Engineering and manufacturing organizations continue to face challenges when managing large repositories of 3D models, including manual shape comparison, inconsistent similarity assessments, and the time‑consuming navigation of extensive part libraries.
Similar Parts Search is a technology demonstration that explores how combining Graph Neural Networks (GNNs) with advanced 3D geometric processing can significantly improve and streamline the discovery of structurally similar 3D parts within large engineering model repositories.
The prototype addresses a core industry challenge: the lack of scalable, objective methods for identifying similar parts across growing, heterogeneous CAD libraries. By learning directly from geometric structure rather than relying on naming conventions or manually defined rules, the approach delivers consistent, repeatable similarity results across teams and product lines.
Beyond addressing core technical challenges—such as manual shape comparison, inconsistent similarity assessment, and the time-consuming navigation of extensive model libraries—the POC demonstrates the potential to deliver tangible operational and economic benefits across engineering, manufacturing, and supply-chain workflows. By providing richer geometric context and reliable similarity detection, Similar Parts Search enables:
- Faster manufacturing cost and lead‑time estimation, by quickly identifying previously manufactured or equivalent parts and reusing validated process knowledge.
- Reduced part proliferation and stock variation, minimizing the number of near‑duplicate components stored across product lines.
- Lower warehouse and inventory holding costs, achieved through consolidation of similar parts and improved reuse strategies.
- Simplified supply and procurement management, enabling more standardized sourcing, better supplier negotiations, and reduced administrative overhead.
- Improved design reuse and engineering efficiency, shortening development cycles and lowering the risk of redundant part creation.
As a proof of concept, Similar Parts Search demonstrates how AI‑driven geometric similarity can form the foundation for custom, enterprise‑grade solutions tailored to specific organizational workflows, data ecosystems, and integration requirements.
The approach leverages domain‑specific datasets, established expertise in geometric algorithms, and high‑quality engineering model collections, including the Mechanical Components Benchmark (MCB)—an open‑source dataset developed by Purdue University and distributed under the MIT License. Together, these elements demonstrate how AI‑driven geometric intelligence can transform both the technical discovery of similar parts and the business outcomes of engineering organizations.
Similar Parts Search offers the following functionality:
- Easily upload 3D models in a 3D viewer and organize them into datasets, keeping shape libraries structured and ready for exploration.
- Automatically prepare 3D models for analysis, with all preprocessing handled behind the scenes so users can focus on results rather than file cleanup.
- Leverage learned geometric characteristics to support more accurate and meaningful similarity comparisons across large part libraries.
- Create and maintain searchable digital profiles of parts, enabling persistent storage and reusability of similarity data.
- Quickly find the most similar parts using intuitive top‑k search, helping users avoid duplicate designs, evaluate alternatives, or identify compatible components.
- Compare similar parts by price, allowing users to assess cost differences and make informed decisions when selecting among geometrically similar options.
- Receiving the AI-generated response. Inspect models directly in the browser through an interactive 3D viewer that supports rotation, zooming, and other standard viewing controls.
- Validate similarity results visually using a synchronized dual‑pane comparison view that displays both the query and matched models side by side, reinforcing transparency and trust in AI‑assisted recommendations.
Watch a short demo video below to see the current functionality of the Similar Parts Search technology demonstration.
If you are interested in learning more about the demonstrated technologies and how they can be utilized for your organization’s needs, please contact us to discuss the details.
Read the full news article.
All third-party trademarks belong to their respective owners. For more details, please refer to the Third-Party Trademarks list on the Privacy and legal notices page.
