Vision systems eyes in the factory

Machine vision has been one of the most innovative areas of automation over the past few years. Bulky, hard-to-program vision systems are a relic for museums of ancient automation civilizations. PC-based systems have become stable, while software advances that include graphical interfaces and software components make them easier to use for more control engineers.

By Gary A. Mintchell August 1, 2002


Special machine vision product section

Machine vision has been one of the most innovative areas of automation over the past few years. Bulky, hard-to-program vision systems are a relic for museums of ancient automation civilizations. PC-based systems have become stable, while software advances that include graphical interfaces and software components make them easier to use for more control engineers.

Trends in electronics are invariably toward smaller, yet powerful products, and machine vision is not an exception to the rule. Someone figured out how to put the vision processor in the camera housing, and vision sensors were born. These compact systems make vision inspection an available tool for almost all control engineers.

Engineers implementing machine vision on an OEM machine or a manufacturing line typically hired a system integrator dedicated to the technology. These engineers were trained and experienced with all aspects of vision and automation. The trouble lies in automation economic cycles.

Fred Molinari, president and ceo of Data Translation (Marlboro, Mass.), relates, ‘Machine vision is tied up with general-automation economics, that is, it goes through boom and bust cycles. A machine vision boom may last three to six years then go into the bust cycle for a few years. Many vision system integrators are relatively small operations that cannot survive the downturn. Therefore, you don’t get longevity in the business.’

An experienced integrator has supplier relationships and knows how to specify various components of a complete system and get them up and running in a timely manner. Consider that many PC problems are caused by conflicts with video cards and that two major components of a vision system are a ‘frame grabber’ card that accepts input from a camera and a video card to drive a high-resolution monitor. An engineer would be wise to choose an integrator carefully to assure that all this will work together in production. During downturns, it may be more difficult to find one of these experts.

Mr. Molinari explains, ‘Manufacturers respond to this situation by providing more standardization of components. On the one hand this makes it easier for general automation engineers to install vision systems. On the other hand, this allows less ability for customization. If you need more cameras for different viewing angles, complexity is added. If the application demands too much of this, then an expert who can customize a system is the way to go.’

There are four basic types of vision systems. At the high end are 3D Moire (pronounced ‘mwaa-ray’) systems. Other 3D systems enable vision guided robotics (VGR), a new and growing application. PC-based systems, with a typical system priced in the $10,000 to $15,000 range are next. Vision sensors, often priced from $5,000 to $10,000, are at the ‘low end,’ at least from a price perspective.

Typical applications for Moire systems are metrology, surface measurement, certain semiconductor assembly inspections, and inspection off line. These machines carry a high price compared to other machine vision systems, but they also perform complex tasks.

Within the last six years, PC-based vision systems began to surpass dedicated vision processors associated with PLCs. Exploiting commercial PC hardware and software advances with continuous product development, systems with great power and stability are now available.

System components

Components required for one of these systems include a PC, usually with a PCI I/O bus, but this could possibly be an embedded, VME, or CompactPCI box as well. A frame grabber is a card specifically designed for connection to a camera and communication through the computer’s backplane. Some models stop at that point with the system using the PC’s processor, while others contain the complete vision processor as well. Other hardware items include a monitor, one or more cameras with appropriate lenses, lighting, and fixturing details. The final requirement is software, providing a development environment and runtime processing, as well as enabling data logging and communication.

Vision sensors, on the other hand, feature simplicity. Products can include everything in the camera housing with ports for power and communication, or they may consist of a small camera connected to a DIN-rail-mounted processor. Up until now, these products have been confined to one camera per system. If the application requires two cameras, then two systems must be purchased. Some companies have recently been previewing systems that allow two cameras on one system.

Obviously, PC-based systems are more complex than vision sensors, and typically more expensive. When should the controls design engineer pick one over the other?

With complexity comes power. In the hands of a skilled integrator, a PC-based system can not only handle more complicated inspections than a sensor, it can be more readily adapted to solve future challenges. Need a second, or third, or even fourth camera? No problem. Many cards come with two ports, and a second card can often be added with just some additional configuration in the software.

A vision sensor’s relative simplicity means that an automated machine or processing line can be implemented more quickly by circumventing the initial system configuration and design process. Need another camera or greater resolution for a gauging application? Now that could be a problem.

