ASPHALT STUDY BY NEURONAL NETWORKS - CORRELATION BETWEEN CHEMICAL ANDRHEOLOGICAL PROPERTIES

In this paper we investigate the prediction of rheological properties of bitumens using some structural parameters calculated from C-13 NMR data. This study was carried out using methods of quantitative structu re properties relationships (QSPR) and more particularly neural networ ks (NN). Such a mathematical tool can find out non linear relations be tween descriptors and properties. Two asphalt rheological properties, m (creep slope at low temperature) and G/sin delta (stiffness at high temperature) were selected, whereas the descriptors are the average m olecular parameters which characterize the hydrocarbon skeleton of bit umens. This work permitted to prove that the skeleton information cont ained in the average molecular parameters could be correlated to the m value but not to the G/sin delta. Thus, the low-temperature rheologi cal behavior appears to be highly dependent on the aliphatic part of t he bitumens.