Performance Modelling of Porous Asphalt Concrete using Artificial Intelligence

ABSTRACT In this paper two models are presented that allow the amount of ravelling in porous asphalt concrete (PAC) wearing courses to be predicted 5 and 8 years after construction. The models are developed using visual condition data that were collected on a number of PAC sections for a period of 10 years. Artificial intelligence techniques such Artificial Neural Networks (ANN) are used for model development. The developed models are in excellent agreement with practice since the artificial intelligence techniques pointed out the importance of the same parameters as was known from practical experience. The models show e.g. that if the bitumen content by mass of PAC wearing courses is less than 4% only a poor performance of that layer can be expected. The developed models are the first ones that allow quantitative prediction the effect of mixture composition, traffic and climate on the performance of PAC.