AI Inspection Catches Defects Before They Become Recalls

Instrumental’s Monitor identifies anomalies and learns whether they’re defects.

Instrumental, a company that produces optical detection systems for manufacturing, has added Monitor, a capability designed to automatically recognize and triage abnormal units as they are being built on an assembly line. Monitor was developed to detect differences and then learn whether those differences are actually defects.

Monitor reviews images of units from the assembly line and automatically identifies things that are different from what it expects to see. Instrumental noted that the underlying technology in Monitor was designed different from traditional computer-vision based approaches, so Monitor is able to work without the need for significant training, golden samples, or rules.

“The key point with this system is we’re building a brain,” Anna-Katrina Shedletsky, co-founder and CEO of Instrumental, told Design News. “We’ve created a way for manufacturers to tell us the difference between different and defective.”

 

Instrumental, Monitor, inspection, optical systems
The images on the top row left reveal missing screws. Once the Monitor system learns these are defects, the system identifies them as defective instantly. Image courtesy of Instrumental.

 

 

According to Shedletsky, the hardware development process for today’s high-tech products is very low-tech. Teams often spend weeks or months overseas on the manufacturing floor, resolving issues one-by-one with incomplete data. Preventable issues can and regularly do slip through the cracks, resulting in expensive and sometimes highly publicized product delays or dangerous failures out in the field.

When the Monitor system spots an anomaly, it goes through processing to see if the oddity is benign or whether it’s an indication of a defect. Once the system learns whether what it’s seeing is an insignificant difference or a true defect, it remembers what it has learned and will make future determination at the edge rather than going through time-consuming processing. “The first time it sees an anomaly it takes a few minutes, but once its learns the difference between different and defective, it pushes to the edge and the data won’t have to go as far,” said Shedletsky.

The idea of having the system learn is to speed the process of identifying issues in manufacturing and reporting them instantly. “In order for this system to be good for mass production, it has to respond in real time,” said Shedletsky. “You have to put a device in a box and the box should be able to tell if its correct or wrong in seconds, not minutes. You have to be aware quickly so you can do something about it.”

Getting the Screws Right

Shedletsky explained that during the development of Monitor, the system was viewing the insertion of a specific screw when it noticed there was trouble with the screwdriver’s calibration. Quite a few screws didn’t get screwed in all the way.

“In the field that would be a problem. Yet Monitor caught the defect in time to avoid a product recall,” said Shedletsky. “A couple units got caught with screws that were not all the way in. It would be hard to detect without an optical inspection system. A well-train observer might have caught it, but maybe not.”

Subtle differences can slip through human inspection, but once out in the field, these differences can be a problem. “These are units that function, but they’re the walking wounded,” said Shedletsky. “They will fail at some point. Monitor provides a performance test station that doesn’t otherwise exist on the line.”

 

 

As the system automatically identifies and aggregates these abnormal units, engineers can see the product and process quality at a glance. The inspection stations are typically deployed at multiple places on an assembly line, giving engineers visibility into the entire assembly process.

Shedletsky will present the program, “How AI Is Changing Manufacturing for the Better,” at the Design and Manufacturing Pacific show in Anaheim, Ca, on February 8, 2018.

 

Rob Spiegel has covered automation and control for 17 years, 15 of them for Design News. Other topics he has covered include supply chain technology, alternative energy, and cyber security. For 10 years, he was owner and publisher of the food magazine Chile Pepper.

Comments (1)

Please log in or register to post comments.
By submitting this form, you accept the Mollom privacy policy.
  • Oldest First
  • Newest First
Loading Comments...