Writing the Ultimate Machine Vision Specification
by Winn Hardin, Contributing Editor - AIA Posted 07/13/2012It’s the line between success and failure. It defines the goals of the machine vision system and what “success” will look like when the vision system succeeds.
And when it comes to customer requests for proposals, it’s also one of the rarest of documents.
“I’d say one out of 10 customers comes to us with a real specification document, and most of the time, the document isn’t viable,” says Robert Eastlund, Vice President of Sales at Graftek Imaging, Inc. (Austin, Texas). “Integrators have to find ways to quickly correct the customer’s misconceptions when it comes to what the machine vision system needs to do to solve their production problem. Many times, coming up with a real specification document takes more time than designing the system.”
Writing a useful specification document is so important that David Dechow, President of Aptura Machine Vision Solutions, LLC (Williamston, Michigan) and an instructor for AIA’s Certified Vision Professional (CVP) certification courses, spends a large chunk of his class discussing how to analyze an application and write an effective specification document. “We help integrators build a checklist of things they need to think about when analyzing the application,” he says.
Specifying “Real” Solutions
When a customer goes looking for a machine vision system, it’s usually because either a new product needs to meet specific customer guidelines, or the plant already has a problem with an existing production line not producing parts to specification. In both cases, getting samples of parts that represent all possible defects as well as variations of acceptable parts can be tricky business.
If a customer is building a new production line, only prototype parts exist. Many times, the manufacturer simply gives a machine vision integrator a list of tolerance requirements specified by their customer and tells the consultant to develop a system that can verify parts meeting these tolerances.
“If you have samples on hand that include the full spectrum of possible failures you need to detect – and not just failures but acceptable variations as well – you can create a formal specifications document as well as acceptance criteria for the final system,” explains Eastlund. “If you don’t have real samples up front, and the integrator does a fixed price bid, they’re setting themselves up for disaster. Because when the customer comes back and says, ‘Oh, by the way, some of the parts are green,’ that red light you specified won’t work at all. Or a customer may not know the difference between detecting a feature and measuring a feature. When we don’t have a specification – and even if we do – one technique we use is to ask to install an image recording system that will allow us to acquire thousands or tens of thousands of pictures of real-world parts. That way, we can determine the critical features ourselves and have a quantifiable idea of where their production quality is before we start and how many variations the machine vision system will need to accept.”
Aptura’s Dechow likes to take his investigations beyond the specification document and sample parts to the upstream production process.
“I love to put imaging systems on production lines to determine manufacturing requirements based on actual production, but it can be difficult to justify a two-day trip to set up a recording system when the overall project budget is only $7,000,” he says. “What really is at issue is the process. I like to walk backwards up the line and see exactly how the part was processed. Many times, walking the line with the engineer, we’ll find out the inspection processes the plant engineer thought were important won’t solve their problem.”
Another benefit to “walking the line” is figuring out exactly where the defect is manufactured in relation to where the customer thinks the machine vision system should be installed.
“Many times, we can save the customer additional money by identifying defective parts before additional value is added,” Dechow says. “Also, seeing the line helps me to address a problem customers rarely consider, which is how to present the part to the machine vision system so we can make the measurement. It’s true that 90% of a machine vision system design is lighting and optics, but the best imaging system in the world can’t find a defect if you can present the defect to the camera. Just because there is an index table at the end of the production line doesn’t mean that’s the best place to inspect it – or that it’s even possible to find the defect without adding more automation to orient the part to the vision system.”
Working with the customer on site also helps to separate critical operations from “wish list” functions that the customer may think are important but can add significantly to the complexity of the design.
“Customers want everything, but they typically need a lot less,” adds David Wyatt, CEO of Automation Doctor Inc. (Mishawaka, Indiana). “Customers are afraid that if they don’t ask for something, they won’t get something that they could have had for ‘free’ or nearly free. It’s human nature. But to the integrator, every added feature reduces simplicity and dependability of the final system. Sometimes, a single requirement that is not very important can add so much complexity that the system is no longer viable.”
Are We There Yet?
Acceptance testing is the step that proves the machine vision system is working as required. It is also a step that is often missing from formal specification documents.
“Integrators and customers have to agree up front on a test method that proves or disproves the vision result,” says Wyatt. One common test procedure is to create a set of parts that represent both defective and good parts and have quantifiable data on defects and acceptable variations. Running those parts through the final system and measuring what is known against the vision system’s results offer a good way to determine if the system is ready for prime time.
Aptura’s Dechow adds that the “challenge set” of parts used to acceptance test the system should regularly be run through the system, and the samples updated as needed. “Parts and part designs change over time; molds age, machines wear down, etc. If you use the same challenge samples you used two years ago, they may not reflect changes in production equipment or design changes. The vision system may be successfully validated, but the challenge samples may no longer represent the customers’ current parts.”
Even a small change in perspective can become a problem for the machine vision system. “It is imperative that the part presentation to the camera be identified and limited in all 6 degrees of freedom,” adds Automation Doctor’s Wyatt. “We typically state +/- X and +/-Y and +/- Z and +/- T in any axis. You would be surprised at how many customers forget how a change in perspective of the camera changes the ability to meet a measurement specification.”
Finally, a good specification will include warranties, spare parts, and operational and maintenance procedures. Integrators may flinch from adding warranties and spare part suggestions beyond those offered by the component manufacturers because no one likes to discuss failure. However, especially when a system includes exotic cameras or parts with long lead times, the customer may think about potential downtime.
“We like to encourage the customer to purchase replacement parts at the same time they purchase the system,” adds Graftek’s Eastlund. “Specifications need to include the total cost of ownership for the system, including how long it will take to get replacements. And it’s not just cameras and lights, but PCs, too. If you’re running a PC with Windows XP, and your system doesn’t fail for three years, now Windows 7 has come and gone. Will your software still work with Windows 8? And what’s the path for upgrade migration? All of these need to be in the specification.”
There are currently no comments for this article.
Leave a Comment:
All fields are required, but only your name and comment will be visible (email addresses are kept confidential). Comments are moderated and will not appear immediately. Please no link dropping, no keywords or domains as names; do not spam, and please do not advertise.