Chances are, many items that you use every day are made with the help of a machine-vision system. Whether it's a razor, soda cans, light bulbs, or the diaper on your baby, machine-vision technology is at the heart of that product's fabrication.
This technology allows computers to "see" by employing a specialized computer interfaced with a video camera. The system captures the image of an object being manufactured, analyzes that image, and then provides information about the object. Machine-vision systems identify products, inspect for defects, gauge part dimensions, and instruct other equipment like robotic arms to take some action—all at up to hundreds of parts per minute (for a better understanding of this technology, see "What Is Machine Vision?" p. 48).
A pioneer in cutting-edge machine-vision technology is Cognex Corp. (www.cognex.com), whose name is derived from Cognition Experts. The company doesn't have one landmark technology, but several (see DRILL DEEPER 7251, "Cognex Milestones," www.elecdesign.com). Last year, it shipped its 200,000th machine-vision system, representing nearly $1.5 billion in cumulative revenue since the company began operations in 1981.
Cognex's latest groundbreaking product, In-Sight, provides end users with no programming background an easy-to-use machine-vision tool that they can install in one or two applications. In-Sight is a comprehensive vision sensor rather than a vision system. It takes the individual machine-vision components—such as a camera, frame grabber, and computer—and integrates them into one package.
It's the first vision product to employ a spreadsheet for configuring vision applications, thereby entirely eliminating the need for complex programming or reliance on an external PC for setup. In-Sight is significant because it bridges the gap between single-purpose "smart cameras," which typically lack the power and flexibility to solve all but the simplest vision tasks, and general-purpose machine-vision systems, which are higher in price and can be more difficult to use.
At Cognex, the emphasis is on research and development. This follows from one of its founders, Robert Shillman, who left his post as a lecturer in human visual perception at the Massachusetts Institute of Technology to start Cognex. He wanted to go into business and fulfill the potential of his research by applying it to industrial automation. So, he took his life savings of $86,000 and invited two graduate students from MIT's famed Artificial Intelligence Lab, Bill Silver and Marilyn Matz, to join him in founding a new business venture. Today, Shillman is president, CEO, and chairman; Silver is senior vice president of research and development and chief technology officer; and Matz is senior vice president of engineering for the company's Modular Vision Systems Division.
"I started Cognex because I wanted to work each day in a particular type of company—one that had smart, energetic people who enjoyed what they were doing, who liked to have goals and to accomplish those goals," says Shillman. "And I wanted a physical environment that reflected the culture of the company, which I wanted to be dynamic. I also wanted to work for a company where the management would tolerate, or better yet, seek out outspoken and assertive people like myself who don't care about office politics but who care deeply about doing the right thing for customers. Well, I couldn't find such a company, so I had to create it."
"Machine vision is as necessary a part of advanced manufacturing as gears and motors," adds Silver. "Without it, many of the products we take for granted simply could not be manufactured, and without those products the modern world as we know it would be very different." The ramifications of machine vision are far-reaching (see "What Can Machine Vision Do?" p. 50).
How can manufacturers get machine vision to work on the factory floor as it does in the lab? Real-world manufacturing conditions, which are rarely evident in the lab, include inconsistent ambient lighting and optical distortion, as well as changes in component shapes and surface characteristics. These variations can interfere with a vision system's ability to "see."
"The answer to overcoming these hurdles is a strong collaboration among the automation equipment supplier, the machine-vision supplier, and the end users," says Matz. "It's critical that the machine-vision supplier is involved throughout the entire development process, so it can not only react to problems that occur but also help companies anticipate unforeseen problems. When the invention of new vision technologies is necessary to handle specific challenges, tight collaboration can help ensure that it begins early enough to avoid impacting the production schedule and the acceptance of the vision machine."
MAKING A DIFFERENCE
The focus on research and development has produced many engineering breakthroughs. Cognex was among the first to identify codes embossed on rubber (tires) and etched on metal (circuit boards). It proved to a doubting industrial world that machine product identification could be done at the speeds required by manufacturing. A series of advances in conducting industrial pattern recognition, such as normalized correlation and then geometric pattern matching, changed the way objects are located in semiconductor manufacturing and electronic assembly. Overall, the breakthroughs increased accuracy in finding objects by a factor of 40 over a 10-year period.
The company made these advances because it has always been willing, from day one, to take risks. Initially, says Silver, it was the "idealism of youth" that sparked their desire to create machine-vision tools, despite the lack of information on the subject. "We saw a problem people wanted solved, and we were crazy enough to do it," he says. "We had our share of good luck too. Had we known how hard it would be, who knows if we would have given it a try?"
That philosophy, seeing a problem and working to resolve it no matter what, still prevails. "We have set the direction for what customers of machine vision expect and what suppliers produce. We've taken risks to produce new things before people had an idea they were possible or valuable commercially," says Silver (see DRILL DEEPER 7252, "A History Of Success," www.elecdesign.com).
