Review of the Back-Propagation Neural Network Method as a Basis for Pavement Performance Assessment

Abstract

The conventional methods of pavement performance assessment indices were established by statistical analyses based on single item and multiple linear regression techniques. These regression models have many deficiencies and are not able to truly reflect the inherent complex nonlinear relationships among the performance indices. However, the Back-Propagation (BP) neural network method with ca comprehensive nonlinear dynamic system is able to address some of these weaknesses. In this paper, the International Roughness Index (IRI), Damage Rate (DR), Structure Strength Index (SSI), Sideway Force Coefficient (SFC), and Rutting Depth (RD) were selected as the five index variables. These variables are considered as some of the most significant factors that affect pavement performance. Additionally, these indices were easily classified as non-dimensional quantities and became input data units in the application of the BP neural network. In the study, Pavement Management Index (PMI) was accordingly sub-divided into five groups representing five grades; namely (1) excellent, (2) good, (3) medium, (4) subordinated, and (5) inferior. In this paper, pavement performance assessment based on the BP neural network method and PMI is presented along with a practical application example; followed by a summary of findings and recommendations