Distributed Fiber Optic Sensing Integrating TCN-Transformer for Damage Evolution Analysis of Asphalt Concrete under Freeze–Thaw Cycles

In this paper, an artificial intelligence-driven distributed fiber optic sensing technology was used to monitor the mechanical behavior of asphalt concrete under different loading forms to address the problem of damage evolution prediction of asphalt concrete. Experiments of static loading, fatigue loading, postdamage unsaturated freezing and thawing cyclic action, and postdamage saturated freezing and thawing cyclic action were designed in order to collect the strain data of the specimens under the four working conditions. Subsequently, the time-series prediction model of the temporal convolutional network-Transformer (TCN-Transformer) was developed to predict the strain distribution of specimens obtained from distributed fiber optic sensors under the four aforementioned conditions. In order to evaluate the accuracy of the TCN-Transformer time-series prediction model in predicting the strain distribution of asphalt concrete under different damage cycles, the root mean square error, the mean absolute percentage error, and the coefficient of determination (R2) were calculated to quantify the accuracy of the model, utilizing the spatial distribution points of the fiber optic sensors as samples for comparing the true values with the predicted values. Subsequently, the performance and prediction capabilities of the proposed model were evaluated across datasets of varying sizes. The performance of the model and the fit of the predicted values to the true values were evaluated for different sizes of datasets.