Quality control using machine vision: what you need to know

At first glance, the wrinkled seat of a car is not a major manufacturing problem. Yet the automaker Ford must inspect every seat on every vehicle for this possible manufacturing defect. It is in order to be able to automate this process that Ford has been using a Computer Vision system since 2019.

But computer Vision is only one of the possible ways of inserting a machine vision system on a production line.

The term “industrial vision” covers all the technological alternatives to visual inspection by humans (computer systems, compact or on-board systems, laser, etc.). Their goal is always the same: to increase the efficiency of quality control to save time and costs.

You should know that off- the-shelf solutions already exist for the implementation of a machine vision solution. But there are some special cases: production lines where too many tolerated exceptions accumulate, high variability in the same part, or even modular fault tolerance.

In those situations where a customized solution must be set up, when is it better to ' use a model of computer Vision? What are the respective advantages of possible machine vision solutions? Check for more here on https://www.dzoptics.com/en/machine-vision/.

Traditional image processing methods applied to quality control

Of automated visual inspection systems were installed on production lines, storage and distribution of large industrial since the 1990s. This type of quality control uses so-called “ deterministic ” algorithms (ie. For which the decision rules are well defined and easily applicable). These excel in the quantitative measurement of a structured scene thanks to their processing speed, precision and reproducibility.

Coupled with the right camera and the right optical resolution, a deterministic inspection system can identify defects invisible to the naked eye. It is a more reliable and faster method than humans. These deterministic solutions are ideal for control involving gauging and precision measurement.

On a production line, a deterministic industrial vision system can perform quality control of hundreds or even thousands of objects per minute reliably and repeatedly. The defects thus controlled may be too small to be visible to the naked eye if the camera allows this precision. This performance goes way beyond what a human being can do.

However, human visual inspection has always prevailed:

in cases requiring learning by example,

and when it is necessary to know how to recognize and tolerate an acceptable deviation during quality control.

The advantages of Computer Vision (Image Recognition) for quality control

Human beings excel when it comes to distinguishing between cosmetic and functional defects. Likewise, it is able to recognize which cosmetic defects may be perceived by the consumer as affecting the perceived quality.

Once trained on a set of data corresponding to its actual conditions of use, a Computer Vision algorithm can reproduce this human adaptability.

Indeed, Deep Learning technologies use artificial neural networks to reproduce human behavior. This allows them to recognize anomalies and defects while tolerating natural variations in complex patterns. Computer Vision methods therefore combine the adaptability of human vision inspection with the speed and robustness of a traditional image processing system.

Compared to a traditional image processing solution, a deep learning solution is more efficient in cases where variations and exceptions are tolerated.

Once the choice has been made between the development of a deterministic algorithm and a Deep Learning solution, it remains to install equipment that will allow the recovery of quality data.

In both cases, the choice of good equipment is essential. A complete machine vision system built to measure for the production line environment is a prerequisite for good data recovery and use.

All the components of a machine vision system must be chosen according to the environment and the object to be inspected.