Deep learning based generative adversarial networks for generating individual jumping load

The existing individual jumping load models are usually based on artificially extracted features, such as shapes of the impulses, impact factors and contact ratios. Meanwhile, the deep learning allows the model to automatically extract features from the samples. Moreover, the emergence of generative adversarial networks greatly promotes the quality of the generated samples. Therefore, based on conditional generative adversarial networks and Wasserstein generative adversarial networks with gradient penalty, a generative adversarial network model for individual jumping loads is proposed. The generator of the model has five fully connected layers and a one-dimensional convolutional layer, while the discriminator has five fully connected layers. To collect real samples, experiments are conducted on Tongji University and Sheffield University. After one million steps, the generator can generate high quality samples that are very similar to the real samples. Finally, by comparing the power spectral density and single degree of freedom response of the generated samples with those of the real samples, it is further proved that the generative adversarial network proposed in this paper can be used to generate individual jumping loads. © 2019, Nanjing Univ. of Aeronautics an Astronautics. All right reserved.