Estimation and Validation of Gaussian Process Surrogate Models for Sensitivity Analysis and Design Optimization

The Mechanistic–Empirical Pavement Design Guide (MEPDG) is a powerful predictor of pavement distress, but it is computationally expensive to evaluate. Analyses that require many MEPDG evaluations, such as sensitivity analysis and design optimization, become impractical because of the computational expense. These applications are important in achieving robust, reliable, and cost-effective pavement designs. This paper develops Gaussian process (GP) surrogate models that, with a trivial amount of computational expense, accurately approximate the results of the MEPDG for each relevant distress mode. The GP is validated in accordance with three model metrics: average predictive percent error, predictive coefficient of determination, and Bayes factor. The GP models are then exploited for sensitivity analysis and design optimization, making these tasks computationally affordable.