Back to Basics: Selecting sensors, machine vision
Ask these questions as a starting point when considering a detection, inspection, or measurement solution using sensors, a smart camera, or a more sophisticated machine vision system. See checklist.
Simple is best, and less is more. When I am creating a detection or measurement solution using machine vision or sensors, I constantly ask myself the question: Can this be done with a simpler solution? See checklist, related links, at bottom of article.
As a vision system integrator, my considerations for selection of a system are (in no particular order):
- Is the system capable of solving the problem?
- Hardware and engineering costs
- Simplicity of use and troubleshooting (What are the customer’s capabilities?)
While the first two items may seem obvious, the third is often overlooked. Too often, I believe, system integrators are so focused on solving the problem that they don’t consider what happens after they have left the picture. Making a system as simple as possible to use and troubleshoot helps to minimize the number of warranty issues and gives a customer a much higher level of confidence in a system’s functionality. If you’ve added a bunch of bells and whistles that are cool but aren’t required for core functionality of a system, you’re going to regret putting that bell or whistle in when you get a warranty call because it doesn’t work.
With those general considerations in mind, let’s now consider three levels of technical solution and where each would fit. In order from simplest to most complex:
1) Sensors (proximity, photoeye, depth sensors, etc.)
2) Machine vision—smart camera
3) PC-based machine vision
Sensors can be used to detect the presence or absence of key features of a part, such as a drilled hole or a bracket. To have successful sensor-based detection, generally a part must be very well fixtured (held in precise position while sensing is performed) and the feature must be “sensor friendly” (detectable by a sensor). Sensor technology is advancing quickly, and some tasks that previously required vision can now be performed by sensors. It’s always worth a call to a sensor company to ask “can a sensor detect this?”
Machine vision smart cameras are those that do not require a PC to operate. All the “brains” are contained in the camera (it does all the vision processing), and a PC is only required for setup and programming. Smart cameras come in a very large variety. The simplest are very inexpensive and are sometimes dubbed “smart photoeyes.” Consider these as an option if a feature you are looking for is very simple to detect visually, but not detectable by a simpler sensor.
Smart cameras used to be limited in pixel resolution and vision tools, but now some of the most advanced smart cameras offer up to 5 megapixel (5 MP) resolution and the most advanced vision tools. Engineering and programming time for a smart camera is low compared to a PC-based equivalent single camera system, but hardware cost is similar. Smart cameras also offer a lot of communication options to interface with a variety of equipment.
PC-based machine vision
PC-based machine vision carries the most flexibility and power. The architecture here is a PC that performs the vision processing and one or more cameras that connect to the PC. PCs have the most computational power and speed, and some vision libraries are able to take advantage of the multiple processor cores some PCs have, giving them the ability to handle extremely high-speed inspections. With a PC-based system the integrator typically writes its own GUI (in something like VB or C#) and as a result also has tremendous “logic power.”
This ability to do logic, make advanced decisions, and control vision program flow makes PC-based vision very flexible and powerful, but also more difficult to change or troubleshoot because it is most often the programmer (a single person) who understands the program well enough to do those tasks quickly. When considering cost, the hardware cost of PC-based vs. smart cameras for a single camera system is similar, but engineering time for a PC-based system is usually more unless you have developed a “canned package” that you can deploy easily.
Once you go to two or more cameras you can realize significant hardware savings because multiple cameras can attach to each PC; having multiple smart cameras does not get cheaper per camera since each has everything needed to stand aloneSo which solution is best for your application? Ultimately it comes down to what you need. Obviously, a system must be capable of doing what it needs to do or you don’t have a system at all. If you’re not sure, talk to a vision system integrator or vision and sensor manufacturers as they can provide insights into the capabilities of their systems and where they fit, beyond what a specification sheet might say.
And always ask the question: Can this be done with a simpler solution? See checklist below.
Kevin Ackerman, M.Sc, PEng, is machine vision and robotics specialist with JMP Engineering Inc., London, ON, Canada. JMP is a 2008 Control Engineering System Integrator of the Year.
www.controleng.com/integrators has more integrators with expertise in vision systems, robotics, test/inspection, and other engineering specialties.
Checklist - Sensors or machine vision: Simple is better
When creating a detection or measurement solution using machine vision or sensors, ask:
- Is the system capable of solving the problem?
- What are the hardware and engineering costs?
- What are available capabilities for operation and troubleshooting?
- Will a simpler solution work? Start by considering sensors (proximity, photoelectric, depth sensors, etc.). If that won’t meet application needs, consider smart camera machine vision (no PC required), and if that won’t work, consider PC-based machine vision.