A Computational Tool for the Reliable Prediction of Pavement Performance based on Temperature Data

Abstract

This study presents a computational tool for predicting pavement performance using long-term temperature data from thermocouple trees embedded in three flexible and two rigid pavement sections at the MnROAD facility. The research leverages spectral and probabilistic analyses to assess thermal behavior and its impact on pavement condition. Temperature measurements, supplemented by weather data, were processed to address missing data and artifacts using compressed sampling, ensuring a uniform 15-minute sampling interval. Spectral analysis techniques based on Fourier Transform and Wavelet Analysis with Generalized Harmonic Wavelets were used to model pavement layers as a cascade of filters, revealing the time-varying behavior of the filters' gain and phase shift, which indicates that they are sensitive to aging, moisture, and compaction. Wavelet analysis provided superior temporal resolution for detecting transient thermal phenomena. A probabilistic framework using Markov Chain Monte Carlo (MCMC) methods estimated thermal diffusivity coefficients, achieving residuals below 1.17 degree C and robust uncertainty quantification. The results highlight distinct thermal responses across pavement layers, with asphalt showing uniform behavior and base/subgrade layers exhibiting environmental sensitivity. Interfaces between layers displayed significant time-dependent changes, potentially linked to densification. Implemented as a modular Python package with Jupyter notebook examples, publicly available on GitHub, the tool offers a scalable solution for pavement monitoring. This research demonstrates that thermocouple-derived temperature data, when analyzed with advanced computational methods, provides reliable indicators of pavement degradation, supporting data-driven infrastructure management decisions.