Development of the non-destructive monitoring methods of the pavement conditions via artificial neural networks

Abstract

Non-structural parameters like surface defects and ride quality were frequently used, as a practical index for the rehabilitation selection process. The key purpose of this study was the assessment of using artificial network technology as a support for decision-makers about paving maintenance concerning the structural condition compared to the conventional, time-consuming, effort, and costly methods. The structural model was established based on the deflections from the FWD, (asphalt and base) layers thickness, surface temperature, precipitation rate, AADTT, traffic volume of class 9 and base layer type. The data used in building the developed ANN model is related to a previous study of flexible pavement structures on the M4 highway in the Russian highways network during the five years 2013-2017. The ANN model was built, trained, and tested by the Matlab program. The focus was on calculating roughness, fatigue, and rutting values as they are the most common pavement distress on site. We used the logistic model equations, developed by the Federal Highway Administration's Long-Term pavement Performance (LTPP) to calculate the three pavement distress that will be used as output variables while training the ANN model. The ANN model presented a high performance in predicting the three pavement distress (fatigue, roughness, and rutting) where the R- squared value was equal (1, 0,999, and 1), respectively for the forecasting sections.