| Abstract (ENG): |
Manufacturing environments face growing challenges due to a diverse and often inexperienced workforce, where language barriers and limited exposure to complex assembly procedures lead to inefficiencies and safety risks. Tradi-tional training programs and supervision-based learning fail to provide real-time, context-aware support, while static solutions like manuals or training videos lack adaptability. Additionally, the retirement of experienced workers results in the loss of valuable tacit knowledge, threatening productivity and quality standards. Manufacturers must ensure workers receive precise, step-by-step task guid-ance without relying on costly human supervision or translation services.
To address these challenges, we propose a Retrieval-Augmented Generation (RAG)-powered Large Language Model (LLM) system for real-time, privacy-preserving multilingual assistance. Unlike conventional Artificial Intelligence (AI) solutions, RAG dynamically retrieves domain-specific knowledge, enhanc-ing accuracy, relevance, and reducing AI hallucinations. Our system lever-ages multimodal LLM capabilities, allowing workers to interact through text, images, and structured data for contextualized guidance. By processing diverse inputs such as scanned documents, assembly images, or operational logs, the RAG-LLM provides tailored, actionable insights.
Recent advancements demonstrate that AI-driven cognitive manipulation tech-niques can enhance robot-assisted workflows, improving efficiency in semi-structured environments. Integrating a locally deployed RAG-enhanced mul-timodal LLM offers a scalable, cost-effective workforce training solution. Ben-efits include reduced errors, improved communication, retention of expert knowledge, increased worker autonomy, and enhanced productivity. Local de-ployment ensures data privacy, security compliance, and reduced dependency on external cloud infrastructure. By addressing skill gaps, knowledge retention, and multilingual support, this system transforms assembly line efficiency, fostering a more resilient manufacturing workforce. |
| Citation: |
Siddiqui, Md Khalid and Engel, Tobias
(2025)
Leveraging Large Language Models for efficient assembly task support in multilingual manufacturing environments.
In: 4th International Symposium on Industrial Engineering & Automation (ISIEA) “Manufacturing 2030 – A Perspective to Future Challenges in Industrial Production”, June, 18-20, 2025, Bolzano (Bozen), Italy, Paper – Abstract No. 160.
|
