Smart data-driven modeling has significantly advanced algorithm research in industrial applications by enabling data collection in virtual environments, effectively addressing the challenges of data labeling in real-world scenarios. Our research focuses on utilizing deep learning techniques to estimate the 6D pose of components in industrial settings. Traditional RGB-D-based pose estimation methods have struggled to meet the high-precision demands of these industrial scenarios. In contrast, point cloud-based approaches have shown promising potential in addressing these challenges. In this study, we introduce PointPET, a novel Transformer-based network designed for 6D pose estimation. PointPET leverages a self-attention mechanism to encode local features, facilitating end-to-end prediction of an object’s 6D pose from input data. Our evaluation of the ICD-4 dataset demonstrates that PointPET can predict position with an accuracy of 1mm and rotation angles within 5°, meeting the stringent precision requirements of industrial applications.