Non-Linear Analysis of Concrete Pavement Construction by the Use of Artificial Neural Networks

An artificial neural network (ANN) model is developed for the estimation of the concrete pavement construction productivity. Data were collected with direct observation over an eight month period. A multi-layer feed forward (MLFF) network with a back-propagation (BP) algorithm is applied for modeling three main sub-tasks: concrete layering, concrete finishing and joints cutting. Two input neurons have been used representing the working width and length of a given surface with a constant concrete thickness and one output neuron provides estimates of the achieved productivity. One hidden layer with three slabs and different number of neurons for each individual model has been implemented, thus resulting in three distinct artificial neural networks for the study of the concrete pavement operations. Different types of scaling functions are incorporated in the models, so as to model the actual data in a more realistic manner. The developed neural network models indicate adequate convergence and generalization capabilities, as shown by the outcome of the validation process. The results are compared with the actual data from the field measurements as well as with estimates stemming from multiple regression (MR) models. It has been shown that for the first two sub-tasks the non-linear description is better, whereas for the third sub-task the regression analysis seems to provide more suitable results. The concrete pavement productivity prediction models developed herein enable planners and estimators to acquire more accurate predictions for different operational scenarios.


  • English

Media Info

Subject/Index Terms

Filing Info

  • Accession Number: 01486873
  • Record Type: Publication
  • Files: TRIS
  • Created Date: Jul 18 2013 1:47PM