Backcalculation of Flexible Pavement Moduli from Falling Weight DeflectometerData Using Artificial Neural Networks

摘要

The goal of this research was to develop a method for backcalculating pavementlayer moduli from FWD data in real time. This was accomplished by training an artificial neural network to approximate the backcalculation function using large volumes of synthetic test data generated by static and dynamic pavement response models. One neural network was trained using synthetic test data generated by the same static, layered-elastic model used in the conventional backcalculation program WESDEF. That neural network was shown to be just as accurate but 2500 times faster. The same neural network was subsequently retrained using data generated by an elastodynamic model of the FWD test. The dynamic analysis provides a much better approximation of the actual test conditions and avoids problems inherent in the static analysis. Based on the amounts of time needed to create the static and dynamic training sets, a conventional program would likely run 20 times slower if it employed the dynamic model. The processing time of the neural network, on the other hand, is unchanged because it was simply retrained using different data. These artificial neural networks provide the real-time backcalculation capabilities needed for more thorough, more frequent, and more cost-effective pavement evaluations. Furthermore, they permit the use of more-realistic models, which can increase the accuracy of the backcalculated moduli. (MM).