Development of a Predictive Model Based on an Artificial Neural Network for the Semicircular Bend Test

Abstract

One of the major distresses in asphalt pavements is fatigue cracking. To avoid premature cracking failure in the field, it is necessary to characterize an asphalt mixture’s fracture resistance in the laboratory before the mixture is produced and constructed. In 2014, Louisiana decided to implement the semicircular bend (SCB) test as part of a balanced mix design procedure. However, fabrication and testing of SCB specimens can take up to 7 days after the asphalt mixture has been successfully designed in accordance with specification criteria. This study developed a predictive model for the SCB that was based on an artificial neural network that used the volumetric properties of asphalt mixture. This predictive model can be used by practitioners during the mixture design process to estimate the critical value of the J-integral, or Jc. To formulate and validate the model, 31 asphalt mixtures representing a wide range of design and production practices were tested with the SCB test. Statistical analysis (Pearson’s correlation, coefficient of determination, and the general linear model procedure) was then used in determining correlations between the dependent and independent variables and in the development of the predicted SCB test model. In addition, multicollinearity among and between independent variables was evaluated. The artificial neural network was used to develop and validate the SCB model. It is shown that the developed model can be used to predict the critical strain energy release rate, Jc, of aged asphalt mixtures with reasonable accuracy.