Monday, March 31, 2008

That joke isn't funny anymore

Machine-vision-system integrators are more than consultants--they actually have to know something about integration

There’s an old joke about consultants that reads something like, "Ask a consultant what time it is and they will borrow your watch and tell you the time." When I first heard the joke, I thought it was quite funny, especially since I have personally been asked questions about how to design machine-vision systems for various applications.

Unfortunately, while the joke may ring true to many considering deploying a machinevision system, it is not as amusing when one considers the disparate disciplines of optics, illumination, image processing, computer science, and mechanical engineering needed to develop such an application. Indeed, this is the main reason that the development of these systems is so challenging and at the same time so frustrating.

Since very few universities and colleges bundle these subjects into a single degree, it is difficult for developers to hire technical staff. Instead, they must rely on pooled knowledge from those with experiences in individual subjects.

At the outset, system development may seem easy. Light the product to be inspected, capture an image of it, and then trigger a reject mechanism should the part fail the inspection. When looked at from 30,000 ft, system development may seem trivial and—to those in management—inexpensive.

When examined from a microscopic level, however, the problem of designing a system becomes more complex. Just choosing a lens to image the subject may result in hours of NRE (nonrecurring engineering) time. Deciding on the type of lens required, optical mount, focal length, and resolution may appear easy, but, because of the lack of detailed specifications offered by many suppliers, an evaluation of a number of lenses may be required—a process that could take days.

Additional complications
This situation is further compounded by the fact that coupling a lens to several different cameras may result in very different images being obtained. As David Dechow, president of Aptúra Machine Vision Systems (Lansing, MI, USA; pointed out in our February Webcast, the different formats of imagers employed by camera vendors may result in varying levels of illumination rolloff or vignetting.

With the move to larger-format imagers, this problem is further exacerbated. Worse, if the digital camera you select does not have dead-pixel correction or flat-field correction, the resulting image may not be usable. As can be seen, simply selecting the correct optics and cameras is challenging. But system integrators face other tasks relating to lighting, choosing the correct software package and operating system, and how these are integrated into an industrial automation system controlled by PLCs. While college textbooks may help students understand the basic principles of all of these subjects, deploying machine-vision systems requires more. Luckily, most integrators are fully aware of this situation.

For those considering deploying a machine-vision system, a visit to an engineering facility may be most valuable. If you are led into a conference room and given a sales pitch, beware! Instead, ask for a tour of the engineering department, where you should expect to see workbenches strewn with optics, lighting, cameras, and half-complete computers. If people there appear busy and frustrated, take this as a very positive sign.

Often, however, potential customers visiting these facilities arrive unprepared, handing the company management a few questions and a part that they would like automatically inspected. Hence, the integrator must probe more deeply into exactly what needs to be inspected, the nature of the production line, the type of lighting used in the facility, and the previously installed computer systems— essentially borrowing the potential customer’s “watch” to ascertain the time. In such situations, having your “watch borrowed” is obviously quite a good idea, since it will only lead to the development of a more effective and efficient vision system.

1 comment:

Jack Vaughan said...

Well spoken! As you rightly describe it, such engineering is near to art. May I ask a question? In the news this week is a story about Google and its latest Image search software, "PageRank for Product Image Search." Apparently these Googlista suggest that despite decades of effort, image analysis remains a largely unsolved problem in computer science. Do you agree or disagree that that is the case. Thanks for your time. Jack Vaughan