This is an implied "either-or" that has arguably now been bridged by a few manufacturers…but remains from the past a distinction to many. Image based perceived as Low End, and Algorithm based as High End.

Image based systems

The old term "image based system"comes from the fact that those machines compare images of entire components.

If the source image deviates too much from the target, it flags a defect. This deviation tolerance is specified as a brightness percentage. These systems are also known as correlation based systems.

Algorithm based systems

Also known as Threshold or Vector systems. these machines distinguish Pass and Fail byapproval against targetpixel data, tolerances and algorithms.

There are two different ways toset the target data:

1. Manually by numericaldata and threshold inputs
2. Automatically by measuring the values from real components.

Correlation based systems are in general less accurate then algorithm based systems. Programming is usually very simple and machines can be marketed as "self learning."

Algorithm based systems are more accurate and more flexible when PCB designs are not perfect.

Due to misunderstanding, an algorithm based system that uses real image datafor programming, is sometimes wrongly interpreted as a correlation based system.

At the risk of oversimplification, AOI machines that can be programmed by using data from real components will set up faster and more easily. For a small EMS company, this is usually a most critical success factor.

The advantage of machines with manualinput of target and tolerance datais that all values can be set very accurately. For High Volume production where programming time is less critical and for special measurement purposes, this can be a requirement.

Algorithm based systems utilize contrast location mechanisms that look for variations in light and dark grays scales to identify the component and the surrounding area. From these location techniques all other inspections are applied such as presence/absence, solder joint, polarity, bridge, and package identification.

Image based systems learn the variables of a group of images and apply those learned variables against all future component inspections.

How good either technique is really depends upon the AOI vendor but the general assumptions and feedback in the field is that algorithm based systems provide higher levels of inspection coverage, lower false fails, but can take a little longer for initial program development. The image based systems are often perceived as easier to get programs up and running but as stated may not provide the best inspection coverage.

As a small EMS company it depends on what you want to achieve and where you see yourself in a few years. If you are only concerned in quick presence/absence analysis an image based system is probably sufficient for your needs. But if you expect to grow and really need good defect coverage especially around solder joint analysis and location an algorithm based system is probably the best solution. It should also be noted that most algorithm based systems have some level or learning capability that allows future programs to be developed much faster after the first few initial programs.

1 & 2 – The main difference between algorithm and image based AOI is the information used by the system to qualify the defects. Image based AOI are working with an “historical” database which contains all the pictures of good and the defective components. This method is very sensitive to the way we store pictures and especially if you do misjudgement of defects.

It is easy when you implement but very unstable in production environment while you are always storing images. Algorithm based AOI are using tools to analyze images and find defects. This is not related to the current image but refers to logical characteristics of the image (like the edge of the components).

Some advanced AOI are using algorithm such as “pattern match” to identify the geometry of the component itself. These AOI are much more stable during the production as they refer to properties of the component and not to a few “stored” images which depend on component and PCB variations.

3 – As a small EMS company, even if the short term benefits of image based AOI can be attractive (easy to implement), I would recommend to plan the use of AOI for mid or long term advantages (stability). You will gain when the AOI will be fully implemented with a unique stable library of algorithms for all the variety of your products.