Man vs Machine: Who Reigns Supreme for Quality Inspection?

The human eye is a particularly complex instrument. Working in combination with the brain, it allows us to perceive our surroundings. We are able to recognize objects in a split second, even when their exact shape varies. We use perspective to analyze our environment and have a wide field of vision capable of focusing very sharply on particular areas. These abilities have gradually evolved over millennia as humans continually adapted to many different stimuli and environments in order to survive.

But our visual perception also has important limitations. For starters, we only have two eyes. They are too slow to see fast-moving objects in detail and they are sensitive to only a limited portion of the light spectrum. Glare and reflection also impede our ability to focus on certain properties of an object, such as size or color, for a long time. In addition, we are quite subjective in how we perceive and store images.

High-speed vision inspection cameras integrated into high-volume assembly line.

No wonder, then, that the human eye cannot perform accurate measurements and is therefore not the ideal instrument to verify product quality.

Machine Vision: More Reliable and Accurate Than the Eye

Machine vision, or imaging-based automatic inspection and analysis, has everything it takes to surpass the human eye when it comes to accurate and reliable product inspection and it is often combined with different technologies. Here are two examples of how a machine vision system can be implemented to ensure the highest quality in production environments.

Example 1: Products moving at a rate of 20 per second have to be thoroughly inspected. The goal is to detect errors with an accuracy of 0.02 square millimeters.

Given the fast pace and the need for long-term reliability, visual inspection with the naked eye is not an option in this scenario. If attempted, nonetheless, such an experiment would involve a whole team of people, which would go against the objectivity of the inspection. Machine vision is the solution: six cameras observe the fast-moving products using very short shutter speeds and brief and polarized light exposure (strobe). This creates sharp images in which defects are perfectly visible. Special software then searches all defects within 50 milliseconds and it can do that 24 hours a day (by using a real-time operating system or FPGA). The result: a system that is superior in every respect to human inspection.

Example 2: Defects of a few microns (μm) must be detected in an area of 20 millimeters (mm) on objects that pass by at a speed of 5 meters (m) per second.

If inspection with the human eye was used in this instance, it would require a single person capable of seeing defects of a few microns on a 2-m surface while the product moves by at a speed of 18 km/h (i.e., one every second). So, here too, inspection with the naked eye is not realistic. The only option is a combination of hightech machine vision such as 8k line-scan technology combined with fast lenses, LED line lights, and super-high-speed, “on-the-fly” image-processing software.

The Human Eye vs Self-Learning Software

There is one thing for which the human eye is sublime: spotting anomalies or defects on products. We see a defect on a product right away if it is big enough. Even though we have never seen the defect before, we immediately notice a scratch on a small object or a torn seam on clothing.

As a rule, we unconsciously perceive the anomaly when we pick up a product, turning it and observing the reflections. This, combined with our exceptional interpretative abilities, makes the human eye almost unbeatable.

In recent years, however, machine vision technology has evolved considerably and now matches our interpretative abilities in many cases. Using complex, self-learning vision algorithms, the current technology is now capable of processing images in the same way the human brain would perform the task. If supplied with a picture library with additional information, intelligent software can teach itself where to find the errors without anyone having to program a single line of code. This additional information can indicate which products are good and which are bad or show where defects are located. Even products with a changed design can be recognized quickly.