Data-driven prediction on critical mechanical properties of engineered cementitious composites based on machine...

Area of Science:

• Civil Engineering
• Materials Science
• Computational Mechanics

Background:

• Engineered Cementitious Composites (ECCs) require precise mechanical property prediction for optimal design.
• Existing methods for predicting ECC properties can be time-consuming and resource-intensive.
• A data-driven approach using machine learning offers a novel solution for efficient property estimation.

Purpose of the Study:

• To develop and optimize machine learning models for predicting key mechanical properties of ECCs.
• To identify the most influential features affecting ECC mechanical properties.
• To establish a control strategy for ECC mechanical properties based on predictive modeling.

Main Methods:

• Compilation and analysis of 1532 ECC datasets.
• Implementation and comparison of four machine learning algorithms: Linear Regression (LR), K-Nearest Neighbors (KNN), Random Forest (RF), and Extreme Gradient Boosting (XGB).
• Application of SHapley Additive exPlanations (SHAP) for feature importance analysis.

Main Results:

• RF model achieved R² of 0.92 for compressive strength and 0.91 for flexural strength.
• XGB model achieved R² of 0.87 for tensile strength and 0.80 for tensile strain capacity.
• Water-cement ratio (W) and water reducer (WR) significantly impact compressive and tensile strength; WR is key for compressive strength, and polyethylene (PE) fiber for other properties.

Conclusions:

• Machine learning models, particularly RF and XGB, provide accurate predictions of ECC mechanical properties.
• Feature importance analysis reveals WR and PE fiber as critical factors influencing ECC performance.
• The developed control strategy offers practical guidance for engineers in designing ECC with desired mechanical characteristics.