On December 22, 2025, researchers at the Korea Advanced Institute of Science and Technology (KAIST) unveiled a breakthrough that could transform the way factories operate, especially those relying on the intricate process of injection molding. The team, led by Professor Seunghwa Ryu and Professor Yoo Seung-hwa from the Department of Mechanical Engineering and the Inno-Core PRISM-AI Center, developed an artificial intelligence system that not only automates the management of factory machines but also serves as a digital mentor for workers on the factory floor.
Injection molding, the method behind most of the world’s plastic products, is deceptively tricky. It involves melting plastic and injecting it into molds to mass-produce identical parts. But even minor fluctuations in room temperature or humidity can spell disaster, causing defects that render entire batches unusable. Traditionally, factories have leaned on the intuition and "gut feeling" of seasoned workers to tweak machine settings and maintain quality—expertise that’s becoming scarce as experienced workers retire and a growing number of foreign laborers, often facing language barriers, enter the workforce.
According to Aju Press, the KAIST team tackled this looming skills gap by designing two interconnected tools. The first is a generative AI engine that acts like a digital brain, analyzing real-time factory conditions—say, the current humidity level—and calculating the ideal machine pressure and speed. The second innovation is "IM-Chat," a digital assistant built on a large language model (LLM). Workers can ask it questions in their native language, such as “What is the best pressure when the humidity is 43.5 percent?” IM-Chat doesn’t just spit out a generic answer; it triggers the AI engine to perform the necessary calculations and then delivers precise settings, complete with explanations drawn from the factory’s technical manuals.
The research, published in the Journal of Manufacturing Systems in April and December 2025, marks the world’s first instance of generative AI technology autonomously optimizing injection molding processes while making expert-level knowledge accessible to anyone on-site. The project was a true team effort, with doctoral candidates Junhyeong Lee, Joon-Young Kim, and Heekyu Kim serving as co–first authors, and was supported by the Ministry of Science and ICT, the Ministry of SMEs and Startups, and the Ministry of Trade, Industry and Energy.
What makes this AI system stand out is its reliability. Previous attempts to automate this process using AI, such as models based on Generative Adversarial Networks (GANs) or Variational Autoencoders (VAEs), struggled with error rates between 23 and 44 percent. The KAIST team’s diffusion model–based approach, trained on months of real factory data, slashed that error rate to just 1.63 percent. In practical terms, this means the AI could set up machines to produce high-quality parts—without any human intervention—at a success rate that rivals or even surpasses the best factory veterans.
But the innovation doesn’t stop with accuracy. The IM-Chat system is designed for the realities of today’s manufacturing floors, where language diversity is the norm. Thanks to its multilingual capabilities, a novice or foreign worker can receive the same level of decision-making support as a seasoned expert, simply by asking a question in plain language. The AI not only provides the answer but also backs it up with references from technical manuals, ensuring transparency and trust in its recommendations.
Professor Seunghwa Ryu described the achievement as “a case where we addressed fundamental problems in manufacturing in a data-driven way by combining AI that autonomously optimizes processes with LLMs that make on-site knowledge accessible to anyone.” He added, “We will continue expanding this approach to various manufacturing processes to accelerate intelligence and autonomy across the industry.”
The implications of this research stretch far beyond plastic. The team envisions applications in other manufacturing sectors—think molds, presses, extrusion, 3D printing, batteries, and even bio-manufacturing. The technology’s adaptability comes from its core architecture, which integrates generative AI and LLM agents through a Tool-Calling approach. This means the AI isn’t just following a script; it can autonomously decide which functions or programs to invoke based on the situation at hand, a paradigm shift in how manufacturing AI operates.
In the words of Professor Yoo Seung-hwa, “This is a case where we solved the core problems of manufacturing by using data-based AI. By combining an AI that can optimize the factory process on its own with a system that can explain that knowledge to anyone, we hope to make many different types of industries more independent and automated.”
The team’s work is also notable for its practical validation. In real-world factory tests, the AI-generated conditions were applied to actual machines, resulting in the successful production of defect-free parts. This hands-on demonstration not only proved the system’s reliability but also highlighted its potential to reduce dependency on human intuition and minimize costly trial-and-error adjustments.
The research papers—“Development of an Injection Molding Production Condition Inference System Based on Diffusion Model” and “IM-Chat: A multi-agent LLM framework integrating tool-calling and diffusion modeling for knowledge transfer in injection molding industry”—were published in the Journal of Manufacturing Systems, a top-ranked international journal in engineering and industrial fields. The recognition from such a prestigious publication underscores the significance of the team’s achievement.
For factories facing the dual challenges of an aging workforce and increasing linguistic diversity, this technology couldn’t have come at a better time. As more experienced workers retire and new, often foreign, laborers take their place, the risk of losing institutional knowledge has been a growing concern. The KAIST solution offers a way to preserve and democratize that hard-earned expertise, ensuring consistent quality and efficiency regardless of who’s on shift.
Looking ahead, the researchers are optimistic about expanding the technology’s reach. The combination of autonomous process optimization and accessible, multilingual knowledge transfer could serve as a blueprint for manufacturing sectors worldwide. As industries continue to embrace automation and digital transformation, tools like the KAIST AI system may become essential for staying competitive in a rapidly evolving global market.
In an industry where a single miscalculation can lead to costly defects and downtime, the ability to blend machine precision with decades of human know-how—now available at the tap of a screen—signals a new era for manufacturing. The KAIST team’s achievement stands as a testament to the power of innovation, collaboration, and the enduring value of making expertise accessible to all.
