Advice: Machine vision applications…subscribers tell all

By Control Engineering Staff April 24, 2007

Oak Brook, IL —A recent survey of Control Engineering subscribers uncovered advice about selection and application of machine vision technologies. A table of showing percentages of machine vision applications among respondents follows below the tips. The May 2007 North American print edition of Control Engineering will include more about the survey, along with leading vendors’ products in: ” Machine vision: now is the time .”

Control Engineering subscribers are asked monthly to provide input about a particular product area. For May 2007, it was machine vision. Among input provided was considerable application advice. A sampling of advice from machine vision product research survey respondents follows, without attribution:

Ability to effectively light the camera area and ability to maintain by our plant personnel has been the biggest challenges to greater acceptance of vision systems. We have had great success with the products we have used and feel the sensors are capable of benefiting us in some applications.

Ability to turn a picture into data for further processing is critical. My applications are mostly for raw materials handling (mining) and for processing evaluations, and some automated guided vehicle applications. Most involve dust and harsh environments. The equipment must be able to take clear pictures and the software must be able to convert those to operating data for a 24/7 continuous processing environment. Allowing one oversize rock to hit a crusher can mean hundreds of thousands of dollars lost, even if no equipment damage results, and more if repairs are added to the lost operating time.

Primary applications for machine vision
Survey respondents were asked
to select all that apply.
Robot motion control 78%
Target identification 43%
Target location 43%
Motion coordination/guidance 38%
Other 3%
Moving vehicle automation 33%
Unmanned vehicle control 19%
Driver aids 11%
Other 6%
Assembly guidance 79%
Pick and place 49%
Mechanical assembly 37%
Failed unit removal 35%
Other 5%
Product inspection 82%
Product dimensional measurement 64%
Product color 35%
Defects 4%
Other 7%
Assembly inspection 69%
Component presence/absence 58%
Component placement 46%
Other 3%
Print quality/readability 52%
Readability 35%
Correct information 32%
Character form 27%
Character placement 26%
Other 1%
Packaging inspection 27%
Packaging fill level 23%
Packaging seal integrity 18%
Other 2%
Barcode 86%
One dimensional 76%
Two dimensional 40%
Character 75%
Feature identification 52%
OCR reading 46%
Other 5%
Source: Control Engineering and Reed Research Group May 2007 Product Research on Machine Vision

Always work with a knowledgeable reseller . Explain your needs and look for suggestions on lighting and lenses. This can save a tremendous amount of research and development time on your own.

As a potential customer, I need to be convinced that the equipment that I purchase is going to accomplish the detection and rejection requirements that are necessary, justify the capital outlay required. There also has to be a performance guarantee in the contract, specifically stating machine capabilities, production throughput, and detection—rejection rates of product and or products being inspected.

As automation increases in the speed and number of products processed per second , we need an integrated machine vision tool that can give us the frames per second (FPS) we need to diagnose problems or increase efficiencies without paying >$150K per camera.

Choosing the correct lighting and lenses is the most complicated part of any machine vision application.

Choosing the correct product for the customer’s needs must be the first priority or anything that follows will cause the customer to be dissatisfied. Getting the customer to understand the limitations of the vision product is the second most important priority. Most of the problems with vision products results from a failure to follow those two rules. There is a lot of poor information out there.

Don’t settle for the cheapest option because it will just cost you more [time and money] in the future.

Do not be reluctant to ask questions about the part you do not understand. We are all going through a constant learning curve.

Focus on ease of use .

In the last few years, the market lines have blurred between leading vision offerings. What was once a cut-and-dry decision between vision systems based on price, performance, and camera count is no longer the case with multiple vendors now offering multi-camera, single-controller options and low-cost, purpose-built units designed for specific tasks. We’ve seen our options explode and while it has made system selection more difficult, it has made is easier for us to find the right system for the task at hand.

It is very important to correctly define the trigger mechanisms for high speed applications to avoid de-synchronizations due to lags with the machinery trigger.

It pays off to start with one project instead of going into a “shotgun” approach, trying to cover all projects or the whole plant. Find an application that would gain a large benefit at the lowest cost to “get your feet wet” and try it out

Look for experience . Make sure the machine vision sensor vendor has done your application before.

Lighting is critical ; the right camera characteristics must match the job. This matching process can be very difficult, depending on the application. Keeping the lighting clean and available is a major cost of these systems.

Linux is best for front-end vision oriented computer systems.

Look for connectivity and integration abilities into an overall automation platform. This greatly increases the marketability of machine vision systems.

Machine vision shows all defects of our final product and improves our quality assurance.

Primary use in robotics vision systems has advanced our product offerings greatly. Small size and price constraints dictate our selection process.

Robust and rugged mounts are a good idea. Use the I/O connections given to ensure system operational readiness and inspection functionality. One way to show a bad implementation of a vision system is to allow the machine to run when the camera is offline!

The success of a vision project hinges on how well the vendor/ integrator understands the inspection requirements in a production environment. Making an inspection work on a bench top under controlled conditions can give you a false sense of security that the same inspections will work under production conditions.

Use of vision systems has eliminated costly defects to our customers and is a tool that, in hindsight, should have been implemented years ago .

Vision systems reduce the amount of manufacturing and packaging errors and reduce losses because they are more accurate than the human eyes and are not distracted or missing from work. Once trained, no retraining is required as regular personnel and backup staff.

We would like to see smart vision sensors with the bandwidth and intelligence to close the motion control loop for high-speed servo-driven devices to the precision and stability of present “contact devices” like shaft encoders.

When selecting a machine vision application be sure that it will work in the environment that it is placed.

When setting up your first machine-vision system, just remember to take extra time to think through all the process points you want included and then plan for future growth before you decide on the system!

When working with vision, you have to use enough equipment to fully do the job . If two cameras just about do the job, but three or four cameras do the job 100%, go the extra mile or you’ll be unhappy.

With all the application experiences and successes that have been accumulated over the years, machine-vision is still not readily accepted as the best solution to difficult problems, even though it is intuitively obvious that it is the simplest, most flexible, most tolerant, and ultimately the most reliable means of solving the problem .

For related information, also see a primer of machine vision tricks: ” Measure Up with Machine Vision .”

Mark T. Hoske , editor in chief,
Control Engineering Weekly News
Subscribe to our eNewsetters (scroll down to select among 12).