By Cesar Peña
Selecting a machine vision platform should be pretty straightforward. Determine the resolution of the camera needed and the lighting type, select a lens, make sure the processor is fast enough for the inspection rate, and verify the camera will communicate with the rest of your system. Write up your requirements, do a Google search for machine vision systems, send your requirements out and select the best proposal.
Obviously, these steps are anything but straightforward. From a technical perspective, they really are just the beginning. As a starting point, and trying not to be self-serving since I work for a system integrator, the most important step is to find a company that can support your needs. This could be a system integrator, a distributor or the product manufacturer.
You also need to determine the technical requirements of your inspection. There is no way to cover all that in a one-page column, but the following list should be a good starting point:
- Assign a vision champion within your organization to take ownership of the vision system, both before and, especially, after it is installed.
- Clearly define acceptance and rejection criteria. In all but the simplest cases, you can't define a "master" part and have the system reject anything that doesn't look like that part. Not every part will be identical; reasonable tolerances should be set. If your tolerances are too tight, not only is your vision application going to be more challenging, you might reject acceptable parts.
- Calculate an ROI to get you in the ballpark for your budget. Sometimes this calculation will lead you to realize that a vision system might not make sense. Or that it makes so much sense that you're wasting money by not acting now.
- Consider camera type. How uniform must the lighting be in the inspection area? It is difficult to illuminate a large area uniformly. By using a line-scan camera for a large-area application such as for glass pane inspection or web inspection, it is possible to uniformly illuminate just a line on the target using a line-scan illuminator, and get incredible resolution. The drawback is that line-scan cameras are typically more challenging to install and set up. For smaller areas, or areas where lighting isn't the biggest concern, area cameras usually make the most sense.
- Color or gray scale? Color cameras are great for inspecting the color of parts. They aren't good at precision measurements because the pixels in a color camera are arranged in a checkerboard pattern, and the resolution for any given color is only half the specified resolution of the camera. In a gray-scale camera, all pixels are equivalent. So a precision measurement made with a 1280x960 camera will be equivalent to a 640x480 pixel measurement made by a gray-scale camera with the same field of view.
- Choose between a smart camera and a PC-based platform. Generally speaking, if the quantity of cameras needed per vision system is less than five, and the application is not very complex, a smart camera is a good choice. A smart camera is basically a camera (optics), a processor and software housed in a single unit. For applications that have multiple cameras, very high performance requirements or very specialized vision algorithm needs, a PC-based solution might be the right approach. The line between smart camera and PC-based vision price/performance continues to blur, so the final answer will depend on individual circumstances.
- When selecting the specific hardware, consider the overall install base of the manufacturer, the experience level of your internal and external resources, and the overall support infrastructure available. For example, if you're building a machine to ship to Asia, can the vision system be supported by local expertise?
Vision system capabilities and ease-of-use have increased significantly over the past several years, and for many straightforward applications it might be tempting to internally develop your solution. Sometimes this makes sense. However, from my experience, choosing the right experts to partner with can be your absolute best decision when implementing a vision system.
Cesar Peña is senior project director at system integrator DMC (www.dmcinfo.com) in Chicago.