Artificial Neural Networks for Predicting the Response of Unbonded Concrete Overlays in a Faulting Prediction Model

Abstract

Transverse joint faulting is a common distress in unbonded concrete overlays (UBOLs). However, the current faulting model in Pavement mechanistic-empirical (ME) is not suitable for accurately predicting the response of UBOLs. Therefore, to develop a more accurate faulting prediction model for UBOLs, the first step was to develop a predictive model that would be able to predict the response (deflections) of these structures. To account for the conditions unique to UBOLs, a computational model was developed using the pavement-specific finite element program ISLAB, to predict the response of these structures. The model was validated using falling weight deflectometer (FWD) data from existing field sections at the Minnesota Road Research Facility (MnROAD) as well as sections in Michigan. A factorial design was performed using ISLAB to efficiently populate a database of fictitious UBOLs and their responses. The database was then used to develop predictive models, based on artificial neural networks (ANNs), to rapidly estimate the structural response of UBOLs to environmental and traffic loads. The structural response can be related to damage through the differential energy concept. Future work will include implementation of the ANNs developed in this study into a faulting prediction model for designing UBOLs.

Mechanistic-Empirical Faulting Prediction Model for Unbonded Concrete Overlays of Concrete

“…In a previous study, corner deflections, full lane width deflection basins, triangular deflection basins, and 2 ft by 6 ft deflection basins on the approach and leave slabs were all evaluated to determine which is able to best predict the development of faulting in UBOLs (8). It was found that the 2 ft by 6 ft basin size is the best at characterizing the entire area most heavily influenced by the load and also accommodates the 6 ft by 6 ft slab size that is becoming more common in overlay design (8,17,18).…”

“…Artificial neural networks (ANNs) are used to calculate the loaded and unloaded slab deflection basins for each axle and temperature loading conditions that are needed. The details on the development of these ANNs are provided in (17). The Totsky model is used to characterize the interlayer stiffness so that the effects of different interlayer materials can be considered.…”

“…Step 2 entailed the development of artificial neural networks so that the slow and cumbersome structural models themselves could be decoupled from the design process. The details of these two steps are not provided here and can be found in reference (11). The focus here will be on Steps 3 through 5.…”

Modeling the Development of Permanent Deformation in Asphalt Interlayers of Unbonded Concrete Overlays of Concrete Pavements

“…ðe p; predicted À e p; measured Þ e p;predicted (5) To ensure the model is sufficient, a model adequacy check was performed. The slope and the intercept of the trendline were checked to ensure linearity and an intercept close to zero.…”

“…The new rigid pavement is modeled as an effective slab, combining the Portland cement concrete (PCC) layer (the overlay) and base layer (asphalt interlayer). Therefore, damage to the interlayer resulting from stresses and deflections imposed by traffic loads is not directly accounted for in the analysis of the effective slab and therefore the predicted performance of the overlay ( 4 , 5 ). A study was conducted to determine the adequacy of using the UBOL module in Pavement ME for predicting the development of distress in UBOL.…”