An efficient dynamic backcalculation procedure for flexible pavement at a network level

This study proposes a procedure including the artificial neural network (ANN) and mayfly algorithm (MA) to backcalculate the layer property at a network level. A four-layer ANN was constructed to build the mapping between layer properties and deflection characteristics extracted from the load time-history and load-deflection hysteretic curves. The weights and biases for each neuron was determined using MA, and assigned to ANN to backcalculate the layer property. A theorical three-layer and field four-layer asphalt pavements were utilized to evaluate the performance of the proposed method. The backcalculated deflection time-history matched well with measured ones, indicating the backcalculated layer property can accurately simulate the pavement behavior. The backcalculated layer property from proposed approach is less scatter and more reasonable than that from iteration-based approach. The superiority of proposed approach is the mapping could parallelly runs for numerous stations, which is more suitable for pavement assessment at a project level.