Artificial intelligence has transformed fields like medicine and finance, but it hasn't gained much traction in manufacturing. Factories present a different challenge for AI: They are structured, fast-paced environments that rely on precision and critical timing. Success requires more than powerful algorithms; it demands deep, real-time understanding of complex systems, equipment and workflow. A new AI model designed specifically for manufacturing, seeks to address this challenge and revolutionize how factories operate.
With support from the U.S. National Science Foundation, a team led by California State University Northridge's Autonomy Research Center for STEAHM has developed MaVila - short for Manufacturing, Vision and Language - an intelligent assistant that combines image analysis and natural language processing to help manufacturers detect problems, suggest improvements and communicate with machines in real time. Their goal is to create smarter, more adaptive manufacturing systems that can better support one of the most important sectors of the U.S. economy.
MaVila takes a different approach. Instead of relying on outside data, like information on the internet, it is trained with manufacturing-specific knowledge from the start. It learns directly from visual and language-based data in factory settings. The tool can "see" and "talk" - analyzing images of parts, describing defects in plain language, suggesting fixes and even communicating with machines to carry out automatic adjustments.
MaVila was trained using a specialized approach that required far less data than typical AI systems - an advantage in manufacturing, where data is often limited or expensive to collect. Therefore, the tool could be more accessible to small and medium-sized businesses that can't afford expensive AI tools, or the expertise required to run them.
Researchers trained MaVila using vast datasets of images paired with descriptive language. Then, they fine-tuned it in a lab setting by showing it pictures of 3D-printed parts with visible flaws, such as blobs, cracks or stringy filaments. In most cases, MaVila correctly identified the defects and suggested better printing settings.
The team also connected MaVila to mobile devices and robotic simulations. This allowed the model to operate in real-time scenarios, such as identifying a machine from a photo and generating step-by-step commands to adjust performance or fix a flaw - something that traditionally requires expert programming.
The development of MaVila was powered by the National Research Platform (link is external) (NRP) Nautilus - a federally funded partnership of over 50 institutions led by experts at UC San Diego that has received continuous support from NSF. To meet the enormous processing demands of training MaVila, the researchers turned to NSF-funded high-performance computing (HPC) systems. These HPC resources allowed them to simulate realistic manufacturing conditions, test edge cases and validate the AI's response and decision-making faster than traditional computing could allow.
This project marks a leap forward in intelligent, adaptive manufacturing. It empowers human workers, increases productivity and strengthens the U.S. position in a fiercely competitive global market. And it reflects years of public investment in computing, collaboration and AI innovation.
Technologies like MaVila are expected to boost domestic manufacturing, fuel economic resilience, and help prepare the workforce for future-ready industries. NSF's support ensures that cutting-edge research translates into practical tools that benefit everyday people and keep America at the forefront of innovation.
The achievement of the researchers reflects the results of years of public investment in computing infrastructure, cross-institutional partnerships and targeted AI research. Through initiatives like the NRP and widespread access to advanced computing resources, the NSF provides researchers with essential tools that accelerate innovation and translate lab research into real-world solutions.
The project represents a significant step forward in autonomous, adaptive manufacturing. By enabling factories to detect issues and optimize operations, MaVila supports human workers by helping them make decisions more efficiently, driving global competitiveness in a constantly evolving world.
Listen to NSF Discovery Files (link is external) wherever you get your podcasts.
