During the last two decades, machine vision has been applied slowly but surely to a variety of manufacturing challenges, all with the goal of improving quality and productivity in the manufacturing process. Semiconductor and electronics manufacturers were early adopters; they currently account for about half of the machine vision applications found on the factory floor. But acceptance is growing quickly throughout the manufacturing sector, with machine vision systems now in place in food processing, pharmaceuticals, wood and paper, plastics, metal fabrication and other industries.

With progress has come some growing pains. Machine vision was first marketed as a new, must-see technology for manufacturing automation in the early 1980s, a lesser player amid the hype surrounding artificial intelligence and automated robotic assembly. The promise of a mechanical system -- hardware and software -- that would emulate the human eye was captivating in concept but created expectations that could not be immediately met. Start-ups were plagued by complex programming requirements, difficult installations, mediocre functionality and low reliability. The technology required to implement a system successfully was simply out of reach for most users.

After some lean years, the outlook is once again bright for machine vision as products have matured, functionality has increased, suppliers have become smarter and the cost and complexity of systems has come down. Ten years ago, machine vision systems cost $40,000 to $60,000, while today they run in the $5,000 to $20,000 range. They also offer vastly improved performance, with much richer data at much higher speeds.

System components at a glance

Typically, a machine vision system is PC-based, using a group of devices to receive, analyze and interpret the image of a real scene. The system makes judgments on the image using predefined criteria set by the user. This information can be used to automate go/no-go inspection decisions, assembly verification, part location and machine guidance, gaging/dimensional measurements, feedback control loops and a host of other tasks.

It is a common misperception that machine vision systems provide generic optical detection and processing capabilities. While every system includes essential functions, most customers require some level of customization in development and should be cautious of vendors claiming to have "one-size-fits-all" solutions. Systems perform best in their own tightly controlled, highly specialized environment.

Application requirements vary drastically by industry, but a number of components are common to every machine vision system. Technology is evolving rapidly in all these areas, creating new opportunities on the manufacturing floor. The following are common components:

  Cameras -- CCD cameras are becoming smaller, lighter and less expensive. Images are sharper and more accurate, and the new dual output cameras produce images twice as fast as previous models. A new generation of CCD color cameras adds another dimension to machine vision by enabling systems to better detect and discriminate between objects, remove backgrounds and perform spectral analysis.

  Frame grabbers -- These specialized  A/D converters change video or still images into digital information. Most frame grabbers are printed circuit boards compatible with the most common types of bus structures, including peripheral component interconnect (PCI), PC-104, ISA, VME and CompactPCI. Today's frame grabbers offer greater stability and accuracy than earlier models, and some can even handle image processing and enhancement on the fly, using digital signal-processing techniques.

  PCs -- With the advent of the PCI bus, the PC has had a major impact on the use of machine vision in manufacturing applications. Personal computers up to then couldn't gather data at a rate fast enough to keep up with machine vision's heavy      I/O requirements, including data transfer rates of 20 MB/second or greater. The VME bus, a specialized architecture for data acquisition and process control, with bus speeds of 40 MB/second, became a development standard instead. However, today's PCs can handle machine vision's demands, with 132 MB/second PCI bus transfer speeds and >100 MHz Pentium microprocessors. PCs are now routinely embedded into equipment on the factory floor. The distributed intelligence made possible by PC technology has contributed immeasurably to the pace and effectiveness of factory automation.

  Software -- Graphical user interfaces and libraries of high-level software modules operating in standard environments such as Windows have eased the development process and made machine vision a user-friendly tool. Leading-edge software suppliers have begun to provide object-oriented application development tools that will speed application development even more.

  New technologies -- High-speed serial data ports like the Universal Serial Bus and Fire Wire (IEEE 1394) will speed data transfer and information throughput, increasing the overall capability of machine vision systems. USB has already been adopted as an industry standard by PC and peripheral vendors, and will make it simpler to connect digital cameras to powerful embedded PCs. However, reaching real-time video rates will require the higher-speed Fire Wire.

Human vs. machine vision

Even though the potential applications for machine vision are exceptionally broad, it is no substitute for human vision, at least not yet. Today's image and processing technology cannot even begin to duplicate the human eye's ability to deliver information to the brain, nor the brain's capacity to process those images and make decisions based on the visual information. For example, grading of lumber, cattle and produce is a very difficult task for a machine vision system. Human eyes -- and judgment -- are often much better at evaluating the subtle nuances and features that contribute to a quality product.

"Noisy" backgrounds with a lot of detail also tend to disqualify machine vision as an observational and analysis tool. Think of berries on a bush or apples on a tree, for example. Machine vision must be told exactly what to look for, and the display of items of interest must be optimized. A system works best under uniform, controlled lighting conditions, with objects positioned to allow little or no background interference and no interfering reflections.

Machine vision proves most successful in the controlled environment of the factory floor, offering some important advantages over human vision in terms of cost, speed, precision and physical demands. Systems can:

  Determine location or the position of an object.

  Measure dimensions within thousandths-of-an-inch accuracy.

  Count items such as pills in a bottle or cells in a petri dish.

  Identify or recognize an object.

  Inspect objects and identify flaws in manufactured goods.

  Verify that an object's quality meets standards.

Machine vision excels at locating and examining objects with hard, well-defined edges and regular patterns. And its high-speed processing capability gives it unquestioned superiority when it comes to looking at parts on today's fast-paced production lines. Although human inspectors can keep pace with visual inspection demands at a rate of a few hundred items per minute, they also tend to get fatigued and miss flaws. With machine vision, thousands of parts often run past a camera per minute and resolve a dozen features on each piece for product conformance -- all in a matter of milliseconds. Machine vision systems ensure repeatable results and can run continuously 24 hours a day, seven days a week.

