A compass is a pretty basic instrument, especially when compared to cell phones, laptop computers, global positioning systems and other modern marvels. But, without that unique tool, Lewis and Clark would never have completed their epic journey through the Pacific Northwest 200 years ago.
Today, many manufacturers are turning to machine vision systems to guide them into the 21st century. While machine vision technology is not new, the latest generation systems feature major improvements over earlier disappointments. New tools, such as digital cameras, LED lighting, and the Camera Link and FireWire standards, help simplify and enhance assembly applications. Many components are now smaller, faster and more robust.
While state-of-the-art machine vision systems are not cheap, they offer numerous benefits to assemblers. For instance, machine vision technology cuts down on retooling costs, lowers labor costs, improves flexibility and boosts quality.
"Machine vision can improve a manufacturer's bottom line more than any other automation technology," claims Joe Rogers, vice president of marketing at PPT Vision Inc. (Eden Prairie, MN). According to Rogers, quality-and the potential cost of defects and recalls-is forcing many engineers to reconsider machine vision technology.
Seeing Is Believing
After 20 years of trial and error, engineers are finally starting to accept machine vision as a cost-effective quality assurance tool. Many of today's newest converts were early adopters of machine vision systems back in the 1980s, when the technology failed to meet their expectations.
Several factors now make vision technology more appealing than in the past. For instance, many manufacturers are under more pressure to trace and validate the products on their assembly lines. As a result, there has been an increase in the use of machine vision for reading identification codes marked directly on parts. Automotive, aerospace, medical device and electronics manufacturers are increasingly using identification codes to error-proof their processes, and to create an accurate history of each part from the beginning of its life to the end.
In addition, lean manufacturing initiatives demand that an inspection system be in place to detect defects and rejects before too much waste or scrap is generated. Vision systems often fill the need for a reliable, 100 percent inline inspection process.
Increases in production volumes, faster production speeds and workforce attrition due to cost-cutting measures often make the use of manual inspection hard to justify. In some cases, the human eye is simply no longer able to detect flaws because products have become so small.
"Machine vision systems save companies money over time," says Kyle Voosen, data acquisition and vision product engineer at National Instruments Corp. (Austin, TX). "With manufacturing becoming cheaper overseas, and customers expecting defect-free products, a company is gambling with its own future if it ignores the long-term benefits of machine vision. If machine vision keeps a defective product out of the hands of a customer, then it has saved more than just money; it's also preserved the reputation of the company and the confidence of a customer."
"The need for productivity improvements ultimately drives demand," adds Philip Colet, vice president of marketing at Coreco Imaging Inc. (St. Laurent, Quebec, Canada). "Machine vision is viewed as the way to deliver improved productivity. However, unlike 15 years ago, machine vision systems are successfully meeting the challenge."
The biggest difference in vision systems over earlier technology is more speed for less money. With faster processors, vision systems can acquire more information from cameras at faster frame rates than before. They can also take advantage of more complex algorithms to verify assembly, while maintaining high speeds.
Today, vision systems are much more reliable than their predecessors. They can acquire and accurately analyze a huge amount of data on time. "Earlier machine vision solutions were not so robust, which left a bad taste with users," says Himanshu Shah, a senior analyst at ARC Advisory Group Inc. (Dedham, MA). "New equipment and algorithms are more robust due to high investment fueled by the semiconductor market."
However, any machine vision system is only as good as the components behind it. There is a critical link between cameras and high-speed interface protocols. As product tolerances become smaller, demand for higher resolution and faster processing speed increases.
New and Improved
Today's machine vision technology is faster and less expensive than in the past. Ten years ago, machine vision could only be justified for assembling expensive items, such as semiconductors and electronic components. Now, practically any manufacturer can justify the cost of the technology, especially for high-speed applications.
"There is still a perception that machine vision is expensive and unreliable," notes Voosen. "Vision earned this bad reputation many years ago, and the [myth] persists. While vision is still coming into its own, the successes of early adopters speak for themselves. In the near future, machine vision will move from being a competitive edge for manufacturing and assembly to a required component-just like the conveyor belt."
However, some vendors "have led manufacturers to believe a machine vision system can be easily deployed for only a few thousand dollars," warns PPT Vision's Rogers. "A recent Automated Imaging Association (AIA, Ann Arbor, MI) market study indicates that while the cost of deploying a machine vision system has been cut in half over the past 10 years, the average deployment cost is still in the $25,000 to $50,000 range."
While the power has increased and the cost has decreased in recent years, the effort required to install and program vision technology has declined dramatically. Indeed, most vision systems now feature much more user-friendly tools that simplify assembly applications.
