Distributed fiber optic sensing and Informer-LSTM prediction for damage evolution of asphalt concrete under freeze–thaw cycling

Abstract

In this study, distributed fiber optic sensing together with deep learning frameworks was developed to accurately capture the strain distribution characteristics of asphalt concrete under different conditions. As the performance and damage evolution of asphalt concrete in low-temperature environments have garnered increasing attention, the effects of freeze–thaw cycles on crack evolution in asphalt concrete were investigated under various test conditions, such as saturated versus unsaturated states and elevated versus reduced temperatures. The informer and long short-term memory (Informer-LSTM) networks time series prediction model was proposed to predict the strain state of the specimen across both temporal and spatial dimensions. The results indicate that the distributed fiber optic sensor effectively monitors local damage during the freeze–thaw cycles of the trabeculae and identifies the location of cracks. Furthermore, the Informer-LSTM model accurately captures and predicts strain distribution at the cracks, indicating the superior robustness and adaptability of the parallel time-series prediction model. This research significantly contributes to improving the durability and safety of pavement structures, providing a scientific foundation for assessing the durability and safety of road infrastructure.