Improved prediction accuracy for compressive strength of recycled aggregate concrete using optimization-based...

Area of Science:

• Materials Science and Engineering
• Civil Engineering
• Computational Intelligence

Background:

• Sustainable construction necessitates accurate prediction of recycled aggregate concrete (RAC) properties.
• Traditional methods for predicting compressive strength (CS) in RAC may lack accuracy and generalization.
• Optimization algorithms can enhance the performance of predictive models for concrete properties.

Purpose of the Study:

• To develop and evaluate a data-driven model for predicting the compressive strength (CS) of recycled aggregate concrete (RAC).
• To compare the performance of an Elite Single Genetic Optimization Algorithm-based Cascade Forward Neural Network (ESGA-CFNN) with other optimization-based neural network models.
• To assess the practical applicability of the developed models for quality control and mix design optimization in RAC production.

Main Methods:

• Development of an ESGA-CFNN model using 272 RAC samples with key parameters: water-to-cement ratio (WCR), water absorption (WA), aggregate densities, and water-to-total material ratio (WTMR).
• Comparative analysis with Particle Swarm Optimization-based CFNN (PSO-CFNN) and Artificial Bee Colony-based CFNN (ABC-CFNN) models.
• Utilized K-fold cross-validation for model development to prevent overfitting and practical validation with 6 RAC samples.

Main Results:

• The ESGA-CFNN model achieved superior performance with a root-mean-squared error (RMSE) of 1.144, R-squared (R²) of 0.991, and an a₁₀-index of 1.000.
• Comparative models showed slightly lower performance: ABC-CFNN (RMSE=1.434, R²=0.987, a₁₀-index=0.982) and PSO-CFNN (RMSE=1.561, R²=0.984, a₁₀-index=0.982).
• Practical validation confirmed the real-world applicability and effectiveness of the proposed models.

Conclusions:

• The ESGA-CFNN model is a highly effective tool for predicting the compressive strength of RAC, crucial for quality control.
• This novel hybrid approach offers enhanced predictive accuracy and generalization capabilities compared to traditional methods.
• The findings support optimizing RAC mix designs for cost reduction, sustainability, and meeting construction standards.