Thermo-Mechanical Pavement Deformation Prediction Using Monte Carlo Simulation and Regularized Neural Networks: A Comparative Study of FFNN and LSTM Models

This study addresses the challenge of predicting pavement performance under the combined influence of traffic induced mechanical loads and daily thermal variations, a critical issue for road infrastructure in Morocco, where harsh climatic conditions and increasing traffic intensities exacerbate pavement deterioration. Traditional monitoring methods, such as visual inspections or simplified mechanical models, remain limited in their ability to capture the complexity and uncertainty inherent to thermo mechanical interactions. In contrast, artificial intelligence methods, particularly neural networks, have shown strong potential for modeling nonlinear phenomena and improving predictive accuracy in pavement engineering. Building on this perspective, the present research develops a predictive framework that integrates two constitutive equations reflecting thermo mechanical interactions, solved through deep learning architectures including feed forward neural networks and long short term memory networks, with and without dropout regularization. The study pursues a dual objective: to compare the predictive performance and robustness of these models, and to assess the reliability of their associated uncertainties, ultimately aiming to provide actionable insights for predictive pavement management and maintenance planning.