Neural Network based model to Estimate Dynamic Modulus E* for mixtures in Costa Rica

Document Type: Regular Article

Authors

Auburn University

10.22115/scce.2019.188006.1110

Abstract

Various dynamic modulus (E*) predictive models have been developed to estimate E* as an alternative to laboratory testing. The most widely used model is the 1999 I-37A Witczak predictive equation based on North American mixtures laboratory results. The differences in material properties, traffic information, and environmental conditions for Latin American countries make it necessary to calibrate these models using local conditions. Consequently, the National Laboratory of Materials and Structural Models at the University of Costa Rica (in Spanish, LanammeUCR) has previously performed a local calibration of this model based on E* values for different types of Costa Rican mixtures. However, further research has shown that there is still room for improvement in the accuracy of the calibrated model (Witczak-Lanamme model) based on advanced regression techniques such as neural networks (NN).
The objective of this study was to develop an improved and more effective dynamic modulus E* predictive regression model for mixtures in Costa Rica by means of NN based models. A comparison of the predicted E* values among the Witczak model, Witczak-Lanamme model and the new and improved model based on artificial neural networks (E* NN- model) indicated that the former not only met the model adequacy checking criteria but also exhibited the best goodness of fit parameters and the lowest overall bias. The findings of this study also supported the use of more advanced regression techniques that can become a more attractive alternative to local calibration of the Witczak I-37A equation.

Keywords

Main Subjects