Backcalculation of the Stiffnesses of Cement Treated Base Courses Using Artificial Intelligence

It is common practice to evaluate the condition of pavements by means of deflection measurements and by back-calculating the stiffness modulus of the various pavement layers. Practice has shown however that the commonly used back-calculation procedures not always result in realistic values for the stiffness of stiff cement treated bases (CTB). This becomes a problem when the stiffness of the base course is specified in contract documents. This study presents a procedure for the prediction of stiffness of CTBs from deflection measurements using two artificial intelligence (AI)-based methods, being support vector machines (SVM) and artificial neural network (ANN). The procedure is based on a data base of deflection bowls of 2880 three layer pavement. The structures consisted of an asphalt top layer, a cement treated base and subgrade. The deflection bowls were calculated using the multi-layer program BISAR. The results showed that SVMs produce better results than ANNs. However, an extra validation using 100 new data points showed the quality of ANN. It is therefore concluded that both ANN and SVM are powerful tools for accurate prediction of stiffness of CTB.