Enhanced Gaussian Process Model for Predicting Compressive Strength of Ultra-High-Performance Concrete (UHPC).

Area of Science:

• Materials Science
• Civil Engineering
• Computational Mechanics

Background:

• Ultra-high-performance concrete (UHPC) possesses superior mechanical properties, especially compressive strength.
• Accurate compressive strength prediction is vital for UHPC mix design optimization but is hindered by data variability and complexity.
• Existing prediction models often struggle with limited data and intricate material interactions.

Purpose of the Study:

• To enhance the Gaussian Process (GP) model for more accurate UHPC compressive strength prediction.
• To address challenges of data dispersion, limited availability, and complex material interactions in UHPC.
• To develop a robust tool for optimizing UHPC mix proportions and reducing experimental costs.

Main Methods:

• Incorporation of Singular Value Decomposition (SVD) for data quality enhancement and feature extraction.
• Integration of Kalman Filtering and Smoothing (KF/KS) to reduce data dispersion and align predictions with physical principles.
• Development of an enhanced GP model combining SVD and KF/KS for compressive strength prediction.

Main Results:

• The enhanced GP model demonstrated superior accuracy in predicting UHPC compressive strength compared to traditional methods.
• Singular Value Decomposition (SVD) effectively improved data quality by extracting critical features.
• Kalman Filtering and Smoothing (KF/KS) successfully reduced data dispersion and improved prediction reliability.

Conclusions:

• The enhanced GP model provides a reliable and interpretable tool for UHPC compressive strength prediction.
• This approach significantly outperforms Artificial Neural Networks (ANN) and regression models in accuracy and robustness.
• The developed method offers a valuable solution for optimizing UHPC mix designs, lowering experimental expenses, and ensuring structural integrity.