Fractal Characterization of Channel-Lining Concrete Based on U-Net Deep Learning under Freeze–Thaw Cycles

Abstract

This study aims to examine the degradation patterns of channel-lining concrete in freeze–thaw cycles in north China. Experimental investigations were conducted to analyze the mechanical properties, surface morphology, and microscopic porosity of the concrete. The U-Net model was employed to perform semantic segmentation on scanning electron microscopy (SEM) images which enhance the accuracy and efficiency of the identification of minute pores. First, its apparent and microscopic pore perimeter area was calculated using the Image Pro Plus image analysis software. The apparent and microscopic fractal dimensions were then computed using fractal theory principles. The experimental results indicated a degradation in the mechanical properties and an increase in the fractal dimension as a consequence of freeze–thaw cycles. The integration of concrete’s internal damage evolution characteristics with the apparent fractal dimension allows for a quantitative analysis of concrete’s damage characteristics when subjected to freezing and thawing. The study also enables the prediction of microscopic damage, facilitating the timely assessment of overall damage to the concrete lining of the channel system. Consequently, it will promptly develop an emergency maintenance plan for the concrete of the channel system.