Prediction of structural deformation of a deck plate using a GAN-based deep learning method

During the manufacturing process, vessels typically undergo structural deformation during the erection stage because of the heavy loads they are subjected to. In this study, the out-of-plane mechanical deformation of a deck plate in an erection stage was predicted using a deep learning model. As model inputs, the initial deformation of the deck plate after assembly, the dimensional information of all the reinforcing structures installed on the deck plate, and the normal reaction force acting along the boundary in the erection stage were used, and the artificial intelligence model was trained to create an image of deformed shape of the deck plate from the inputs in a supervised manner. Three different types of commercial vessels were used in the training, and the training data were supplied from a nonlinear buckling finite element method (FEM) simulation. The adopted deep learning model was a convolutional neural network-based generative adversarial network, which was designed to translate the input images to the deformation predictive image. The deep learning model successfully predicted the 3-D deck distortion, and the accuracy reached 99.7794% (R-Squared) with respect to the FEM results. The prediction time was within a few seconds. •An image-based deep-learning model for predicting deck plate deformation during ship manufacturing is proposed.•Images containing initial deformation, reinforcing structures, boundary forces were used as inputs.•Nonlinear FEM buckling simulation was used to prepare data.•A convolutional neural network-based generative adversarial network was employed.•An R-squared accuracy of 99.7794% is achieved.