Prediction of Compressive Strength of Partially Saturated Concrete Using Machine Learning Methods

Area of Science:

• Materials Science
• Civil Engineering
• Concrete Technology

Background:

• Marine environments pose significant challenges to concrete durability.
• Accurate estimation of concrete compressive strength is crucial for structural integrity.
• Existing methods may not fully account for varying saturation and salinity conditions.

Purpose of the Study:

• To recommend criteria for estimating concrete compressive strength in marine environments.
• To evaluate the effectiveness of non-destructive testing (NDT) parameters and design parameters.
• To compare machine learning models for predicting compressive strength.

Main Methods:

• Cylindrical concrete specimens from three design mixtures were tested.
• Specimens were subjected to various saturation levels and saline solutions.
• P-wave velocity, S-wave velocity, electrical resistivity, density, and water-to-binder ratio were measured.
• Artificial neural network (ANN), support vector machine (SVM), and Gaussian process regression (GPR) were employed.

Main Results:

• ANN demonstrated the highest estimation accuracy (R-squared).
• GPR yielded the lowest root-mean-squared error (RMSE).
• A prediction model using P-wave velocity, electrical resistivity, and water-to-binder ratio showed practical viability.

Conclusions:

• A combined approach using NDT and design parameters provides reliable compressive strength estimation.
• The recommended model offers a practical solution for assessing concrete in marine conditions.
• Further research could explore additional parameters and environmental factors.