Use of artificial neural networks for predicting skid resistance of hot mix asphalt concrete (HMAC) pavements
MetadataShow full item record
Skid resistance plays an important role in the design of surface courses of Hot Mix Asphalt Concrete (HMAC) pavements. Without sufficient skid resistance, necessary friction cannot be mobilized between the vehicle tire and the wet pavement and this could lead to hydroplaning. When the tire hydroplanes the vehicle is no longer under the driver's control and such a situation can result in an accident. Although various factors, such as the quality of tires and driver skills may also influence a vehicle's potential to skid, it is die responsibility of the pavement engineer to ensure that skid resistance of the pavement surface is maintained at an adequate level during die design life of the pavement. This, however, is not a simple task, as skid resistance is not a well-understood phenomenon. Skid resistance is not only influenced by properties of the aggregate used in pavement construction but also by several environmental factors and the mix designs of the surface course. The traditional approach to ensure sufficient skid resistance involves controlling the quality of the coarse aggregate used in the bituminous mix. Aggregate quality control is generally accomplished based on their performance in laboratory tests. One of the most relied upon tests is the Polish Value Test, which provides an indication of the skid resistance that the aggregates will be able to provide. However, questions have been raised regarding the reliability in using results from a single test procedures such as the Polish Value Test. Researchers have therefore relied on empirical data to create models that best explain skid resistance. A common method that has been used involves multiple regression models. This method has the drawback that the model is determined a priori and the data is tested to see how well it fits on the model. The approach used in this research overcomes this drawback. By using Artificial Neural Networks, or ANN, the model is not determined a priori. By providing the network with sufficient and carefully selected example data sets with known outputs, the network is allowed to learn the relationships among the variables. After the network has been trained, it will then be able to predict skid numbers for unknown output values. This feature of ANN is based on the biological neuron of the human brain. It is this feature that lets the network generalize and predict the skid number for a given set of pavement parameters. Another aspect of skid resistance that was studied in this research was the effect of skid resistance with respect to climatic changes. Several studies in the past have attested to the fact that skid numbers vary seasonally within a year as well as within a season depending on environmental factors such as rainfall and number of dry days prior to a significant rainfall. Prior to this study no comprehensive study had been done to evaluate climatic effect on highways in Texas. This study investigates the seasonal and environmental effects on pavements in Texas and develops a model to normalize skid numbers. The data required for this study was gathered over a period of three years from 1995 to 1997 in 55 pavement sections across Texas. Several laboratory and field tests were conducted over this period and a database was developed from this effort. This database formed the data source used in developing the architecture for the ANN as well as in determining the seasonal effect on skid resistance.