Machine-Learning-Based Probabilistic Model and Design-Oriented Formula of Shear Strength Capacity of UHPC Beams.

Area of Science:

• Civil Engineering
• Materials Science
• Structural Engineering

Background:

• Designing Ultra-High Performance Concrete (UHPC) beams for shear is complex due to the interplay of multiple design factors.
• Existing methods may not fully capture the synergistic effects influencing shear capacity.

Purpose of the Study:

• To develop accurate predictive models for UHPC beam shear capacity.
• To quantify the uncertainty associated with shear capacity predictions.
• To provide practical design tools for engineers.

Main Methods:

• Compilation of a large, curated database of laboratory shear tests on UHPC beams.
• Development and validation of machine learning models to predict shear capacity.
• Formulation of a simplified, design-oriented equation.

Main Results:

• Machine learning models achieved accurate point predictions for shear capacity.
• The best models provided a 95% prediction band, with approximately 95% of test results falling within this range.
• A user-friendly design formula was developed for spreadsheet implementation.

Conclusions:

• The developed machine learning models and design formula enhance the prediction of UHPC beam shear capacity.
• These tools offer engineers improved capabilities for screening options, checking designs with uncertainty margins, and selecting conservative values.
• The transparent and implementable approach supports preliminary sizing, verification, and assessment of UHPC members in alignment with common code variables.