The potential applications for machine vision reach far beyond even those areas where human vision can be applied. These include conditions where light levels are too low or too bright for human vision, or where nonvisible electromagnetic radiation such as X-rays or infrared is required. Machine vision systems can be applied in manufacturing clean rooms and can survive environments too hazardous for humans.

The "make-or-buy" decision

Once manufacturers determine that machine vision can be an effective tool for their application, they must decide the best path to take in configuring a system. Larger companies with skilled engineering staffs may pursue their own solution, assembling components purchased from various vendors or even using new technology. However, a steep learning curve, lack of industry standards and time-to-market pressures make the in-house approach largely impractical. The vision system meant to add value to a product can become a serious drain on time, energy and resources. Expert help must be called in to solve the problem.

Outsourcing is a megatrend seen across all market segments as companies find that purchasing a custom-engineered machine vision system entails less risk than designing and manufacturing it themselves. System integrators and value-added resellers have the integration expertise necessary to provide application-specific solutions based on a thorough review of the requirements. Many specialize in serving a particular market niche such as food processing or pharmaceutical manufacturing. This allows them to focus their attention on a smaller range of needs.

Even then, putting together puzzle pieces from a variety of component vendors remains a costly, time-consuming task, mainly due to a lack of industry standards. According to industry analyst Nello Zeuch of the Automated Imaging Association, the cost of components accounts for less than one-third the cost of a machine vision system. The rest goes toward custom development, system integration and installation.

Moreover, the real costs of product development often hide in the lost opportunity cost of not getting a product to market on time. Studies show that, in today's fast-paced markets, the opportunity cost of a six-month delay in product development can far exceed both a 50-percent development cost overrun and a 10-percent increase in manufacturing costs. With the help of an experienced system integrator or VAR, schedules are more likely to be met.

Having the support of an outside source -- long after a product delivers -- adds another advantage. Many companies don't have the in-house support required to get a system back up and running should problems arise. Nor do they have the expertise necessary to upgrade the system later with newer technology. A capable third-party supplier may willingly take responsibility for the whole system, providing an invaluable source of technical assistance and advice.

Working toward standards

The machine vision industry consists of a large number of relatively small firms operating without the benefit of industrywide standards. The AIA is working to correct this lack of direction. But progress remains slow without the benefit of a few large and influential companies to drive the standards.

Meanwhile, users can help ensure that a machine vision system meets their needs successfully. Whether the objective is making accurate measurements, controlling an operation or merely capturing an image, users must work through the design considerations with their supplier. That means understanding the technology, the application and the limitations. Being able to quantify desired results is valuable, too. For example, what system throughput is required? What tolerances?

Finally, industry changes are to be expected and welcomed. A few key trends are shaping the future of the machine vision market. On the supply side, these include standardization and greater specialization among suppliers of components and services. Outsourcing will gain preference as companies continue to downsize and move away from developing custom, proprietary solutions in-house. And hardware prices are dropping rapidly. To be sure, overall system costs -- which include application software, integration and installation -- aren't keeping pace due to the complexity of developing a custom fit. But better tools and interface standards will help the industry as a whole lower prices and meet its growth potential.

Another factor that is having positive impact on the industry is users' growing sophistication. More comfortable with machine vision and with technology in general than just a few years ago, they are better equipped to communicate their needs to the industry. Vendors in turn are able to deliver more effective solutions. That interplay is what makes machine vision the exciting field it is today, full of more opportunity than ever to make good on a concept that was the stuff of science fiction just a decade or two ago.

About the author

George Fabel, Ph.D., is president and CEO of Imagenation Corp., an international leader in the design and manufacture of machine vision components and subsystems. Prior to joining Imagenation, Fabel was president and CEO of CAChe Scientific, a chemical design application software company. He also served for 10 years as director of the Imaging Research Laboratory at Tektronix Inc.

Case Study:

In-line Inspection
Made Easy and Accurate

CyberOptics Corp. of Minneapolis has helped put 3-D machine vision to work on production lines at some of the world's largest corporations. CyberOptics designs and manufactures intelligent sensors and systems for high-production, noncontact dimensional measurement. It is the world's leading supplier of solder paste inspection systems for printed circuit board assembly plants using surface mount technology.

CyberOptics supplies the Sentry 2000, which is used right on the production line at PCB assembly plants to inspect the solder paste on boards immediately after screen printing. By detecting defects before components are placed, the Sentry 2000 helps users increase yields and reduce scrap and rework. Typically, the system is placed over the existing conveyor belt, directly after the solder paste screen printing operation. Using 3-D machine vision technology, the system captures a view of the area selected by the user and, from that image, extracts information pertaining to solder paste height, area and volume at multiple sites per board. Images are stored, allowing operators to go back and view the image of a rejected board after the fact to better understand the problem.

Before automatic in-line machine vision systems were made available, human inspectors would have to pull a board off-line and bring it to a manual inspection system that used a laser sensor to make the measurements. This method not only demanded a skilled operator, it meant measurements were made off-line rather than in-line, as the Sentry 2000 does, at less cost than competitive machine-based solutions. Current customers are end-users such as IBM, Motorola and Lucent.

The frame grabber is a critical component of the system, says Rick Cash, system engineer for the Sentry 2000. He evaluated frame grabbers from a number of companies and chose Imagenation Corp.'s PX500 PCI bus frame grabber for "very high accuracy at a reasonable price."

Cash points out that long-term vendor relationships and good support are also vital to CyberOptics' success, and has found Imagenation to be a partner that responds to their needs.