Resilient modulus (Mr) is a critical input for pavement design as it is the main property used to evaluate the contribution of subgrade to the overall pavement structure. Considering this, practitioners need simple and accurate ways to determine the Mr of in-situ subgrade without the need for expensive and time-consuming testing. The objective of this study is to develop a generalized regression prediction model for in-situ Mr of subgrades, compare it with established prediction models, and assess the model’s predictions on pavement performance using the Mechanistic-Empirical Pavement Design Guide (Pavement ME). The prediction model was built using field data from 30 pavement sections studied in the Long Term Pavement Performance (LTPP) Seasonal Monitoring Program where backcalculated modulus from falling weight deflectometer testing, in-situ moisture contents, and subgrade material properties were considered in the model. Based on the results, it was found that liquid limit, plasticity index, WPI (the product of percent passing #200 and plasticity index), percent coarse sand, percent fine sand, percent silt, percent clay, moisture content, and their respective interactions were significant predictors of in-situ Mr values. The findings showed that the generalized regression approach was able to predict Mr more accurately than predictions from the Witczak model. To assess the application of the predictive model on pavement performance, three LTPP sections located in New York, South Dakota, and Texas were analyzed to predict the rutting performance based on Mr values obtained from the developed generalized prediction model and those obtained from the current Pavement ME model and then compared with rut depths measured in the field. The findings showed that, for coarse-grained subgrades that have a low degree of plasticity, the generalized regression model predicted rutting performance similar to the embedded Pavement ME model. For fine-grained subgrades, the developed model tends to predict lower rut depths which were closer to the field measured rut depths. Overall, the generalized regression approach was successfully applied to create a simple, practical, cost-effective and accurate Mr prediction model that can be used to estimate the stiffness of subgrades when designing and evaluating pavements.
Development and Evaluation of a Learning Level Prediction Model Based on the Growth Curve About the Long-Term Learners