George Blackwell, senior director of end-user marketing at Cognex (Natick, Mass.), discusses applications for each system. ‘High-end systems are best for 3D application for positioning and orientation, multi-camera systems, situations where you need to see different parts of the same object, detecting defects in parts with patterns or difficult surfaces, high-accuracy gauging, and difficult or complex part identification. Sensors are great for code reading, multiple discrete inspections, OCR/OCV [optical character recognition/optical character verification] especially where code size and distance are known, standard measurement and gauging within system resolution, feature presence/absence, continuous web processes, and some robot guidance.’

What if a system designer didn’t have to adapt a general-purpose system for a specific type of application?

Steve Geraghty, Coreco Imaging (Bedford, Mass.) director of Intelligent Products Division (IPD), notes, ‘IPD is currently developing a series of application-specific products. These ‘vision appliances’ will be configurable through a standard Web browser interface and focus on specific solutions. Specific applications will include inspecting labels on packages or locating, measuring, or controlling manufactured parts.’

Lighting important

Asked about the latest technologies, Mr. Geraghty responds, ‘A wider variety of lighting choices than ever before is now available, and choosing the best type is a key element of success. Deep UV lighting, for example, illuminates ever-smaller features-even down to the micron level-making it ideal for printed circuit board inspections, while laser illumination techniques address 3D measurement applications.’

A solution to the ‘tastes great, less filling’ dilemma of PC-based vs. sensor could be found in a packaged solution from one supplier. This would eliminate the need to choose components from various companies then try to make them work.

Fabio Perelli, product manager for stand-alone systems and frame grabbers at Matrox Imaging (Dorval, Quebec, Canada), reveals, ‘We have integrated the various vision components into a self-contained platform. It provides image capture, processing, display, storage, networking, and general purpose I/O points while supporting Ethernet, serial and parallel ports, and USB. Users can compile programs through Microsoft Visual Basic/C++ tools.’

Applications for vision systems can be found almost everywhere. Mr. Perelli notes several recent applications including a medical 3D, ultrasound diagnostic imaging system for checking fetuses. A robotic application aids in airplane wing assembly, while another vision system inspects a hydraulic mining machine bucket for missing teeth.

Jason Mulliner, vision product manager for National Instruments (Austin, Tex.), suggests that users evaluate their current and future needs. He says, ‘OEMs that serve high-tech industries must have constant upgrades to their machines to maintain competitive advantage. Therefore, they want software, computer platforms, and hardware that are scalable, enabling continuous process improvement without necessitating system redesign.’

Some applications that Mr. Mulliner has seen include sensor testing, optoelectronics inspection/alignment, laser beam profiling, and cell counting.

An important consideration in choosing a vision system is communication and I/O connectivity. DVT (Norcross, Ga.) has championed Ethernet connectivity and has added the EtherNet/IP protocol of Open DeviceNet Vendors Association and ControlNet International, enabling communication to Rockwell Automation’s ControlLogix and SLC 5 controllers.

Programming and configuration are important considerations. DVT uses a ‘click-and-drag’ setup in software and includes a ‘network neighborhood’ function for assigning IP addresses. Omron Electronics (Schaumburg, Ill.) uses a familiar drop-down file menu format. Cognex offers a spreadsheet template for configuring an application. PPT Vision, as detailed in the accompanying sidebar, has developed a graphical paradigm where the user drags icons representing tools and functions on the screen and connects them in logical order.

Speaking of tools, Mark Sippel, vision product marketing manager at Omron Electronics, defines several available tools including ‘Classification’ for sorting packages by label or size, ‘Fine Matching’ for determining small imperfections in label graphics and text, ‘Edge Width and Position’ for part gauging accuracy, ‘Edge Code Defect Detection’ for finding surface flaws, and ‘OCR/OCV’ for date/lot code reporting.

Most vision systems also are capable of 1- and 2-D (also called Data Matrix) bar-code reading.

One other system type is for complex vision guided robotics. Babak Habibi, Braintech (North Vancouver, British Columbia, Canada), says that the time for regarding vision as just another commodity to be integrated after the fact into a manufacturing cell must be changed. He states, ‘Under the old model, integrators are expected to assemble robot and vision commodities to create individual solutions for end-users. Suffice to say that the outcomes of this model have been less than stellar. We believe that it is time for vision guidance to become an integral part of standardized functional manufacturing production cells.’

Mr. Habibi offers a caution when approaching a vision-guided robotic (VGR) application, elements of which are similar to all vision implementations. ‘While image processing is an essential part,’ he says, ‘VGR also requires expert knowledge from such fields as robotics, 3D geometry, and manufacturing processes.’

