Artificial Neural Network Modeling of Pavement Performance using Expert Judgement

ABSTRACT The application of Artificial Neural Networks (ANN) in pavement performance modeling for a case where the period of data collection is relatively short is presented in this report. For this purpose, the Canadian Long-Term Pavement Performance (C-LTPP) database is used to develop a roughness prediction model. To overcome the deficiency of poor predictive results beyond the period of the available data, a simple approach based on expert judgements is proposed. The result of the proposed deterministic approach is compared with those of the Bayesian probabilistic method. The latter is developed by the Canadian Strategic Highway Research Program. It is shown that, for the case under study, the neural network roughness prediction model utilizing expert judgements seems to be more realistic when compared to the Bayesian model. The reason is attributed to the fact that for the neural network model, the expert judgements are used only as a complement to the existing data. Therefore, neither data is extrapolated nor a judgement is used when actual data exists.