How is Industrial Machine Vision Different from Academic Computer Vision?

Industrial Machine Vision Testing in the Lab

How is Industrial Machine Vision Different from Academic Computer Vision?

  • Rob Couture

Industrial machine vision is a whole different ball game from the academic computer vision you read about in your textbooks.

 

Environmental factors must be taken into account

 

In the harsh reality of the manufacturing environment, algorithms must be adaptable and robust enough to handle natural variations and fringe cases over an extended period of time. We're talking about dealing with vibration, temperature variability from 10 C - 45 C (50 F - 100 F), dust, EM noise, changes in surface texture of the product, and positioning of the product that becomes more and more lax as fixtures start to wear. It's not for the faint of heart.

 

Limited data availability

 

To make matters worse, you can't always rely on having a ton of data to work with when creating an industrial vision system. Getting good samples isn't the problem, it's having the end user collect and hold examples of non-conforming products that can be a challenge. And once the images are collected, the data needs to be manually sorted to develop and test the inspection algorithms. It's a real headache.

 

Problems tend to be well defined

 

But hey, at least the problems tend to be well defined. We're talking about detecting the presence/absence of all truffles in a chocolate box, measuring dimensions of a spark plug, reading the lot/date code on the bottom of baby formula, or making sure production isn't putting peanut butter ice cream in a vanilla labeled container. These are practical problems that need practical solutions.

 

Interdisciplinary skillset required

 

And don't think you can just get away with being a computer science or math whiz when it comes to industrial machine vision. You need an interdisciplinary skillset that encompasses optics, electrical/computer engineering, mathematics, computer science, and manufacturing process/design. We're not playing around here.

Speed and accuracy matters

And let's not forget about speed and accuracy. In the lab, algorithms that take a few seconds to run are no big deal, but out on the production floor, you don't have that luxury. Consider a system where the inspection is 4 parts per second, which is 240 parts per minute or 14,400 parts per hour. If the industrial system rejects 1%, half of which is a false reject, you're scrapping 72 parts per hour. Let that sink in…. Assuming two shifts, 8 hours each, 5,760 parts of good product are thrown out in a week. 

In the food and beverage industry, that product goes straight to the landfill. They are often not allowed to recycle it for animal feed.

So let's break it down.  4 parts a second, so got 250 milliseconds between the leading edge of the first product and the leading edge of the second product to get everything done.  In that time span the image needs to be acquired, get the data off the sensor, transfer it over Ethernet, decode it and load it into memory, and post the results to the control system.  That means your algorithm may have about 150 milliseconds to execute.  That doesn't leave a lot of time for heavy image processing on an edge device sitting on the manufacturing floor.

 

It is better to light it right, than write around it!

 

Up to 80% of an industrial vision project's success is in lighting and optical design. You can't just write around hot spots, uneven illumination, motion blur, optical distortions, overhead sodium lamps, or dirt with some magical filters or cutting-edge algorithms in the software. The purpose of lighting in machine vision is to provide the necessary illumination to capture high-quality images that are suitable for the algorithms to run efficiently and with repeatable results during production.  To achieve this, lighting must be strategically designed and optimized to enhance the contrast of the features that are of interest, suppress the contrast of everything that is not important, and minimize external influences.

 

How Can We Help?

Industrial Machine Vision systems help to ensure the quality of your product and can be designed so that they are maintenance friendly.

From retrofits to turnkey solutions, extracting meaningful data or making an existing system more reliable; we are here to help you meet The Standard.


Fill Out a Project Brief