Certainly understanding vision is only part of the equation of a successful implementation. To integrate machine vision successfully, it is essential to understand the manufacturing process and the control scheme. Then pick the system that fits overall manufacturing objectives. If in doubt about making it all fit together, then search out a qualified vision system integrator.

Comments? E-mail

Vision system designed with integrators in mind

PPT Vision’s (Eden Prairie, Minn.) Impact machine vision ‘micro-system’ is priced competitively with vision sensors, yet possesses the power and flexibility of large, dedicated vision systems, the company says. Its high-speed processor and sub-pixel software algorithms offer a broad range of inspection tasks including Pattern Find, Blob, Edge Find, OCV, OCR, Gauge, and Contrast Sensing. The system can achieve 60 full-frames per second image capture rate, with greater speeds available using the built-in support for partial scanning. DSL (Digital Serial Link) family of high-speed cameras, including new high-resolution and micro-head cameras are supported.

OEMs and system integrators are offered low-level access to the system’s algorithm and function libraries, allowing them to develop custom inspections. Point-and-click GUI software allows quick generation of operator interfaces. The system is DIN-railmountable.

Inspection Builder software enables speedy configuration of inspection programs through a graphical user interface. Icons representing actual inspection algorithms and controls are arranged and configured on the user interface screen to provide inspection of each part. Users can also set up reusable program segments to accomplish specific tasks.

The processor acts as a server for inspection data, allowing results and images to be collected and set up for multiple-location custom viewing and display via Ethernet connectivity. Control Panel Manager software provides panel building graphic components that allow users to customize control panels. Off-the-shelf Java, Active X, or other third-party controls can be added to the system’s library. Interactive touchscreen buttons, image and data displays, and password protection schemes can be created using the same drag-and-drop programming method.

Other features include system calibration that allows multiple systems to be set to achieve defined, cross-factory performance standards; and a watchdog timer that identifies situations that trigger continuous inspection failure and shuts down the inspection and process.

Bob Kuhl, PPT’s director of product development notes, ‘Impact was designed to give system integrators a vision system that can be quickly and easily applied to many different applications while providing the flexibility and power of fully customizable operator interfaces and a high degree of control using advanced inspection algorithms.’

Injecting quality assurance into pharmaceutical manufacturing

SV Research is a U.S.-based company with offices in Harrisburg, Pennsylvania, and Markt-Schwaben, Germany, that designs and builds general-purpose machine vision systems. Completely configurable and designed for high-speed applications, the SVIM machine vision system is based on a dual Intel Pentium processor running a standard embedded Microsoft operating system, and acquires images via a Coreco Imaging (Montreal, Quebec, Canada) Viper Series frame grabber. SVIM systems are used in applications ranging from automotive engine block inspections to glass manufacturing; however, the majority of SV Research’s machine vision systems are used in the pharmaceutical industry for quality assurance inspections.

Bausch & Stroeble (Ilshofen, Germany), a machine builder that manufactures equipment for filling and processing pharmaceutical products, supplies equipment to major pharmaceutical manufacturers worldwide. SVIM machine vision systems perform FDA-required quality assurance inspections on injected drugs. Products are fed into a machine that presents them to cameras connected to a SVIM vision system. The system inspects as many as 50 images of each product as it is processed at speeds of 300-600 per minute. One of the challenges in this application is the need for absolute verification of these images, which are acquired in a burst mode (up to 12 images in 100 msec). The Viper frame grabber’s event stamp feature addresses this challenge. An internal 32-bit counter increments upon receiving an external event and ‘tags’ each acquired frame with a time stamp for subsequent processing and tracking. By reviewing these unique identifiers, the vision system can accurately identify any faulty product, track it to the correct image, and reject it immediately from the production line.

‘Standardizing on Coreco’s Viper Series frame grabbers gives us a variety of acquisition options, each of which uses the same interface software,’ said Ron Lawson, president, SV Research. ‘This allows us to efficiently implement solutions that are the best fit for each customer’s application.’

Since installing the first SVIM machine-vision systems in 1999, more than 400 SV Research SVIMs have been implemented in pharmaceutical manufacturing companies throughout the United States and Europe, fulfilling FDA requirements and helping to ensure the highest product quality.


Several companies have recently released new vision products. See products from these companies, and more, on

Adept Technology

American Eltec

Concepts in Computing




Wintress Engineering