APPLICATION OF ARTIFICIAL NEURAL NETWORK (ANN) IN DEVELOPMENT OF PREDICTION MODELS FOR PAVEMENT PERFORMANCE AND MATERIAL PROPERTIES

This dissertation presents the development and application of Artificial Neural Network (ANN)- based prediction models for Dynamic Modulus (E*), Dynamic Shear Modulus (|Gb*|, Phase Angle (b), Soil-Water Characteristics Curve (SWCC) parameters, and International Roughness Index (IRI). The IRI prediction model considering climatic and traffic conditions of Texas with data from the Long-Term Pavement Performance (LTPP) database with R 2 = 0.92 can be utilized by Local transportation agencies. An E* prediction model with three neurons, using 7400 data points obtained from 346 mixtures with R2=0.82 can bypass the need for laboratory tests. ANN-based |Gb*| and b prediction models were also developed with seven neurons and three neurons respectively. Both are independent of each other and perform better than the previous two models. ANN showed promising results in predicting SWCC parameters like af and cf. The models with equations are easier to use. Therefore, these models can be integrated into standard design procedures like MEPDG