Characterization of hydration and dry shrinkage behavior of cement emulsified asphalt composites using deep learning

摘要

This paper introduces a study of characterizing hydration and dry shrinkage behavior of cement emulsified asphalt composites (CEACs) through a deep-learning framework. The deep-learning framework consisted of two parts: generative adversarial networks (GANs) and deep neural networks (DNNs). The GAN part was first developed to map the design parameters of a CEAC to its X-ray powder diffraction (XRD) spectrum and scanning electron microscope (SEM) images. The DNN part was then designed to predict the dry shrinkage behavior of the CEAC based on its design parameters and the outputs of the GAN part. Finally, the effectiveness of the deep-learning framework was tested by 36 groups of CEACs. The results showed that the outputs of the GAN part, synthetic XRD spectrums and SEM images, were close to the measured data. Thus, the synthetic data were capable of characterizing the hydration processes and the microstructure of the CEACs. The DNN predicted the dry shrinkage ratios of the 36 groups with a 2.70% average error, demonstrating its high and stable precision. The feature vectors in the DNN provided a new method to characterize the effects of the design parameters on the dry shrinkage ratios. From the distribution of the feature vectors in a two-dimension space, we found that the curing time had the most significant effects on the dry shrinkage ratios, followed by the aggregate grading and the contents of cement and emulsified asphalt. (C) 2020 Elsevier Ltd. All rights reserved.