The degradation of the concrete pore structure is a major concern among scholars due to its negative impact on durability. This study conducted a salt freezing test on polyacrylonitrile fiber concrete (PANFRC) using a composite freeze-thaw medium consisting of Na2SO4, MgSO4, and NaCl (5 % + 5% + 3.5 %). The pore volume fractal dimension was determined by analyzing the T2 spectrum obtained through nuclear magnetic resonance technology. The study explored the influence of different types of pores on frost resistance and developed an artificial neural network-based machine learning model to predict macroscopic frost resistance using the pore volume fractal dimension. The experimental findings indicate that plain concrete exhibits a density loss of 4.67 % upon failure, whereas the density loss in PANFRC is reduced to less than 3 %. The frost resistance durability index is observed to improve as the fiber mass fraction increases within specific ranges. Notably, when the fiber content surpasses 1.2 kg/m³, the durability factor (DF) can be effectively doubled. Drawing from the concept of freeze-thaw cycles, a novel frost-thaw erosion damage model was developed, which incorporates the influence of fiber content under the combined action of freeze-thaw cycles and erosion. The analysis of pore dimensions reveals that the fractal characteristics become more pronounced with the increase in pore size, although this does not extend to the micro-pore range (i.e., those smaller than 0.1 μm). In comparison to other machine learning algorithms, the ANN demonstrates its superiority by establishing a meso-macro transfer prediction model with enhanced generalization capabilities.
Read full abstract