Multivariate Machine Learning Framework for Predicting Electrical Resistivity of Concrete Using Degree of Saturation...

Area of Science:

• Civil Engineering
• Materials Science
• Electrical Engineering

Background:

• Apparent electrical resistivity (ER) is a key indicator of concrete's moisture state and pore structure.
• Understanding the relationship between ER and concrete properties is crucial for durability assessment.

Purpose of the Study:

• To investigate the influence of moisture and pore structure on concrete's apparent electrical resistivity.
• To develop a predictive model for ER using material parameters and machine learning.

Main Methods:

• Continuous ER measurement during controlled wetting/drying cycles for six concrete mixes.
• Evaluation of machine learning models (Gaussian Process Regression, Neural Networks) with parameters like degree of saturation (DS), porosity (P), water-cement ratio (WCR), and compressive strength (f'c).
• SHAP analysis to determine parameter influence and formulation of nonlinear multivariate regression models.

Main Results:

• ER showed an exponential decrease with increasing DS (R² = 0.896–0.997), confirming DS as the dominant factor.
• Machine learning models, especially Gaussian Process Regression and Neural Networks, accurately predicted ER when all parameters were included.
• SHAP analysis highlighted DS as the primary influencer, with P and WCR offering secondary contributions.

Conclusions:

• Apparent electrical resistivity is strongly linked to moisture dynamics and concrete pore structure.
• An integrated experimental-computational approach provides a physically interpretable and robust framework for non-destructive concrete durability assessment.