<p>Accurate 6D pose estimation plays a pivotal role in achieving precision assembly within industrial manufacturing scenarios. While existing methods predominantly focus on generic objects, they often adopt evaluation metrics that inadequately reflect the stringent precision requirements for industrial component assembly. To address these limitations, we propose PGA6D, a novel keypoint voting-based network specifically designed for industrial precision assembly. Our framework features three key innovations: First, we develop an efficient pillar-based architecture that achieves real-time performance in simultaneous object center detection and category recognition while serving as an effective point cloud-RGB filter for subsequent processing stages. Second, we introduce an instance-aware positional encoding module that enhances feature representation by explicitly incorporating spatial relationships between object instances and their surroundings. Third, recognizing the insufficiency of conventional evaluation metrics, we propose the additional contour lines (ACL) metric, a geometrically rigorous assessment criterion better aligned with industrial assembly requirements. Comprehensive experiments demonstrate that our method establishes new state-of-the-art performance on both the conventional ADD(-S) metrics and the proposed ACL benchmark. Notably, the ACL metric reveals critical performance differences between methods that traditional metrics fail to capture, particularly in high-precision assembly scenarios. This work advances pose estimation methodology while providing more appropriate evaluation tools for industrial applications.</p>