EVALUATION OF RELATIONSHIPS BETWEEN PAVEMENT DISTRESS PARAMETERS

Abstract

A Road pavement continuously deteriorates under the combined actions of traffic loading and environment. The ability of the road to satisfy the demands of traffic and environment over its design life is known as performance. Modeling the relationship between. distresses is useful for efficient pavement management system of the transportation infrastructure. The relationship between these parameters will be very much useful to the road deterioration studies and design and data analysis of flexible pavements. In general, the evaluation of pavement distress parameter of a flexible pavement is carried out by using statistical based modeling techniques such as linear and non-linear correlation and regression analysis. This traditional method is complex due to non linear relationship among the roughness and distress parameters. The predictive ability of these models goes down as the number of explanatory parameters increases. The motivation for adding neural network is relevance to problem and capable of mapping complex nonlinear relationship among the modeling parameters. Also the approach of neural network modeling in non-parametric, and do not require any assumption about the functional form of the underlying distribution data. Therefore these modeling tools are attractive and produce better results than the traditional modeling techniques. This dissertation demonstrates the relationship between pavement serviceability and roughness and also relationship among distress parameters and pavement roughness. For this purpose statistical correlation techniques including linear and non-linear modeling approaches._ and Back-propagation neural network models were developed. By using MATLAB and SPSS software tool, models were developed to understand the relation among roughness and pavement distress of various flexible pavements of national highways. Network Survey vehicle data of four National Highways section available at Road Development Planning Management Division of CSIR-CRRI, New Delhi was considered. The relative importance of various input variables used in the present study were evaluated through partitioning of weights algorithm. The same data sets which were used to develop traditional statistical models have also been used for calibrating and validating the ANN models. The comparative analysis is made between these two modeling approaches. On the basis of the results obtained in this study, ANN based pavement performance predicting models are better than the traditional methods.