Method to predict the interlayer shear strength of asphalt pavement based on improved back propagation neural network

Abstract

• The oblique shear test and 90° vertical shear test device were designed. • The interlayer shear tests of asphalt mixtures with different composite structures were carried out. • The BP neural network prediction model for the shear strength of asphalt mixture specimens were established. • The prediction model were verified by experiments. Given the complexity of the factors that affect the interlayer shear strength of asphalt pavement in typical steeply-sloped sections of areas with seasonally frozen soil, to predict and evaluate interlayer shear strength more accurately and rapidly, this study obtains experimental data by designing indoor direct and oblique shear tests, and establishes a three-layer improved back propagation (BP) neural network model to predict the interlayer shear strength of pavement with a structure of 6–20-1. The model uses 6 influencing factors of specimen combination types-tack coat type, a tack coat dosage, shear angle, temperature, and loading rate-as the input layer. The neural network trained, verified, and tested 230 sets of oblique shear test data, and completed the neural network’s universality test. The research results show that the predictive value of the shear strength under different viscous layer oil dosages and temperatures is very consistent with the results of universal testing experiments. Different types of specimens’ interlayer shear strength increases first and then decrease as the amount of tack coat increases, and compared with base asphalt and emulsified asphalt, SBS-modified asphalt has the best interlayer shear resistance when used as the tack coat, and the optimal dosage is 1.2 kg/m 2 . Temperature is under a significant negative correlation with the specimen’s interlayer shear strength. As the temperature rises from 20 to 60℃, the interlayer shear strength of different specimen types decreased at a faster rate, and the shear strength at 58℃ was only 15 %-30 % of that at 20℃. The constructed BP neural network prediction model has good convergence and superior performance. The model prediction error does not exceed ± 0.5, and the prediction accuracy is high (R 2 = 0.99). At the same time, the shear specimen’s shape does not affect the use of the interlayer shear strength prediction model, which can be utilized to predict the interlayer shear strength of the asphalt mixture specimens.