Thermal deformation prediction for additive manufacturing of thin-walled components based on multi-layer transfer learning

Abstract This paper presents a thermal deformation prediction method for additive manufacturing of thin-walled components based on multi-layer transfer learning (MTL). The printability is forwardly designed via multi-objective optimization (MOO) by evaluating scanning length, spot amount and segment amount, accompanied by support material. To avoid the burdened and time-consuming simulation of FEM for various geometric characteristics of thin-walled components, the feed-forward multi-layer perceptron was constructed as the main structure of MTL to rapidly obtain temperature and deformation distributions of manufactured parts. The proposed method is verified by the SLM of mechanical unshrouded turbine. The metallographic diagrams of manufactured components were generated to observe the fabricating quality and verify the effectiveness of the MTL-based method. The metallographic experiment of the fabricated piece proves that the main microstructure of the cross-section of molten pool is spindly columnar crystals. The cross-section morphology and size of the molten pool is different due to different process parameters, making the width of grain is about 1µm. The proposed method is especially useful for metal 3D printing under uncertainty.