Pretrained AI detects factory defects using fewer examples, potentially easing costly setup

edited by Gaby Clark, reviewed by Robert Egan

This article has been reviewed according to Science X's editorial process and policies. Editors have highlighted the following attributes while ensuring the content's credibility:

Add to Preferred Sources CLIP’s architecture and workflow. Credit: Journal of Quality Technology (2026). DOI: 10.1080/00224065.2026.2640171 A University of Dayton statistician and an international team of researchers are showing how artificial intelligence can help factories detect flawed parts without the expensive step that keeps the technology out of reach for most manufacturers.

Factories have used cameras and software to catch bad parts for years. The problem has always been what it takes to get started. Before an inspection system can recognize a crack, scratch or bad weld, the system needs to be fed thousands of labeled photos of those exact flaws. Gathering them takes months. For a small or midsize manufacturer, the cost alone can end the conversation.

Ying-Ju Chen, UD associate professor of mathematics, played a key role on a team of researchers from universities in the U.S. and Europe whose study, published in the Journal of Quality Technology, showed that a model built by OpenAI can inspect manufactured parts after seeing a relatively small number of examples of a particular defect. Because the model already was trained on hundreds of millions of everyday images, it doesn't have to learn what a defect looks like from scratch.

"Automated visual inspection is not new, but many existing approaches require large amounts of labeled data, which can be difficult and costly to collect," Chen said. "Our goal was to investigate how well a pretrained AI model could identify defects using only a small number of examples and whether the approach could work across different manufacturing inspection problems.

"I am excited about what this research could mean for manufacturing and aerospace companies in and beyond our region."

The team tested the approach on several manufacturing inspection problems, showing how a pretrained AI model can be adapted using only a modest number of images. The study also provides practical guidance on how to apply and evaluate the method, including its strengths and limitations.

Fadel Mounir Megahed et al., Adapting OpenAI's CLIP model for few-shot image inspection in manufacturing quality control: An expository case study with multiple application examples, Journal of Quality Technology (2026). DOI: 10.1080/00224065.2026.2640171 Key concepts Trustworthy machine learning AI-enabled digital twins Machine learning methodologies Provided by University of Dayton Who's behind this story? Gaby Clark MA in English, copy editor since 2021 with experience in higher education and health content. Dedicated to trustworthy science news. Full profile →

Bachelor's in mathematical biology, Master's in creative writing. Well-traveled with unique perspectives on science and language. Full profile →