ANN-Based Fatigue and Rutting Prediction Models Versus Regression-Based Models for Flexible Pavements

Abstract

Roads are exposed to continuous deterioration because of many factors such as traffic loads, climate and material characteristics. In Middle East countries, incredible investments have been made in constructing roads that necessitate conducting periodic evaluation and timely maintenance and rehabilitation (M&R) plan to keep the network operating under acceptable level of service. The M&R plan necessitates performance prediction models, which represent a key element in predicting pavement performance. Consequently, there is always a need to develop and update pavement performance prediction models specially for fatigue and rutting distresses, which are considered the most major distresses in asphalt pavement. On the other hand, Artificial Neural Network (ANN) is considered the best solution to developing such models with high accuracy due to its brilliant mechanism in training, testing and evaluating the data. In addition, the ANN approach has the flexibility to change many parameters such as number of neurons, hidden layers and function type to obtain more accurate predicted models. The scope of this paper is to develop ANN-based fatigue and rutting prediction models for asphalt roads. The ANN-based models were developed using MATLAB 2017b software based on actual field data obtained from Long-Term Pavement Performance (LTPP) database. The models were developed for both wet and dry non-freeze climatic zones. Results indicated that the ANN approach can be used in predicting both fatigue and rutting distresses with high accuracy as compared with the developed statistical models’ approach, which were also developed in this study for both fatigue and rutting distresses.