Predicting compressive strength of hollow concrete prisms using machine learning techniques and explainable...

Area of Science:

• Civil Engineering
• Materials Science
• Computational Mechanics

Background:

• Accurate compressive strength (CS) estimation is crucial for masonry structure design.
• Traditional destructive testing for CS is time-consuming and resource-intensive.
• Developing non-destructive predictive models for CS is highly desirable.

Purpose of the Study:

• To develop and evaluate machine learning (ML) models for predicting the CS of hollow concrete masonry prisms.
• To compare the performance of various ML algorithms, including Multi Expression Programming (MEP), Random Forest Regression (RFR), and Extreme Gradient Boosting (XGB).
• To identify the most significant factors influencing CS through explainable AI (XAI) analysis.

Main Methods:

• A dataset of 159 experimental results was compiled from published literature.
• Input parameters included masonry unit strength, height-to-thickness ratio, mortar strength, and a ratio of material properties.
• ML algorithms (MEP, RFR, XGB, AdaBoost) were trained and validated; accuracy was assessed using RMSE, OF, and R².

Main Results:

• Extreme Gradient Boosting (XGB) demonstrated the highest accuracy with an R² of 0.99 and the lowest Objective Function (OF) value of 0.0063.
• Multi Expression Programming (MEP) and Genetic Programming (GEP) provided empirical equations for CS prediction.
• XAI analysis revealed that the strength of masonry units is the most influential parameter for predicting CS.

Conclusions:

• ML-based models, particularly XGB, offer a highly accurate and efficient method for predicting the CS of hollow concrete masonry prisms.
• These models can significantly reduce the need for expensive and time-consuming laboratory testing.
• The findings provide a practical tool for engineers and researchers in masonry design and analysis.