Research Area Description
With the rise of digital transformation and cloud computing in the manufacturing industry, on-demand cyber manufacturing services have begun to emerge with the goal of connecting geographically distributed designers with manufacturers through online interactive design tools and instant job quotes. Present-day cyber manufacturing services (e.g., Mfg.com, Proto Labs, Xometry, Fictiv) have ignited the transformation of the manufacturing eco-system, but are also suffering from limited manufacturing capabilities, poor manufacturing quality and scalability. These challenges stem from the lack of a systematic method to capture highly variable and difficult-to-scale manufacturing know-how and “tribal knowledge” to produce a required product, which traditionally are acquired by human experts through years of experience.
To this end, our efforts in Cyber Manufacturing as a Research Area aim to innovate and develop computational methods to capture critical manufacturing process capability knowledge from data to answer the following research questions: 1) How can we learn the relevant capabilities of manufacturing processes and resources (e.g., machines) from design and manufacturing data, and how effective is this learning? (2) With manufacturing knowledge derived from data, what methods can be used to automatically tailor or redesign products to guarantee manufacturability and enable automated process planning? These research questions are addressed by developing state-of-the-art computational tools using deep learning methods, which enable future evolution of cyber manufacturing services.
Projects
Federated-Learning Based Manufacturing Capability Selection for Cyber Security in Cyber Manufacturing Services
The emergence of platform-based manufacturing has substantially disrupted the way discrete parts are sourced and manufactured. However, the centralized business model of the manufacturing platforms has raised concerns for data ownership and cyber security of independent manufacturing suppliers. Contrary to centralized platforms, Cyber Manufacturing Service aims to connect designers with geographically distributed manufacturing suppliers by matching the design intentions with the manufacturing capabilities of candidate suppliers in a Cyber Manufacturing Marketplace. One of the key challenges in realizing the vision of Cyber Manufacturing Service is the lack of a computationally efficient mechanism for manufacturing capability search while maintaining data security of the proprietary datasets. In this work, we propose a federated learning-based deep unsupervised part retrieval model (FL-DUPR) to train a global embedding model without directly accessing proprietary datasets of suppliers.
Deep Unsupervised Assembly Supplier Matching
As product assemblies become more complex and manufacturing supplier networks continue to diversify, identifying suppliers with the appropriate manufacturing capabilities has become increasingly challenging. Conventional supplier recommendation approaches rely on manually defined rules or supplier-declared information, which often provide only coarse descriptions of capability and fail to capture the detailed geometric and structural characteristics of modern assemblies. In addition, assembly-level variation in shape, topology, tolerance requirements, and production scale creates a high-dimensional decision space that is difficult to model with traditional feature engineering. These limitations highlight the need for data-driven methods that can automatically extract capability information from assemblies and use it to support supplier identification. To address this need, we study a deep unsupervised assembly supplier matcher (DU-ASM) that learns assembly-level capability representations directly from multimodal design and manufacturing data and uses them for supplier retrieval in a fully data-driven manner.
Data-Driven Manufacturing Capability Modeling for Process Planning
There has been extensive research in the cyber manufacturing field aimed at solving capability modeling and process selection. Most of this work has focused on process-level capability modeling, answering questions such as: What processes can be used to manufacture a feature on a part? Does this part require milling, drilling, casting, and so on? However, there has not been much focus on manufacturer-level capability modeling. We know that different manufacturers, despite having the same machine tools and equipment, do not exhibit similar capabilities because they possess different sets of knowledge and experience. Therefore, in this ongoing research area called practical capability modeling, we want to bring the human influence in capability into the equation and provide a data-driven solution to address this problem. In addition, we are interested in gathering manufacturing process-planning knowledge to help manufacturers develop data-driven methods to have model that generates process plan for a given part more efficiently.
Select Publications
Zhao, C., & Melkote, S. N. (2023). Learning the manufacturing capabilities of machining and finishing processes using a deep neural network model. Journal of Intelligent Manufacturing, 1-21.
Zhao, C., Dinar, M., & Melkote, S. N. (2023). Deep learning and sequence mining for manufacturing process and sequence selection. International Journal of Production Research, 1-22.
Yan, X., Wang, Z., Bjorni, J., Zhao, C., Dinar, M., Rosen, D., & Melkote, S. (2023). Process-aware part retrieval for cyber manufacturing using unsupervised deep learning. CIRP annals.
Yan, X., Williams, R., Arvanitis, E., & Melkote, S. (2024). Deep learning-based semantic segmentation of machinable volumes for cyber manufacturing service. Journal of Manufacturing Systems, 72, 16-25.
Park, S., Rosen, D. W., & Melkote, S. N. (2025, August). Deep Learning-Based Multi-Level Retrieval of Assembly CAD Models. In International Design Engineering Technical Conferences and Computers and Information in Engineering Conference (Vol. 89206, p. V02AT02A028). American Society of Mechanical Engineers.
Contact Information
- Suyoung Park: spark3026@gatech.edu
- Farzad Vatandoust: farzad.vatandoust@gatech.edu