24 Prediction of Straining Actions in Rigid Pavements Dowel Bars through Artificial Neural Networks

A heavily instrumented test section was constructed on Corridor ‘H’ in West Virginia (Robert C. Byrd's Highway near Elkins) consisting of 30 consecutive concrete slabs. The test section allows long term monitoring of the performance of full-scale slabs and joints. This investigation puts into perspective the mechanistic behavior of slabs on grade from the early age of construction. Various parameters such as strains, temperature profiles, joint openings provide key-performance and reliable data needed for analysis of rigid pavements response to environmental loading. In this paper time histories from the test section are used to verify the use of dynamic artificial neural networks (ANNs) in predicting straining actions of load transferring devices consisting of dowel bars. In this respect, the dowel bars deformations in response to temperature and moisture variations are predicted using Time Delay Neural Networks as well as Elman∕Jordan Networks. The study shows that ANNs are capable of correlating the dowel bar bending moments to temperature gradients measured across the slab cross sections. In this study, comparing the results of ANNs and field measured bending moments indicate that Elman∕Jordan networks provided slightly better match than Time Delay Neural Networks (TDNN).