Thermal cracking prediction using artificial neural network

Thermal cracking of asphalt concrete is a form of pavement distresses that is of a great importance for northern Canadian jurisdictions where extreme cold conditions and temperature fluctuations are encountered. However, predictions of this distress have not been accurate enough which translated in premature road failures. This is mainly due to the absence of robust mechanistic constitutive models, which translated into reliance on empirical formulations and indicators with inherent limitations. This paper presents a scheme to overcome such shortcomings. It calls for taking advantage of the wealth of field information available from Long Term Pavement Performance (LTPP) sites and using an analytical approach to predict thermal cracking of pavements. This study presents the artificial neural network (ANN) technique as a promising method that can help designers predict thermal cracking based on data accumulated (over the years) from LTPP sites. Several ANN models were trained and tested using simple parameters such as road structure, material properties and environmental conditions as input to predict thermal cracking. Results of ANN simulations showed the potential of the technique to effectively predict low severity thermal cracking of LTPP sites that were not included in the training and testing phases and to delineate the critical factors governing this distress form.