Machine learning-driven prediction of mechanical properties of lightweight concrete based on experimental data

Area of Science:

• Materials Science
• Civil Engineering
• Computational Mechanics

Background:

• Lightweight concrete (LWC) offers reduced self-weight for structural applications.
• Predicting mechanical behavior of expanded polystyrene (EPS)-based LWC is complex due to material interactions.
• EPS incorporation as aggregate replacement presents challenges in performance prediction.

Purpose of the Study:

• To develop an integrated experimental and machine learning framework for predicting EPS-based LWC mechanical properties.
• To establish a reliable Artificial Neural Network (ANN) model for mix design optimization.
• To reduce experimental effort in characterizing EPS-based LWC.

Main Methods:

• Experimental investigation of EPS replacement levels (0-500 L/m³) on LWC properties.
• Data augmentation using physically constrained trends to enhance model robustness.
• Development and validation of an Artificial Neural Network (ANN) model for property prediction.

Main Results:

• Density reduced from 2380 kg/m³ to 1720 kg/m³ with increased EPS content.
• 28-day compressive strength decreased from 41.2 MPa to 28.6 MPa.
• ANN model achieved R² ≈ 0.998 for compressive and splitting tensile strength predictions.

Conclusions:

• The ANN model accurately predicts mechanical properties of EPS-based LWC.
• The framework provides an efficient tool for optimizing LWC mix designs.
• The study demonstrates the potential of machine learning in materials science for concrete technology.