Convolutional neural networks for predicting creep and shrinkage of concrete.

• CNN models for predicting creep and shrinkage of concrete are researched. • The constitutive law of concrete based on the proposed CNN models is presented. • An ABAQUS user subroutine of concrete creep and shrinkage is developed. • The performance of CNN models is better than the B4 model. • Developed user subroutine predicts well the long-term deformation of a RC beam. The problem of long-term deformation caused by creep and shrinkage (C&S) needs to be concerned in the design and service of concrete structures. Although various models have been developed to predict the C&S of concrete, more accurate and reliable prediction methods are still needed. The models of C&S based on convolutional neural networks (CNNs) are proposed in the study. The performance of the CNN models is verified by using 906 sets of creep experiment data and 1114 sets of shrinkage experiment data in the Northwestern University (NU) database. Besides, the K-means clustering algorithm is introduced to divide the data set into the training set, validation set, and test set, and the problem of uneven distribution of the data set on the time scale is overcome. Finally, the incremental viscoelastic constitutive law of concrete based on the developed CNN models is proposed, and the ABAQUS user subroutine for simulating C&S of concrete is developed. The availability of the user subroutine is validated by the creep and shrinkage test of a reinforced concrete beam. The research can provide reliable methods for the rapid prediction of C&S of concrete and the simulation analysis of long-term deformation of concrete structures. [ABSTRACT FROM AUTHOR]