Automatic Leveling in DMLS Based on Deep Neural Networks

This study aims to explore how deep learning technology can be utilized to achieve automatic leveling in the Direct Metal Laser Sintering (DMLS) process. By transforming the complex task of automatic leveling into an image classification problem, we developed and implemented a priority-based strategy to relabel the training dataset, ensuring the quality and relevance of the data. On this foundation, we evaluated various deep learning architectures, including classic models such as ResNet and VGG, as well as the more recently emerged Vision Transformer (ViT) models.To further enhance the model’s performance, we drew on previous research findings and introduced the Convolutional Block Attention Module (CBAM) attention mechanism into the ResNet18 architecture. The addition of CBAM enabled the model to focus more effectively on key features within images, thereby significantly improving its recognition accuracy in the task of automatic leveling. Experimental results demonstrated that the improved model achieved notably higher accuracy in automatic leveling compared to models without the CBAM mechanism, fully validating the tremendous potential of deep learning methods in realizing precise automatic leveling in the DMLS process.The method proposed in this study not only greatly increased the level of automation and reduced the frequency of required human intervention but also significantly enhanced the consistency and reliability of the DMLS process. This provides strong support for future technological innovations and lays a solid theoretical foundation for practical industrial applications.