Virtual mix design: Prediction of compressive strength of concrete with industrial wastes using deep data augmentation

The adding of industrial wastes, including blast furnace slag and fly ash, to concrete materials will not only improve the working performance, but also significantly reduce the carbon emissions and promote the green development in civil engineering area. The traditional material designs are mainly indoor laboratory-based, which is complex and time-consuming. In this study, a virtual material design method, including deep data augmentation methods and deep learning methods, was employed to predict the compressive strength of concrete with industrial wastes. Three types of Generative Adversarial Networks (GANs) were employed to augment the original data and the results were evaluated. The test was conducted based on a small experiment dataset from previous literature, comparing with traditional machine learning methods. Test results show that the deep learning methods have the highest accuracy in compressive strength prediction, increasing from 0.90 to 0.98 (Visual Geometry Group, VGG) and from 0.83 to 0.96 (One-Dimensional Convolutional Neural Network, 1D CNN) after deep data augmentation, where the prediction accuracy of Random Forest (RF) and Support Vector Regressive (SVR) in traditional machine learning algorithms increase from 0.91 to 0.96 and from 0.78 to 0.86, respectively. In addition, a lightweight deep convolutional neural network was designed based on the augmented dataset. The results show that the lightweight model can improve the computation efficiency, reduce the complexity of the model compared with the original model, and reach a great prediction accuracy. The proposed study can facilitate the concrete material design with industrial wastes with less labor and time cost compared with traditional ones, thus can provide a cleaner solution for the whole industry.