Creep Data Generation Method Based on Physically Constrained Variational Self-encoder

Polyurethane is a commonly used material in the industrial field, which is prone to creep under the action of such mixed stresses as high temperature and high pressure. Thus affecting the stable operation of the equipment and obtaining reliable data to study its creep process has become an important challenge in this field. To address this problem, a combination of material physical characteristics of the condition of the variational self-encoder virtual sample generation method was proposed in this paper. Firstly, the necessary features to increase the diversity of features of the input samples were extracted according to the creep characteristic curve from the original data. Secondly, the loss function in the network was constructed through the physical information. Finally, the encoder was used to generate a specific condition of the creep curve. The experimental validation proved that the creep data under specific conditions are more accurate, which can provide a certain reference for models that need to be trained by a large number of different conditions of creep data.