Machine Learning-based Prediction of Concrete Deck Deterioration Using Spatial Features of Electrical Resistivity Data in Full-scale Bridge Testing

Monitoring the condition of concrete bridge decks is essential for maintaining performance, extending service life, and supporting sustainable infrastructure. This study presents a novel machine learning (ML) approach for assessing concrete deck deterioration by analyzing spatial statistical features of electrical resistivity (ER) data. ER data were collected over 850 days through 14 measurement cycles in a full-scale accelerated bridge testing program at Rutgers University, designed to simulate realistic field conditions. Concrete deterioration was evaluated using dynamic modulus values obtained from ultrasonic surface wave measurements. Principal spatial features—average, coefficient of variation, and entropy—extracted from ER data, corrected for non-destructive factors (temperature, age, and salinity), were employed to train various ML models. The support vector machine (SVM) achieved an F1 score of 0.82, surpassing traditional threshold methods by 60%. The results demonstrate the effectiveness of this ML approach in tracking damage progression and enabling early anomaly detection. These findings are significant for network-level assessments of large bridge decks and provide a foundation for more robust deterioration assessment frameworks.