ABSTRACT
The falling weight deflectometer (FWD) is an internationally used device for measuring deflections and evaluating the quality of pavement. But FWD can only take measurements of a limited number of kilometers of road per day. On a large scale, this method is expensive in terms of time and resources. Moreover, in some cases, data sets obtained from FWD are not sufficient, or they may be limited or missing, which makes the planning of maintenance and rehabilitation measures more error-prone. Aiming to overcome these limitations and to improve the method, in this work a new artificial neural network (ANN) approach was developed in order to calculate the deflection at any arbitrary point on the entire route to complement and substitute experimental measurements. A feed-forward ANN model was developed in the MATLAB computing environment based on backpropagation by a multilayer perceptron (MLP) network for asphalt pavement. A MLP neural network model was chosen, because it can deal with those cases, in which FWD deflection data is not sufficiently available. These networks are particularly suited to modeling complex data, as provided by FWD, due to the capability of ANN to learn complex nonlinear behavior. With at least 150 data sets, a model can be trained through ANN, that has a mean square error of less than 1 percent. Fewer measuring points are needed because the missing data can be calculated from the formulation. Thus, this method offers a great potential for the optimization of the traditional measurements in two ways: First, it optimizes measurement costs, and second, it significantly enhances the accuracy of road maintenance planning. Besides, it helps solving the problem of limited or missing data sets.
