Transfer length data of pretensioned concrete generated by Generative Adversarial Network

10,000 fake data were generated using GAN based on the probability distribution of the real transfer length test data. The discriminator and generator networks consist of a total of 10 layers, and each layer is made up of 100 nodes. As with the ANN model, batch normalization and a 20% dropout rate were applied to each layer, and the PReLU activation function was used in the layers other than the output layer. In order to shorten the training time, Adam was used as an optimization algorithm (or optimizer), and the learning rates of the discriminator and generator were set to 0.01 and 0.0033%, respectively, to perform the training process up to 39,000 epochs. The result showed that the distributions and correlations of the fake data are similar to that of the real data