Prediction of Welding Deformation in Stiffened Plate Structures Using a Data-Driven Def-Gan Model

Predicting welding deformation fields before implementing is crucial for improving the overall quality of ship and offshore structure construction. However, conventional numerical approaches and experimental tests are limited in their ability to efficiently predict welding deformation due to the high resource consumptions and modeling limitations. To address this issue, this paper proposes a data-driven analysis framework based on a deformation generative adversarial network (Def-GAN). This artificial neural network model adopts an adversarial architecture consisting of two sub-models, specifically a generator and a discriminator, enabling it to conduct precise and efficient predictions under varying welding conditions. An extensive database comprising welding conditions and deformation distributions of stiffened plate structures has been established to train and test the Def-GAN model. The deformation data within the database are obtained through numerical analyses. Additionally, customized image encoding algorithms are developed to transform the welding conditions and deformation data into input and output image samples, respectively. The results indicate that the Def-GAN model achieved a mean squared error of 0.003 and R-squared accuracy of 0.998 when compared to the numerical results for the test dataset. In addition to its application in predicting deformation fields of welded structures, the Def-GAN model serves as a promising tool for optimizing welding schemes.