MAKING IT EASY
Through the years, Cognex has experienced success selling to both OEMs and end users by recognizing that end users have different needs and requirements than OEMs. End users are interested in solving a particular problem one time, not in repeatedly applying the same solution in different applications as is the case with OEMs. End users want something easy to use, because they have just a few days to devote to the project and they might only install it once or twice. They also need a tool that doesn't require complex programming skills, and of course, it has to be available at a low cost.
To cost-effectively address this need, Cognex had to create a tool with enough flexibility to solve problems in a variety of applications. "To make a machine-vision tool without complex programming for end users was a real challenge. You gave up enormous amounts of flexibility. That's been a constant machine-vision problem," says Silver.
Cognex met that challenge with In-Sight, which was introduced in 2000. "We start with state-of-the-art machine-vision algorithms, then add a unique interface based on a spreadsheet," says Silver. "You don't have to be a highly trained engineer to use one. It is a way for someone who is not an engineer to ask for fairly sophisticated operations to be performed. We then married that to machine-vision tools we pioneered."
End users definitely like it. Jim Gallion of BesPak, maker of drug delivery devices, says, "Our technicians need to be able to train the vision system on a new reference image for each tool, so the system knows what the clean mold is supposed to look like. The spreadsheet makes this a two-minute process for them." Figures 1, 2, and 3 demonstrate In-Sight at work in diversified applications around the world.
The design challenge with In-Sight was making something significantly less expensive and easier to use than previous systems, while maintaining flexibility. "It was the use of the spreadsheet that was the key ingredient that made that possible," says Silver.
But using a spreadsheet also presented a few challenges, says In-Sight creator John McGarry, senior vice president of In-Sight Products. "Conventional spreadsheets are intended to evaluate individual snapshots in time. We had to come up with a way to extend the standard spreadsheet model to include the idea of continuous streams of data," he says.
"While spreadsheets are famous for making 'what if' calculations," says McGarry, "they don't remember the results of prior 'what ifs.'" Cognex solved the problem by implementing clocked data storage in the form of a function that simulates shift registers, delay lines, latches, accumulators, etc. Unlike conventional spreadsheet formulas, where evaluation is triggered by changes in data dependencies, In-Sight's clocked data-storage functions only update on a specific clocking event. This subtle difference significantly extends spreadsheet capability.
"If we want to know how many failures occurred in the last 100 cycles, we can insert a counter in the spreadsheet to accumulate failures, or we can direct inspection results into a shift register," says McGarry. Other In-Sight spreadsheet functions enable the spreadsheet programmer to devise queries like "How many cycles ago was the last failure?" In-Sight can monitor, for example, the standard deviation of values over an interval in time.
"We can use clocked storage to dynamically adjust parameters that affect the current inspection cycle," says McGarry. In effect, the spreadsheet's results can easily be made to depend on the results of previous calculations.
Another issue that McGarry faced was controlling which cells execute a command. In a spreadsheet, any formula or dependency cell changes when the referred source changes. Cognex created a conditional basis for cells that could disable a particular cell based on a condition resulting from the condition in another cell. That gave users the flexibility to dynamically alter an application's behavior.
The next challenge was the user interface. It needed ease-of-use for nonprogrammers in an environment not always friendly to keyboards. Factories are often short on desk space and have a lot of dust flying around. This time, the solution was a handheld control pad like those used in the Sony Playstation. It had four buttons on the right to affect different types of controls, four on the left to navigate, and four in the center for special effects when viewing and triggering acquisitions. "The ability to use a game pad is almost as universal: Combining it with the spreadsheet resulted in a powerful and almost fun way to program," says McGarry.
Hardware has also changed. By integrating the three parts of a vision system (frame grabber, camera, and processor) into one package, Cognex was able to create a product customized for machine-vision applications. In-Sight's intelligent cameras are highly specialized. Today's die-cast package meets the IP67 standard, looking nothing like the surveillance cameras once borrowed from the security industry. Its 64-bit microprocessors are approaching gigahertz clock speeds. Notes McGarry, "They don't consume a lot of power. That helps us make them rugged."
Cognex must adapt its tools for the environment in which they will operate. In the early days of machine vision, RS-232 ports provided serial communications with host systems. These systems simply needed to give a Yes or No response: High bandwidth was unnecessary. Next came standard buses like VME and then PCI (Peripheral Component Interconnect), a local bus standard for PCs. Now everything in manufacturing is Ethernet, so In-Sight is entirely Ethernet driven.
The elements of a spreadsheet basis and game-pad controls integrated in a rugged housing that tied into an Ethernet network succeeded. "In-Sight is growing faster than anything we've seen in the past. It grew to more than a quarter of our total revenue in 2003," says Silver.
Although Cognex has several new products scheduled for launch the first half of this year, Silver is very reluctant to predict the future in machine vision. "That's extremely dangerous. Asking an expert is often no better than throwing darts at a dartboard," he says. Without a doubt, it's moving in the direction of inexpensive intelligent sensors.
"We have hardly begun to scratch the surface of machine-vision applications and technology," says Silver. "We have examples in the world of human vision that are so far in advance of anything we can make with machines, that I don't expect us to catch up in my lifetime or my children's."