Machine vision can greatly benefit quality, but only when all of the components-and people-work together. From the initial project specification to the final implementation, it can be a complicated process. And a high level of technical expertise is not enough: you must be able to articulate exactly what you want. Ned Lecky, president of Lecky Integration (Little Falls, NY), compares it to a college course. If the professor doesn’t say what’s going to be on the midterm, students don’t know what to expect. But if the professor explains the format and what to study, students will how to prepare. Similarly, if an end user tells an integrator what the system needs to pass the final test, there is a better chance it will pass. “To me it all seems like common sense, but unfortunately it’s not so common,” Lecky says. “If goals aren’t explicit in the contract, everyone makes up their own.”
The Project SpecificationIt all starts with a project specification. It is a truth universally acknowledged that product specifications are inadequate. In theory, it explains everything that is needed. In practice, there are always gaps. “The first thing to realize is that the specification is always incomplete,” says Perry West, founder and president of Automated Vision Systems Inc. (San Jose, CA). “Nobody knows what’s missing the first day; you just know that things are missing.”
In his thirty years working in machine vision, West, now a machine vision consultant, has seen a range of specifications, from customers who provide instructions with nothing more than vague comments and hand waving to those that are write specifications down to minute details.
Integrators often say that finding out what customers want can be the biggest challenge in integration projects. John Nagle runs a machine vision consulting business in Cedar Park, Texas, and says this is often not as straightforward as it seems. For example, the customer may say that they want a part inspected. Simple enough, but what happens when the parts get jammed? A customer may assume that this will be included in the final software package, but it may not be if they don’t mention it. “Every industry has a very specific set of problems that an integrator will have to negotiate when they’re doing a machine vision project,” Nagle says.
For best results, stay away from the latest buzzwords and simply specify what’s needed. David Dechow, president and founder of Aptúra Machine Vision Solutions (Williamston, MI), says the fundamentals haven’t changed: the machine vision end user should be aware of the available technology and how it suits the application. Just because a shiny new technology is available doesn’t mean it is right for you. “A lot of people in the end user community tend to gravitate toward the latest buzzword,” Dechow says. “Fifteen years ago it was infrared lighting. Now people ask me for a camera with a liquid lens. End users will latch onto a buzzword and think it’s the great thing for them.”
Will It Work?Norman Axelrod of Axelrod Associates (New York, NY) reminds clients that specifications that are too tight could reduce yields. The machine vision system should improve the current quality levels without reducing yields unnecessarily. End users should ask themselves a few important questions. What are the smallest defects that the system must detect? And what is the cost of letting defects go by?
Keep in mind that even things like 99.9% accuracy can mean one mistake per minute.
“Machine vision is really subtle stuff,” says Lecky. “A lot of people just use nebulous terms that mean totally different things to different people.”
To improve the chance of success, bring integrators into the process. They may be able to help you locate defects 10 steps earlier in the workflow. Have them walk the process back up the line and see what they discover. “Quite often, we find ways that we could incorporate machine vision to catch the problem at a much less expensive point than we’ve thought,” Dechow says.
While cynics may think that the integrator is ramping up costs unnecessarily-and this can happen-integrators often have seen dozens of similar projects. If end users are open to spending more money upfront, they may receive a vastly improved system. Kerry Van Iseghem, co-founder of Imaging Solutions Group of NY Inc. (Fairport, NY), was once called in to help a company find a lower-cost camera with higher resolution. “They were looking to lower the cost of a camera, but it’s a lot more fun to look at whole system to save more money,” Van Iseghem says. He came back with an expanded proposal that redesigned the company’s entire system, which cost more upfront saved them more money.
Machine vision software may seem like an ethereal expense, and it’s difficult to see the costs involved so ask questions to learn what the integrator is doing and why. It helps if the integrator paints a picture of what the system and software will look like. “We can describe it in minute detail in a War and Peace size document and they won’t read it,” Nagle says. “You have to give pictures.”
And after asking questions, don’t forget to listen. A month ago, Dechow had an application run into issues after explaining to the machine builder exactly where to put the light and camera. But when he arrived on the scene, he found they had decided to move the light over five inches. “Moving a light five inches is not good for machine vision,” Dechow points out. Compromises in lighting can doom a whole project.
As this type of situation illustrates, communication is critical. “So as painful as it is, I really like weekly status calls,” Lecky says. “Get everyone together, even the technical people. That way there are no surprises. Nobody likes any surprises in these kinds of projects. If done well, it should be really boring.” In addition, he shares Google Documents with clients to show projected complete dates and test results. “That level of transparency really helps,” he says.
No matter how you handle communication, frequently check in on progress, and try to be available for questions. When these inevitably crop up, the integrator will need someone who can get the answer quickly.
Once a system is finally in place, be sure that the integrator provides easily understandable instructions. Ideally, newcomers should be able to run the system with a minimum of training. “Like a pilot, you have a checklist,” Axelrod says. “These are experts and they have a checklist because they could forget something and it could be a key something.”
PaymentThe budget obviously will be a factor in setting up the application, but so is the payment schedule. Some integrators require an initial payment of 20 to 40% with an order.
West says some companies have a policy not to send an initial payment, saying that the integrator should have the money to run their business. But integrators want customers to be committed to the project, so the initial payment can contribute to that. The idea is to make sure a customer has something to lose if the project doesn’t go as planned. “It’s certainly an area where people have different opinions,” says West.
The manufacturer obviously doesn’t want to be stuck with a system that doesn’t work, and the integrator doesn’t want to be out a lot of time without any compensation.
The Finish LineHow do you know if a job is finished? “It’s so hard to know when ‘pencils down’ time really has arrived,” says Lecky. “A lot of engineers would be perfectly happy to work on something until they retire.” They could spend 40 to 50 hours a week on the project for the next 20 years. But of course, that is not how things work. Ideally, you, the end user, should decide when the project is done. The integrator should have explained the system, provided training and ensured that everything works as planned. If the integration went well, you may have found a partner for future projects. V&S
Tech TipsDo your research. Try to solicit the opinions of more than one integrator to get the best solution.
Provide a clear project specification to get the best chances of success.
Work with the integrator and allow them to provide suggestions that may smooth the process.