Analysis of Structural Performance Evaluation of Asphalt Concrete Pavement Based on Unsupervised Learning

The structural performance evaluation of asphalt concrete pavements is critical for ensuring traffic safety and optimizing maintenance strategies. Traditional nondestructive testing methods often rely on complex models and extensive manual intervention, limiting their efficiency and real‐time applicability. This study proposes an unsupervised learning‐based approach to enhance the accuracy and automation of pavement performance assessment. A forward model of pavement dynamics under falling weight deflectometer (FWD) loading was established using the spectral element method (SEM) to generate time‐history deflection curves. A convolutional autoencoder (CAE) was then employed as an unsupervised learning model, trained exclusively on normal pavement data. Performance anomalies were detected by evaluating reconstruction error (ReError) and structural similarity (SSIM) between input and reconstructed deflection signals. The results demonstrated that the proposed method effectively identified abnormal pavement conditions, with ReError increasing and SSIM decreasing significantly as the dynamic modulus or thickness of pavement layers degraded. Linear relationships between dynamic modulus and deflection were observed for one‐ and two‐layer pavements, while three‐layer pavements exhibited nonlinear behavior due to asynchronous parameter variations. The unsupervised learning framework provides a robust, data‐driven tool for qualitative pavement performance evaluation.