Predicting Ultra-High-Performance Concrete Compressive Strength Using Tabular Generative Adversarial Networks.

Area of Science:

• Materials Science
• Civil Engineering
• Computational Science

Background:

• Ultra-high-performance concrete (UHPC) research is extensive.
• Traditional methods struggle with UHPC's complex, nonlinear properties and mixture composition relationships.
• Advanced predictive tools are needed for UHPC materials science insights.

Purpose of the Study:

• To develop accurate machine learning models for predicting UHPC compressive strength.
• To utilize a comprehensive experimental database and advanced data generation techniques.
• To gain insights into UHPC strength development and influential parameters.

Main Methods:

• Employed state-of-the-art machine learning techniques.
• Utilized a database of 810 experimental UHPC observations and 15 input features.
• Generated 6513 synthetic data points using tabular generative adversarial networks for model training.
• Trained and validated models including random forest, extra trees, and gradient boosting regression.

Main Results:

• Developed machine learning models demonstrated outstanding predictive performance for UHPC compressive strength.
• Models successfully generalized predictions on unseen experimental data.
• Parametric studies provided valuable insights into UHPC strength mechanisms and parameter significance.

Conclusions:

• Machine learning offers a robust approach for predicting UHPC properties.
• Synthetic data generation enhances the training of predictive models.
• This study advances the understanding of UHPC nonlinear materials science and engineering properties.