Prediction method for void content of aggregate based on neural network model

The void content of aggregates (VCoA) considerably influences the mechanical response of asphalt concrete; the rapid prediction of VCoA can aid in designing high-performance concrete. To predict VCoA, the size and shape parameters of aggregate particles are measured using image analysis processing, and the results are used as inputs in a neural network model. Void content data is non-linear and interlacing; therefore, the Elman and back propagation neural networks are used to predict VCoA owing to their non-linear mapping abilities. The correlation coefficient in the data set and maximum prediction error are used to compare the robustness of the two models. Results indicate that the Elman neural network model can obtain a low prediction error and stronger correlation between predicted and simulated values when choosing the appropriate number of hidden layer nodes. Considering complex shapes of practical aggregates, the predicted VCoA using the Elman neural network model was corrected and verified via experiments; the results show that the error between the predicted VCoA after the correction and the actual VCoA is within 1%, which meets the accuracy requirements of engineering application. Thus, combining imaging technology and the neural network is an effective approach to predict VCoA.HIGHLIGHTSVoid content can be predicted by the particle size and particle morphological parameters and the deviation between the predictive value and the true value was within 1%.The rapid prediction model of void content based on the Elman neural network was established.EDEM simulation software was used to construct abnormally-shaped aggregate particles and to carry out void content simulation experiments.