Predictive Modelling for Concrete Failure at Anchorages Using Machine Learning Techniques

Area of Science:

• Civil Engineering
• Structural Engineering
• Materials Science

Background:

• Concrete anchorage is critical in construction.
• Concrete cone breakout is a quasi-brittle failure mode in anchors under tensile load, posing risks due to lack of warning.
• Accurate reliability assessment of anchors against concrete cone failure is essential.

Purpose of the Study:

• To develop and evaluate machine learning (ML) predictive models for the tensile breakout capacity of concrete anchors.
• To compare the performance of Gaussian Process Regression (GPR) and Support Vector Regression (SVR) against existing semi-empirical models.
• To assess the suitability of ML models for reliability frameworks using a novel explainability concept.

Main Methods:

• Utilized a dataset of 864 experimental anchor tests.
• Developed predictive models using Gaussian Process Regression (GPR) and Support Vector Regression (SVR) machine learning algorithms.
• Assessed model efficiency via statistical comparison to state-of-practice semi-empirical models and a new Model Explainability concept based on Analogous Rational and Mechanical phenomena (MEARM).

Main Results:

• The developed GPR and SVR models demonstrated high predictive accuracy for tensile breakout capacity.
• ML models showed improved precision and reduced uncertainty compared to the current state-of-practice semi-empirical model.
• The MEARM concept provided insights into the ML models' decision-making processes.

Conclusions:

• Machine learning models, GPR and SVR, offer a more precise and reliable approach to predicting concrete anchor tensile breakout capacity.
• The developed ML models show potential for use as General Probabilistic Models in structural reliability assessments.
• Further research can refine ML applications for enhanced construction safety and design standards.