Faulting Prediction Model in Jointed Plain Concrete Pavement and determining the parameters affecting this failure with Artificial Neural Networks

One of the essential functional failures in concrete pavements is faulting. Predicting this failure can be used in various fields such as pavement design and pavement management systems. In this study, powerful tool of artificial neural networks has been used to predict this failure. Initially, using 32 input variables including traffic, weather and structural data, the artificial neural network architecture was determined by trial and error and then the specified architecture was properly trained. Among these 32 variables, in addition to the variables used in previous studies, new input variables that have not been studied so far, such as Poisson's ratio and elastic modulus of concrete slabs, have been considered. Then, with a new method, 19 important variables were identified and a new neural network model with 19 variables was constructed. The values of correlation coefficient, mean square error and mean absolute error for the model with 32 variables and 19 variables are equal to 0.97, 0.45, 0.43, 0.95, 0.54 and 0.6. Finally, using the random forest method, the importance of 19 variables was determined, of which the four most important variables are the annual cumulative number of days with precipitation greater than 12.7 mm (24%), elastic modulus (14%), Pavement life (12%) and base thickness (10%). elastic modulus is one of the most important input variables if this variable has not been studied in previous studies.