating shrinkage control of concrete within steel tubes, aiming to minimize shrinkage-induced stresses and improve the long-term structural integrity of CFST columns. Concrete shrinkage within confined steel tubes creates complex stress interactions between the shrinking concrete and the restraining steel tube wall, potentially leading to internal cracking, reduced bond strength, and premature corrosion initiation. The ANN approach enables the identification of optimal mix design parameters that achieve compensating shrinkage, where the shrinkage of the concrete is offset by the expansion of supplementary cementitious materials or shrinkage-compensating admixtures.

Core Technical Content

The ANN model is trained on experimental data from CFST specimens with various concrete mix designs, steel tube geometries, and curing conditions. The input parameters typically include:

The output parameter is the shrinkage strain (με), with the ANN model achieving prediction accuracy of R² > 0.90 and average absolute percentage error (AAPE) < 8%.

Key findings include:

• The optimal shrinkage-compensating admixture dosage is approximately 3% to 5% of cement weight for achieving near-zero shrinkage within steel tubes.
• The steel tube confinement effect reduces free shrinkage by 40% to 60% compared to unconfined concrete specimens.
• The combination of silica fume (10% to 15%) and moderate shrinkage-compensating admixture (3% to 5%) achieves the best balance between shrinkage control and strength development.
• The shrinkage-induced hoop stress in the steel tube wall can reach 30MPa to 50MPa, which is significant relative to the yield strength of Q345B steel.

Mechanism Analysis

The shrinkage behavior of concrete within steel tubes involves several mechanisms:

1. Chemical shrinkage: The hydration of cement produces a volume reduction that is partially compensated by the expansion of calcium hydroxide and other hydration products. In UHSC or high-strength concrete with low water-cement ratios, chemical shrinkage is more pronounced.
2. Drying shrinkage: The loss of water from the concrete pore structure causes capillary tension and subsequent shrinkage. Within steel tubes, the drying shrinkage is partially restrained by the steel tube wall, creating internal tensile stresses in the concrete.
3. Thermal shrinkage: The temperature decrease during curing causes thermal contraction of the concrete, which is restrained by the steel tube, generating compressive stresses in the steel tube wall.
4. Creep shrinkage: The sustained shrinkage stress causes time-dependent deformation (creep) of the concrete, which gradually reduces the shrinkage-induced stresses over time.

The ANN model captures the nonlinear interactions between these mechanisms and the mix design parameters, providing a predictive tool that can identify optimal combinations for compensating shrinkage.

Standards and Quality Control

The shrinkage test for concrete within steel tubes follows a modified procedure compared to standard shrinkage prisms:

1. Cast concrete into steel tubes with end plates sealed to prevent moisture loss from the ends.
2. Cure for 7 days under standard conditions (20°C, 95% humidity).
3. Install displacement transducers at multiple locations along the tube length.
4. Monitor shrinkage strain at 1, 3, 7, 14, 28, 56, and 90 days.
5. Compare with free-shrinkage reference specimens of identical mix design.

Engineering Practice Integration

The practical application of ANN-predicted shrinkage control involves several steps:

1. Mix design optimization: Use the ANN model to predict shrinkage for various mix designs and identify combinations that achieve compensating shrinkage (net shrinkage strain < 100με at 90 days).
2. Steel tube design adjustment: Account for shrinkage-induced hoop stresses in the steel tube design. For Q345B steel with yield strength of 345MPa, the shrinkage stress should not exceed 30% of yield strength (approximately 100MPa) to ensure adequate safety margin.
3. Construction sequencing: Implement staged concrete placement for long columns to reduce thermal and shrinkage stresses. Each stage should be allowed to cure for at least 7 days before the next stage is placed.
4. Post-placement monitoring: Use strain gauges or fiber optic sensors embedded in the concrete to monitor shrinkage development and validate the ANN predictions.
5. Quality assurance: Perform ultrasonic testing at 28 days to verify the soundness of the concrete fill and detect any shrinkage-induced cracking at the steel-concrete interface.

Key Questions and Reflections

A critical question is the generalizability of the ANN model across different concrete mix designs, steel tube geometries, and environmental conditions. While the model achieves high accuracy for the training dataset, its extrapolation capability to untested parameter combinations must be validated through additional experimental work. The ANN approach should be viewed as a decision-support tool rather than a replacement for experimental verification.

Another important consideration is the long-term durability implications of shrinkage control. While compensating shrinkage reduces internal stresses and cracking, the use of shrinkage-compensating admixtures may affect the long-term chemical stability of the concrete. Calcium hydroxide-based admixtures, for example, may increase the alkalinity of the pore solution, which is beneficial for steel passivity but may affect the compatibility with certain supplementary cementitious materials.

Study Insights and Implications

This study demonstrates that ANN-based prediction is a powerful tool for optimizing shrinkage control in CFST columns, enabling the identification of mix designs that achieve near-zero shrinkage within confined steel tubes. The key implication for steel pipe manufacturers and structural engineers is that shrinkage-induced stresses are a significant but often overlooked design factor that can compromise the long-term integrity of CFST columns. The ANN approach provides a systematic methodology for addressing this issue through mix design optimization rather than relying on empirical rules of thumb. Future research should focus on extending the ANN model to predict the combined effects of shrinkage, creep, and thermal cycling on the long-term behavior of CFST columns under sustained and cyclic loading. The integration of ANN predictions with finite element analysis could provide a comprehensive design framework that accounts for all major degradation mechanisms in CFST structures.