A hybrid methodology integrating non-destructive testing data and neural network models for estimating the compressive strength of concrete in real structures

Assessing the compressive strength of existing concrete structures is challenging, particularly under field conditions. Non-destructive testing (NDT) methods are recommended to support this estimation; however, equipment is usually based on correlation curves derived from standardized specimens, which may not reflect local materials and practices. Therefore, it is essential to develop methodologies suitable for real structures. This paper proposes a hybrid methodology that combines non-destructive tests, calibrated and validated by destructive tests, with artificial intelligence techniques. Four concrete structures with different design strengths were built and subjected to rebound hammer, pull-off, and pin penetration tests at different ages. Based on these data, mathematical and computational models were developed and evaluated. Results showed that artificial neural networks (ANNs) outperformed traditional statistical methods, such as polynomial regression, by reducing prediction errors. The proposed methodology proved feasible for estimating compressive strength in the field, achieving error levels comparable to controlled laboratory conditions.