Reduce Manual Defect Detection by Repurposing Images

Whether it is an automotive manufacturer inspecting paint applications or a medical device manufacturer ensuring equipment alignment during assembly, today’s industrial environments use a variety of images for many different purposes. Why not put these images to work in defect detection and classification?

Repurposing existing images can help manufacturers reduce waste, improve yield, and decrease the burden of manual defect classification. It can also increase the overall return on investment of equipment when that equipment is not processing defective material. This process is also described in detail in our recently published white paper, Faster, More Accurate Defect Classification Using Machine Vision.

At Intel, we use a variety of defect classification methods, such as applying machine vision and deep learning to our image processing. Repurposing images for gross defect identification at Intel started as a challenge to reduce the number of people performing manual inspection. Using images that were originally captured for other purposes, we are now able to both decrease the amount of manual defect detection and classification, and also conduct defect classification in new areas. For example, unit staining occurs when there are problems with the epoxy process, and this stain is visible in images that we capture to monitor equipment component health. Where we previously inspected all materials manually, we can now use these images for gross defect detection.

Using OpenCV, we separate only the materials that require manual inspection and classification for that process―greatly reducing the volume of items inspected manually without compromising product quality. We are essentially maximizing the use of computers to segregate only marginal units for manual inspection.

Focusing Employees on Other Tasks

Inspection is a highly specialized job, and currently the team reviews millions of images per week. With machine vision and machine learning, we have been able to reassign some employees to other areas in the factory, and in some instances, retrain employees for other tasks inside the factory, like machine operation. This is possible due to the reduced demand for manual inspection.

In addition, we can now identify defects in some use cases in the assembly process where we previously did not conduct any defect identification at all. Furthermore, we have been able to achieve this without purchasing new equipment or investing in major upgrades to existing equipment.

Repurposing images in our assembly and test factories is one of many solutions Intel uses for defect detection and classification. To learn about our solution for highly sensitive automatic defect classification (ADC) built on Intel® Xeon® processors, read the IT@Intel white paper: Faster, More Accurate Defect Classification Using Machine Vision.

Published on Categories Machine Learning, ManufacturingTags , , , ,
Duncan Lee

About Duncan Lee

Duncan Lee is a Principal Engineer for Intel Manufacturing IT with 25 years of IT experience. He currently focuses on manufacturing efficiency and quality through the usage industry 4.0 principles, integrating various AI/Analytics/IT/automation solution stack to improve the efficiency and productivity of Intel’s Assembly and Test manufacturing. Recent successful projects include driving factory machine uptime through early defect signals detection and analysis to facilitate predictive preventive maintenance. In addition, Duncan works on projects involving the usage of vision analysis and AI to identify potential problems or defects in units. Duncan is a member of Intel’s Systems and Artificial Intelligence Patent Approval Committee. He received his Master of Science in Electrical Engineering from the Wichita State University in Kansas.