Session: 03-04-02: Artificial Intelligent Applications in Manufacturing II

Paper Number: 167138

Adaptive GenAI-Empowered Manufacturing Training Using LLMs and GraphRAG 

The rapid advancements in smart manufacturing technologies have significantly increased the demand for a highly skilled workforce with evolving expertise such as artificial intelligence or machine learning, necessitating a shift from manually operating machinery to interpreting real-time data and supervising autonomous systems. However, existing workforce training methods have struggled to keep pace with evolving technological requirements. Traditional training approaches, based on standardized curricula and rigid instructional frameworks, lack the flexibility to update in real-time to reflect new tools, processes, and best practices. Moreover, the lack of real-time adaptability and personalized feedback hinders workers from acquiring the competencies necessary for smart and connected manufacturing environments. Furthermore, high costs and logistical constraints associated with classroom-based training further exacerbate these challenges, particularly for small and medium-sized manufacturers (SMMs) with limited resources. As a result, projections from Deloitte indicate that the U.S. manufacturing sector could face a shortage of 2.1 million skilled workers by 2030.

Artificial intelligence, particularly supervised learning, has enhanced workforce development by automating and augmenting a broader range of tasks beyond structured, repetitive work. However, traditional AI relies on rigid rule-based models that require extensive labeled data and lack adaptability. Overcoming these challenges necessitates a paradigm shift in workforce training—integrating contextualized problem-solving exercises and fostering creativity in decision-making to create more interactive and adaptive learning experiences.

Generative AI (GenAI) presents a transformative opportunity, as it can autonomously generate advanced contextual content—including texts, codes, images, and simulations—by leveraging pretrained large foundation models such as OpenAI's GPT-4 Turbo, Meta’s LLaMA 2, and Google’s Gemini models. Additionally, GenAI can interact and adapt to human workers in a natural way via texts and images, facilitating personalized and informed training contents, providing instant contextual feedback, and enabling interactive problem-solving and self-paced learning. Moreover, by enabling automatic updates to training materials to reflect state-of-the-art tools, technologies, and industry standards, GenAI enhances accessibility, supports continuous skill acquisition, and ensures that training is responsive to technological advancements, and workers’ skills remain up to date.

Building on these strengths, the objective of this study is to develop a GenAI-driven workforce training solution - Adaptive GenAI-Empowered Manufacturing Training (AGEMT) that integrates GenAI, particularly large language models (LLMs), with curriculum design and instructional frameworks to deliver highly personalized and adaptive learning experiences. Leveraging state-of-the-art LLMs, AGEMT presents an interactive training method that integrates real-time learner performance tracking, automated adaptation of instructional materials, and intelligent tutoring systems. By dynamically adjusting training content based on individual progress, providing real-time problem-solving assistance, and generating domain-specific knowledge representations and explanations, AGEMT enhances comprehension and easiness of skill acquisition. This approach enables context-aware guidance through complex tasks while offering instant feedback to reinforce learning outcomes, fostering a well-informed, continuously evolving, and self-paced learning experience.

We aim to implement AGEMT in real-world human-robot collaborative assembly and inspection tasks by utilizing LLMs to generate structured, step-by-step operational instructions. Furthermore, GenAI-driven feedback mechanisms systematically assess learner performance, identifying strengths and areas requiring improvement to dynamically tailor instructional content and problem-solving support for enhanced skill acquisition. To evaluate the effectiveness of AGEMT, we identified key performance metrics, including learning retention rates, training time reduction, skill acquisition accuracy, and adaptability to real-world manufacturing scenarios. By harnessing GenAI, our work contributes to bridging the skills gap in manufacturing, democratizing access to personalized training, and enhancing workforce adaptability in response to rapidly evolving industrial technologies.

Presenting Author: Xingyu Li Purdue University

Presenting Author Biography: Dr. Xingyu Li is currently an Assistant Professor in the School of Engineering Technology at Purdue University, West Lafayette. Before joining Purdue, he was an Adjunct Assistant Research Scientist at the Department of Mechanical Engineering at University of Michigan - Ann Arbor. Dr. Li received his Ph.D. degree in Mechanical Engineering from the University of Michigan – Ann Arbor in 2018. His research interests include smart manufacturing systems, supply chain management, deep learning, artificial intelligence, and optimization.

Dr. Li has authored 40+ papers in distinguished journals such as CIRP Annals - Manufacturing Technology, Scientific Reports, Journal of Manufacturing Systems, International Journal of Production Economics, Reliability Engineering and System Safety, European Journal of Operational Research. Dr. Li is currently a CIRP Research Affiliate, an ASTAR Visiting Fellow, an AnalytiXIN Fellow, and a corresponding expert in Engineering, he is also a recipient of the Best Reviewer of OMEGA in 2023, Journal of Manufacturing Systems Outstanding Reviewer, Best Paper Award at the 2019 IEEE Ai4i, Ford COVID-19 Innovation Challenge Award and Presidents Health and Safety Award. Dr. Li chaired or served as a member of the scientific and organizing committees for international conferences like DigiTwin 2024 for Manufacturing Sustainability, Safety, and Resilience, IEEE ReMAR 2024, CIRP CMS 2024 and 2025, and CIRPe 2023 and 2024. He has chaired sessions at notable conferences such as INFORMS 2023, IEEE AI4I 2023, NSFC-RGC 2023, and NAMRC 2024.

Authors:

Xingyu Li Purdue University
Wei Ye Purdue University
Dazhong Wu University of Central Florida
Yanchao Tan Purdue University
Nathan Hartman Purdue University
Martin Jun Purdue University
Ragu Athinarayanan Purdue University