A welding sequence optimization method of multilayer thin-walled structures via combined architecture of convolutional long short-term memory-UNet and non-dominated sorting genetic algorithm II

Numerous welding seams in multilayer thin-walled structures of ship blocks could include thousands of welding sequences and lead to various structural deformations, significantly undermining the manufacturing quality. Welding sequence optimization based on numerical finite element (FE) simulations needs repeated model modification and calculation, facing challenges of time-consuming cost. Thus, this paper proposed a novel welding sequence optimization method based on a combined architecture of convolutional long short-term memory-UNet (ConvLSTM-UNet) and non-dominated sorting genetic algorithm Ⅱ (NSGA-Ⅱ), reducing welding deformation of multilayer thin-walled structures of ship blocks. The ConvLSTM network was used to extract the spatiotemporal characteristics of welding seams, and then welding deformation was rapidly predicted by the UNet network. NSGA-II was employed to automatically generate thousands of welding sequences, which would be input to the ConvLSTM-UNet network for fitness calculation. The multi-objective function consisted of distortion unevenness of each layer and the maximum flatness was applied for fitness evaluation and regeneration of new welding sequences. The optimized welding sequence could reduce the maximum deformation of the multilayer thin-walled structure of ship blocks up to 40.8 %.