"Machine vision systems are easier to use than 10 years ago," claims Ed Roney, manager of vision products at FANUC Robotics America Inc. (Rochester Hills, MI). "The vision, setup and application tools are significantly better. Application-specific vision systems, like systems for robotic guidance, where the task of the vision system is constrained into the application and task at hand, offer the best opportunity for a simple implementation. In most cases, the vision system can be applied ‘out of the box,' requiring no custom programming-just setup."
"In the past, the needed software may or may not have been present," adds Reno Suffi, marketing manager for sensors at Omron Electronics LLC (Schaumburg, IL). That forced end users to find creative ways to solve their particular application. In contrast, Suffi says today's software tools have "fleshed out" many of the more common applications, such as edge detection and center of gravity, and these are now part of any standard tool sets.
"This ease of use increases performance reliability, which results in a wider level of user acceptance," explains Suffi. "Therefore, we are seeing an acceptance of low-end, user-friendly vision systems." Several years ago, Suffi says the market was dominated by complex application-specific systems. Today, there is a trend toward general-purpose, lower-end, lower-cost vision systems.
"The industry has come a long way, driven by both demand, as well as the advances in processing capabilities of computers, and the increase in speed and resolution of cameras," adds Sunderraju Ramachandran, industry manager for test and measurement and semiconductors at Frost & Sullivan Inc. (San Antonio).
"Use of commercial PC technology has dramatically reduced cost of ownership," Ramachandran points out. "Improved graphic user interfaces have resulted in easier to use systems that no longer require highly skilled operators. Setup and programming of systems are also easier, enabling less-skilled technicians to operate and maintain machine vision systems. New machine vision technology being developed is increasingly flexible, giving users the option to tailor use to needs at any point."
For instance, vendors are starting to embed knowledge into software, rather than making end users rely on the traditional point-and-click interface. These smarter, configurable tools create a friendlier development environment.
"With the increased quality of configurable machine vision software, being a computer programmer is becoming less important," claims National Instruments' Voosen. "You do not need a degree in computer science to build a machine vision system anymore. Engineers are now able to use packages to prototype and deploy their applications."
For instance, Voosen's company offers a configurable program called Vision Builder that does not require programming. This easy-to-use software allows end users to build, benchmark and deploy integrated machine vision applications. It can communicate inspection results with PLCs, robots, servers and operators.
However, some experts believe there is still a lot of progress to make in the area of ease of use. "By nature, machine vision is one of the most complicated technologies you can find in an assembly plant," notes PPT Vision's Rogers. "Users have to deal with lighting, optics, high-speed image processing and software."
Fortunately, there is a broad range of machine vision technologies available today, ranging from high-end PC-based systems down to very low-level embedded sensors. "The distinction between these high-end, midrange and low-end systems is much more seamless," claims George Blackwell, director of product marketing at Cognex Corp. (Natick, MA). "Consequently, end users today may base their selection on more than just performance. [Engineers] are more likely to consider factors such as how it integrates into the overall system, what kind of operator interface they want in the system or environmental considerations."
One problem that continues to plague machine vision technology is widespread confusion over interface standards and protocols. End users can choose between Camera Link, FireWire, Gigabit Ethernet and Universal Serial Bus. Each standard has pros and cons.
- Camera Link was developed for high-performance vision applications. It provides a standard connection between cameras and frame grabbers, while at the same time reducing cabling costs.
- FireWire, also known as IEEE 1394, is a high-speed serial bus that was developed to link digital cameras to personal computers. It features plug-and-play usability and uses a low-cost PC interface.
- Gigabit Ethernet has been around for more than 20 years and is widely used for local-area network applications. It is scalable and easy to implement.
- Universal Serial Bus (USB) was developed to connect printers and other peripherals to PCs. It uses a shared bus and supports asynchronous and isochronous transfers.
"There is confusion in the marketplace between competing digital standards," says Coreco's Colet. "Unfortunately, the governing bodies of these organizations have done little to alleviate this confusion."
"But, we are much better off now than we were four years ago," says Voosen. "Back then, customers needed custom cables every time they wanted to connect frame grabbers to the proprietary connectors on cameras. With Camera Link and FireWire, connectivity is much simpler, but the choices still remain."
According to ARC Advisory Group's Shah, more options are confusing only when end users do not understand the basic technology. "The especially fast-changing nature of networking will always place doubts in users' minds about whether the right technology is being used for their applications," explains Shah. "Understanding your application requirements in detail will make the process of selecting from a variety of technologies and components easier."
Keith Russell, director of sales at Redlake MASD Inc. (San Diego), claims the choice "is not as difficult as one may think. The selection of the interface is done by determining the data throughput requirements and cost limitations. I don't think we can make an assertion as to which is the best standard or protocol."
"Everything starts with the application," adds Kevin Taylor, sales manager at ISRA Vision Systems Inc. (Lansing, MI). "There are many different ways to communicate inside a plant. [The decision typically] depends on the application's specific needs, such as the length of cable between the machine vision processor and PLCs or robotic controllers."
