Durability evaluation of concrete structure under freeze-thaw environment based on pore evolution derived from deep learning

Concrete exhibits significant variability, making it challenging to accurately determine many parameters, especially in quantifying damage caused by freeze-thaw cycles. Consequently, current methods for assessing concrete damage in freeze-thaw environments are often subjective and insufficient. This study integrates artificial intelligence with durability diagnostics by employing deep learning image recognition algorithms and X-CT scanning technology to develop an intelligent segmentation model for micropores inside concrete. The analysis focused on the evolution of pore structure parameters inside concrete under freeze-thaw cycles. By introducing fractal theory, the study examines the correlation between the fractal box-counting of internal concrete pores and the macro index under freeze-thaw cycles. A method utilizing the fractal box-counting dimension of internal pores as a damage variable to evaluate concrete's freeze-thaw durability is proposed. Results indicate that the intelligent segmentation model established using the U-Net3 + deep learning algorithm effectively captures and quantifies the complex internal pore information in concrete. This provides a comprehensive and intuitive approach to exploring the evolution of internal pore structures in concrete under complex service environments. As freeze-thaw cycles increase, the fractal box-counting dimension of internal pores in concrete gradually increases, and the pore structure transition from ordered to disordered. The complexity of pore space distribution also increases. A strong linear correlation exists between the fractal box-counting dimension at the micro level and concrete's macro index at the macro level. The new concrete freeze-thaw degradation assessment method proposed in this study can be used to accurately evaluate the freeze-thaw damage status of concrete.