Prediction of freeze-thaw damage of asphalt concrete based on distributed fiber optic sensors and KAN-Transformer fusion model

Asphalt mixtures are widely used in road construction but are vulnerable to damage caused by freeze–thaw cycles. This study introduces a novel approach to real-time monitoring using distributed fiber optic sensing (DFOS) technology and proposes an innovative deep learning fusion network architecture combining Kolmogorov-Arnold Network (KAN) and Transformer models. Real-time monitoring of asphalt beams during freeze–thaw cycles is achieved through DFOS, which collects data on strain, temperature, and other critical physical parameters. The KAN-Transformer model, along with Transformer-based time series models, is employed for data processing and feature extraction to detect and predict minor changes in material properties during freeze–thaw cycles. The results demonstrate that the KAN-Transformer model outperforms the traditional Transformer model, which suffers from limited parallel processing capability and sensitivity to hyperparameter tuning, leading to improved accuracy in predicting freeze–thaw damage evolution. This study not only validates the superior damage prediction accuracy of the KAN-Transformer model but also offers an efficient method for field applications in asphalt concrete damage prediction.