Strength Prediction of Cement-Stabilized Steel Slag Using Deep Learning and SHAP Analysis

Area of Science:

• Materials Science
• Civil Engineering
• Geotechnical Engineering

Background:

• Steel slag utilization in construction materials is crucial for sustainability.
• Understanding the mechanical properties and microstructural evolution of cement-stabilized steel slag (CSSS) is essential for its effective application.
• Predictive models for CSSS strength are needed to optimize mix designs and ensure structural integrity.

Purpose of the Study:

• To investigate the influence of curing age, steel slag content, and gradation on CSSS mechanical properties.
• To elucidate the microstructural mechanisms governing CSSS strength development.
• To develop and validate a deep learning model for predicting CSSS unconfined compressive strength (UCS) and splitting tensile strength (STS).

Main Methods:

• Experimental analysis including SEM-EDS and XRD for microstructural characterization.
• Development and application of a CNN-GRU-Attention deep learning model for strength prediction.
• Statistical analyses (SHAP, Pearson correlation) to identify key strength determinants.

Main Results:

• CSSS strength increases nonlinearly with curing age, peaking at 60% steel slag content.
• Steel slag incorporation enhances AFt formation and gel network densification, with carbonation and secondary hydration contributing to long-term strength.
• The CNN-GRU-Attention model demonstrated high accuracy (R² > 0.94) and robustness in predicting UCS and STS, outperforming benchmark models.
• Cement content was identified as the primary strength driver, with steel slag content showing a threshold effect.

Conclusions:

• The study provides a comprehensive understanding of CSSS mechanical behavior and microstructural development.
• The developed deep learning model offers a reliable tool for predicting CSSS strength, aiding in practical engineering applications.
• Findings emphasize the importance of controlled gradation and judicious steel slag content for optimizing CSSS performance.