Artificial Neural Network Model for Prediction of Depth Temperature of Asphalt Layers Using LTPP Data – Case Study: Ohio, USA

One of the critical environmental factors that affect the deformation of flexible pavements is the depth temperature of asphalt layers. This is due to the viscoelastic behavior of the asphalt mixtures. The stiffness of the asphalt layers has a significant effect on the structural capacity of flexible pavements. This property is a function of the asphalt layer temperature and changes daily and seasonally. As the temperature increases, the stiffness of the asphalt layer decreases, which increases the stress in the base and subbase layers of the pavement. Therefore, the pavement response to the applied loads is affected by the depth temperature. Hence, the depth temperature of asphalt layers is one of the most important and main factors in the analysis, design, and rehabilitation process of flexible pavements. Some predictive models have been developed to determine the depth temperature of asphalt layers in pavement maintenance and rehabilitation activities. These models, as an alternative to field and laboratory measurements of this factor, are low-cost, rapid, and simple methods to determine the depth temperature of asphalt layers. It should be noted that these models are based on the limited field and laboratory data, therefore, there is a need for developing new models for prediction of the depth temperature of asphalt layers in different traffic and climatic conditions. The objective of this study is to develop a model for predicting the depth temperature of asphalt layers based on climatic data. In recent years, Artificial Neural Networks (ANNs) have shown good performance as a useful tool for modeling physical events. The modeling method used in this study is a Back-Propagation Neural Network (BPNN) model that predicts the average hourly depth temperature of asphalt layers based on several variables, including the time of the day, desired depth from the pavement surface, average hourly air temperature, average speed and direction of the wind, minimum air humidity and total solar radiation. Data was extracted from the Long-Term Pavement Performance (LTPP) database. After extracting and preparing raw data, all the needed data were acquired from different data tables and linked to each other in a database. As a case study, data points collected from pavements in Ohio, USA, has been used for modeling. Also, to ascertain the presence or absence of multicollinearity between independent variables, the Pearson correlation test has been conducted. For this reason, the maximum speed and direction of the wind and maximum air humidity parameters were removed from the data set. According to the results of the Pearson correlation test, the average hourly air temperature has the most powerful impact on the average hourly temperature of the asphalt layer depth (correlation=95.2%). After training and testing the neural network, the performance of the developed model has been evaluated, and results were compared with a non-linear quadratic regression model. The results show that the developed model is more accurate than the regression-based model. In addition, the ability of the developed model in predicting the depth temperature of asphalt layers based on existing climatic data with a very good prediction accuracy (R2=0.96) and very low bias and error has been shown. Furthermore, the performance of the developed model has some restrictions for the prediction of depth temperature of asphalt layers. Other factors such as material characteristics can be scrutinized and applied to enhance the performance and applicability of the model.