Most observers agree that FireWire covers the most assembly applications. For instance, FireWire cameras use only one six-pin cable and do not require frame grabbers. However, they are generally slower than Camera Link cameras, which are more widely available and offer much higher performance.
Camera Link is based on technology used in flat-panel displays. It was developed four years ago by a group of camera and frame grabber vendors, and is administered today by AIA. Camera Link includes features that specifically address the unique needs of machine vision systems, such as standard cabling, data formats, triggering, high-speed acquisition and camera control.
"Today, the Camera Link standard appears to be the best suited for machine vision applications, especially in assembly, with an increasing number of industry participants developing components for this specification," says Frost & Sullivan's Ramachandran. "This provides an increased choice of components, resulting in a more optimal system."
"For higher speeds or greater resolution, Camera Link is the way to go," adds National Instruments' Voosen. "There are no camera standards for USB or Gigabit Ethernet, so these cameras have not taken off."
One of the most critical components of any machine vision system is its cameras. Two trends are working to simplify camera connectivity: Consumer buses and embedded vision systems.
"With cameras focusing on new, fast consumer buses, there is really no need for a specialized frame grabber," says Voosen. "Also, with compact, embedded vision systems, the need for a separate frame grabber no longer exists. These compact vision systems acquire and process images using real-time operating systems. They do not rely on PCs and Windows to analyze images."
Many machine vision applications today use digital cameras, which are not bound by the resolution and speed specifications of analog cameras. They also offer higher resolutions, greater pixel depths and more variable frame rates.
"Digital cameras are compact, with advanced integrated electronics and packaging," says ARC Advisory Group's Shah. Frame grabber functionality within the camera lowers the overall cost of the vision system. Embedded intelligence enables the camera to process images and network with PC-based controllers or PLCs.
Digital cameras typically offer high image quality, very high-speed acquisition and higher resolution. They often do not require a frame grabber, and digital signals are less prone to degradation than analog signals. However, cable lengths are usually shorter than with analog cameras. In addition, digital cameras carry a higher price tag than a typical analog camera.
Many machine vision users are eager to eliminate the confusion surrounding image transfer protocols. Products that contain the imager and image processor in a single package are now available. They're known as "intelligent cameras" or "smart cameras." These products are just beginning to offer all the features of full-scale vision systems.
"There is a big difference between digital cameras and smart cameras," says PPT Vision's Rogers. "Digital cameras have been widely deployed for a long time. They were part of a natural evolution from using analog-based video signals to digital video signals. This improved image quality and reduced noise. Smart cameras, on the other hand, contain the image processing hardware right inside the camera."
Earlier intelligent cameras did not have the horsepower of full-fledged vision systems, so they represented a small part of the overall machine vision market. "But, with improving power, speed and image processing software, this is the most promising category," Rogers points out. "Intelligent cameras now offer more advanced features, such as automatic recipe control, subroutines and multitasking. The integrated nature of intelligent cameras means that manufacturing engineers can avoid the protocol debate altogether."
Despite all those benefits, a debate is raging over the pros and cons of digital cameras and smart cameras. "Smart cameras do relatively well in simple applications, such as barcode reading and low-resolution measurement applications," says Redlake's Russell. "But, for more demanding applications, digital cameras are the better choice. A digital camera offers the flexibility of the interface, higher resolutions, and your choice for a machine vision processing engine."
"Digital cameras, whether they are Camera Link or FireWire, offer a number of advantages over smart cameras, namely scalability, performance and choice," adds National Instruments' Voosen. "When a higher performance imaging sensor is introduced, smart camera users are forced to replace their entire vision system to upgrade the image sensors. Also, with smart cameras, users must choose between only a few possible sensor options, such as color, high resolution and high speed." Consequently, smart cameras tend to be small, simple to program and inexpensive.
"Many applications cannot be solved by one of a few predefined smart cameras," claims Voosen. "Conversely, there are hundreds of digital cameras on the market that may provide a better image for a given application. These cameras cover many types of performance, from very high speed to very fine sensitivity." In addition, smart cameras are generally not available for specialty applications, such as infrared, line-scan and X-ray inspection.
While the debate over cameras will continue, one thing is certain: New technology is changing the traditional roles of camera manufacturers and vision board suppliers. As they struggle to add value, some vision board companies are manufacturing their own cameras and a few camera companies are manufacturing vision systems integrated with their cameras.
"As software becomes an ever important component in a machine vision application, camera manufacturers are forming strategic partnerships with the vision software companies," says Russell. "The line between vision board manufacturers and camera manufacturers is blurring." As vendors search for ways to add value, Russell predicts there will be more strategic partnerships, acquisitions or mergers in the machine vision industry.