Types of vision recognition

Phil,

In the considerations you make for choosing an AOI tool, you should check what a vendor delivers today and what will it be able to deliver in the future. As the role of AOI in the SMT production line evolves from "detection" to "prevention" and from a "single AOI tool in a line" to "multiple AOI tools", you should consider the ability of a vendor to support these new concepts - as those are the real major revolutions for AOI in our industry.

You may also ask and consider other important questions such as: Is your vendor capable of providing a solution that not only detects faults but analyzes AOI results and can provide alerts on a process drift or trend? Does the vendor provide a backbone solution to handle multiple AOI tools in a line in an efficient, effective way? Can your vendor support your global operation? Is AOI the vendor's main business or is it a minor activity? Is your vendor stable enough to withstand the difficult economic times of the industry?

Back to your specific question:

Your question can be divided into two subjects - image acquisition (what you referred to as "greyscale") and image processing (what you referred to as "algorithyms").

Image acquisition:

AOI systems can be easily divided into two image acquisition groups: systems which employ a single top-view camera (the majority) and systems which use multiple cameras (the minority). In general, based on an ultimate truth of AOI which states that "you can not avoid what you can not see", systems that use multiple cameras can deliver a higher degree of performance simply because they "see" more. These systems are more complex, can detect post-solder defects unlike 2D and are usually referred to as the "high end".

Almost all the systems today (both with a single top camera and with multiple cameras) use area CCD cameras in the seen light wave length to acquire images. You may find very few vendors who use a line-CCD or the special TDI (Time Delayed Integration) technology. One vendor is inspecting by Infra Red imaging, which may be good for special needs. The reason people choose AOI systems that use area CCD cameras is simply because it is a known and proof technology that can "do the job" and can serve current and future market requirements.

Image processing:

Here you may find more variety. Again you can draw a line between lower end solutions and higher end. The lower end solutions use an image comparison which compares the acquired image to a "golden" image (perfect image of the board which is stored in the system data base) and process the differences. Higher end systems use much more sophisticated methodology to analyze the acquired images. Such systems are based on algorithms which extract parameters (or "descriptors") from the images; the extracted parameters represent or "model" the elements in the image. The parameters then can be processed and accumulated over time into a statistical modeling database for an enhanced, more robust and better performing image processing.

To summarize:

You may view systems which use CCD cameras as what you called "norm" and this is a safe choice. Systems using statistical-based image processing algorithms can deliver higher performance. Their algorithms and setup are more flexible and they can be adapted to changing requirements. If you want as safe as possible capital expenditure, look for a platform that can offer you an upgrade path. If you choose a high-end solution you may safely assume that it will serve today and tomorrow market requirements better than the lower end solutions.

This message was posted the OnBoard Forums

reply »