Manufacturing AI Automation AI AI Use Case Machine learning Azure

Published:
Est. reading time: 4 minutes
Author: Steph Locke

With the pace of output on machines ever increasing, quality control becomes a lot tougher. Using AI to detect defective products sooner can help scale your quality processes and avoid significant stops. The regularly viral tomato sorter types of videos show that basic sorting of raw materials has been around for a while and is pretty decent. But what about more complex goods like electricals or where the defect is a bad print?

Improve operations

Learn from our business case studies

Businesses working smarter with us

Steph Locke

Technologist and consultant with a track record of delivering transformation of businesses into data science and AI companies.

More

With the pace of output on machines ever increasing, quality control becomes a lot tougher. Using AI to detect defective products sooner can help scale your quality processes and avoid significant stops.

The regularly viral tomato sorter types of videos show that basic sorting of raw materials has been around for a while and is pretty decent. But what about more complex goods like electricals or where the defect is a bad print?

In these sorts of circumstances we need to rely on computer vision (CV) – the use of cameras and artifical intelligence – to perform the task of “seeing and interpreting”. The value of using CV is that a computer can perform calculations many times faster than a human can, and can run in environments that may be unsafe for humans.

Continue reading to understand more about computer vision and how you can use it for defect detection.

What is computer vision (CV)?

Computer vision machine learning models are complex algorithms designed to recognise certain patterns in images or video streams.

We typically build CV solutions by taking images of the things we want to be able to identify and training a model of things present in the image that are most predictive of the thing we’re trying to recognise.

This doesn’t really learn like humans do and looks at different levels of pixel groups to perform an optimisation process. As a result, when we build CV solutions we need to make sure we don’t do things like show good products in boxes and bad products in a bin because the CV solution will basically detect bins first as the easiest way to tell if a product is bad.

What sort of defects can I detect with CV?

CV for defect detection can be used across most sectors of manufacturing including:

  • automotive
  • electronics
  • materials
  • metals
  • food and beverages
  • pharmaceuticals

For most of the industries you can identify visual non-conformities like misprints, incorrect colours, dents, scratches, and more.

NEC, who make a visual inspection system, produced this handy diagram explaining the areas within the different sectors that can benefit from computer vision.

How do I get started with CV?

There are advances in machines with computer vision integrated solutions but new hardware is expensive!

I recommend starting with defect detection solutions that use non-invasive cameras and report problems to humans to pick the defective products off the line. This allows you test the concept cheaply, apply it in a targeted fashion, and use relatively commodity hardware.

I mentioned NEC above as a solution provider in this area and there many others in this growing field, including landing.ai a startup founded by Dr. Andrew Ng, one of the leading advocates for AI.

Can I make my own computer vision systems?

Yes! You could start with a kit as cheap as a small webcam and Raspberry Pi to build your own solution.

There are open source (ie free to use) frameworks for computer vision that would allow you to entirely train your own defect detection models and you can use relatively low-end hardware for actually hosting the solution close to your machines.

These sort of solutions are possible, so much that people have even put these Raspberry Pi implementations on drones for defect detection on the move.

It can definitely be done in a super cheap way, but that being said, doing everything from scratch is pretty daunting.

Starting to be rolled out by Microsoft is their Azure Percept device. Azure Percept can be used to process video data and perform actions based on the results of a machine learning model, including the use of their off-the-shelf and custom computer vision models. These are lower code solutions to build a pilot project than the full DIY solution but give you a strong foundation from a major software vendor.

Aside of defect detection, what else can I use computer vision for?

Visual recognition tools can be used to aid robots and machines with things like label placement, package inspection and sorting.

Computer vision can also be used to track health and safety on the shop floor, from identifying individuals who are not wearing safety equipment to spotting contamination risks, the system may even intervene, blocking access when noncompliance is identified, or halting dangerous machinery to prevent injuries.

Visual recognition systems can be used for text and barcode reading which has several applications in manufacturing, such as checking the right parts are being used or sorting and tracking item through the factory. Computer vision can also be used to guide operators through complicated processes such as assembling an item, by checking codes on parts and interacting with the operator through gestures.

Excerpt from our larger AI in manufacturing article

Improve operations

Learn from our business case studies

Businesses working smarter with us