Deep learning-based detection of part deformation for laser powder bed fusion process

During the manufacturing process of parts made by laser powder bed fusion (LPBF), the dimensional accuracy of the part is one of the most important factors affecting the quality of the part and therefore in-situ real-time online monitoring of the part deformation is essential. To address these issues, a complete visualization method for LPBF part deformation detection has been developed in this study. Aiming at the problem that traditional digital image processing methods and machine vision used in the existing LPBF image segmentation are greatly affected by illumination and image quality, this paper uses a semantic segmentation model based on deep learning for the segmentation of powder bed background and parts. By comparing the segmentation performance of FCN, U-Net and DeepLabV3 models in different configurations, the U-Net(32) model with 32 basic convolutional kernels was chosen as the base framework, which were reduced by half compared to the U-Net model. The model has the advantages of effective segmentation, faster speed and smaller model, satisfying the basic requirements of real-time and easy to be deployed in industrial environments. In addition, this paper investigates the deformation intervals of parts without significant deformation and the variation pattern of deformation during the growth of warpage defects, which can provide reference information for feedback control in the future.