Application of Neural Network Model to Forecast Short-Term Pavement Crack Condition: Florida Case Study

Abstract

Certain highway agencies use the crack index (CI) to enumerate pavement cracking and determine the rehabilitation priorities. Thus, accurate forecasting of CI is essential for pavement rehabilitation budgeting. Currently, mechanistic-empirical and purely empirical models are popular tools for forecasting pavement cracking. However, with a large data dimension, it is difficult to select appropriate mathematical function forms for the above models. This paper summarizes the results obtained from a case study in which single-year and multiyear back-propagation neural network (BPNN) models were developed to accurately forecast the short-term time variation of CIs of Florida's highway network. The BPNN models exhibited a remarkable ability to learn the historical crack progression trend from the CI database and forecast future CI values. The BPNN model was then validated by comparing the forecasted CIs with measured CI data for the year 1998. Lastly, the BPNN model results were compared to those of a commonly used autoregressive model and the BPNN model was seen to be more accurate than the autoregressive model. Hence, the BPNN models can be expected to make a significant impact on the efficiency of rehabilitation budget planning in particular, and pavement management systems in general.