Optimised Machine Learning and Statistical Modelling for Predicting the Design Strengths of Hardened 3D Printed...

Area of Science:

• Construction Engineering
• Materials Science
• Computational Mechanics

Background:

• 3D Concrete Printing (3DCP) offers significant advantages in construction, including design flexibility and efficiency.
• A major barrier to 3DCP adoption is the lack of standardized design provisions for structural elements.
• Predicting the mechanical properties of hardened 3D-printed concrete is crucial for developing these provisions.

Purpose of the Study:

• To develop statistical and machine learning models for predicting the design strengths of hardened 3D-printed concrete.
• To analyze experimental data to derive design strength equations for 3DCP.
• To benchmark the performance of statistical models against machine learning algorithms.

Main Methods:

• Compiled and analyzed a comprehensive database of experimental results for 3DCP materials (2012-2025).
• Employed regression-based statistical modeling to derive design strength equations.
• Developed and benchmarked machine learning algorithms using the same dataset.

Main Results:

• Statistical models showed good predictive accuracy for compressive (R²=0.83) and flexural (R²=0.67) strengths.
• The shear strength model exhibited higher variability (R²=0.39) due to limited data.
• Machine learning models confirmed the robustness of statistical equations, achieving high R² values for compressive (0.93) and flexural (0.94) strengths.

Conclusions:

• The developed models provide preliminary design strength predictions for hardened 3D-printed concrete.
• This framework contributes to establishing robust structural design guidelines for 3DCP.
• Facilitates wider adoption of 3D Concrete Printing in the construction industry.