Potato chips, cell phones, razors, and baby diapers are just a few of the many everyday items made with the help of industrial machine vision. In fact, any product that is manufactured in either high volume, high accuracy, low cost, or a combination of all three can be made better, faster, and less expensively using machine vision.

Since the company’s founding 25 years ago, Cognex has relied on standard computer components developed for high-volume applications unrelated to machine vision. That meant using minicomputers in the early 1980s, switching to microprocessors in the mid-1980s, and employing ASICs starting around 1990.

Figure 1: A Vision Area Network can be easily uplinked to plant and enterprise networks, allowing any workstation with TCP/IP capability to display results, images, statistical data, and other information.

When the high-volume applications had switched to desktop PCs, Cognex maintained a dual-pronged approach to vision system development, adapting vision tools for both environments in which they were required to operate: ISA-bus PCs and VME-based rack systems. Both backplanes provided the bandwidth needed to capture and analyze images of manufactured items, and communicate information about the object to the automation system. But these early systems required special-purpose hardware on which to run the vision algorithms until Intel introduced its MMX instruction set in 1997.

Figure 2: Cognex’s highest performance vision sensors utilize advanced PatMax geometric pattern matching technology to reliably and accurately locate parts.

“Since then, we don’t see many customers creating new VME-based systems, though it is still used in some legacy systems,” said George Blackwell, director of product marketing for PC Vision at Cognex. Since Intel Pentium-class CPUs took center stage, most vision systems run vision algorithms, or tools, directly on the CPU of the host PC. The majority of PC-vision systems today use the PCI local bus standard for PCs. Yet, just like the past transition from ISA to PCI, during the next couple of years, PCI will yield to PCI Express.

“We expect high-speed, high-resolution, and line-scan applications to be the first to take full advantage of PCI Express,” said Blackwell. PCI Express offers tremendous advantages for handling large quantities of image data. For example, a four-channel frame grabber over PCI Express offers eight times the bandwidth of a four-channel PCI frame grabber.

Multi-core CPUs and Digital Standards

With little effort from vision suppliers, PC vision systems become faster as CPU speeds increase. But the days of ever-faster CPUs have given way to a trend toward multi-core CPUs, according to Blackwell. “PC vision manufacturers must now optimize their systems to take advantage of multi-core CPUs to increase performance. Customers requiring multi-camera vision systems will see the greatest performance improvement as images from separate cameras can be processed in parallel on separate CPU cores.”

Today’s PC-based machine vision systems also benefit from emerging digital communications standards such as FireWire (IEEE 1394) and Camera Link. Camera Link cameras enable higher resolution and higher-speed imaging to help customers address the most challenging manufacturing applications, including fine defect inspection, precision alignment and measurement, and continuous process inspection.

“First, the camera drivers are built into the Cognex software to simplify integration,” Blackwell explained. “Cognex FireWire systems also use the same application development software, identical vision tools, and proven acquisition engine as other Cognex PC vision systems. And finally, our FireWire implementation is optimized for the high speed typically required in machine vision applications.”

Software Bridges

Cognex is leveraging open-architecture CAD interfaces to provide a new setup option that gives users the ability to train parts directly from geometric descriptions, providing an alternative to image-based training when an acquired image is unavailable or unsuitable for training purposes.

“Instead of relying on noisy or cluttered training images that can sometimes limit alignment performance due to variations in lighting and object shape,” explained Blackwell, “users can import DXF format CAD data to create a synthetic definition that retains essential pattern features necessary for alignment, while excluding component details that are likely to vary.”

This all-digital solution provides users with the ability to import CAD data for alignment, to directly edit high-level geometric descriptions for the purpose of optimizing alignment, and to generate and refine geometric descriptions in a semi-automatic fashion using specialized machine-vision software tools. Because an alignment tool can only be as accurate as the model with which it has been trained, these advances will improve accuracy and process yield, reduce setup time, and raise the level of performance users can expect from their machine vision systems.

The use of synthetic training has already proven very useful for a variety of applications, including fiducial alignment, surface mount device placement on printed circuit boards, computer peripheral component assembly and inspection, and fiber optics component inspection. Cognex also applies synthetic training to eliminate time-consuming training in its ProofRead optical character verification (OCV) system that includes a library of fonts for the most common printers.

Networked Vision Sensors

Vision sensors are systems that are self-contained and don’t require the use of a PC, VME, PCI, or similar architecture to run vision tools. While low cost and ease of deployment remain the key attributes of sensor-based platforms, during the last several years, vision sensors have become more capable.

“Because the DSPs in today’s vision sensors are so powerful, they can run very robust algorithms such as [geometric pattern searches], that in the past you could only run on more powerful PC-based systems,” said Kris Nelson, senior vice president of Vision Sensors for Cognex.

Traditionally, vision sensors have been used to indicate if a product is good or bad, but with networking, more vision sensors are being used for control functions, including quality control, process control, machine control, or robot control.

Today, Ethernet is a key enabling ingredient in the way people use vision on the factory floor. Cognex and DVT vision sensors are examples of cameras that offer built-in Ethernet networking capabilities that enable users to link multiple vision sensors across the factory. These sensors typically come with integrated software for managing vision activity remotely and sharing vision results data with all levels of an organization.

Networking vision sensors provides a number of important benefits. First, it enables vision sensors to communicate with PCs, PLCs, robots, and other factory automation devices. Second, it allows data and images to be archived for trend analysis and continuous process improvement. And finally, networking allows vision sensors to increase manufacturing agility by automating procedures such as changeover for mixed-model processing.

This article was written by John Lewis, public relations manager at Cognex. For more information, contact Mr. Lewis at 508-650-3000.


NASA Tech Briefs Magazine

This article first appeared in the June, 2006 issue of NASA Tech Briefs Magazine.

Read more articles from the archives here.