Vision Systems See New Technology Bring Lower Prices
Want to banish the "magic" from machine vision integration and pay less? Machine vision systems today reflect commercial microprocessor advancements as well as software enhancements to make systems easier, more powerful, and less expensive. The power of Intel's Pentium microprocessor with MMX (multi-media capability) has allowed many suppliers to leverage commercial technology that is les...
KEY WORDS
Machine control
Machine vision
Computer boards
Artificial Intelligence
Want to banish the “magic” from machine vision integration and pay less? Machine vision systems today reflect commercial microprocessor advancements as well as software enhancements to make systems easier, more powerful, and less expensive. The power of Intel’s Pentium microprocessor with MMX (multi-media capability) has allowed many suppliers to leverage commercial technology that is less expensive than custom ASICs.
Many suppliers are using technology to base complete systems on a PC platform. This drives down hardware cost, allows interface with other Microsoft Windows-based software, and provides configuration software with a familiar user interface. Component miniaturization has allowed several suppliers to market very low cost systems that are just a step above smart sensors. These are finding wide use in general manufacturing. Fuzzy logic and neural networks—high-end programming buzz words from the last few years—have found their way into vision systems in exciting new ways.
OEMs providing vision as part of an automation system can now add more functions at less cost with additional ability to customize interfaces and communications. Many systems are becoming easy enough to set up, program, and operate that end-users now are installing vision for general applications themselves. One of the most promising new tools is a method part analysis.
Geometric recognition
The common picture analysis method is known as Normalized Grayscale Correlation (NGC). NGC was a good analysis tool for finding images but had some drawbacks. Analysis was slow and subject to errors caused by light variations around the object. An algorithm for geometric analysis existed but required too much computing power to be economical to implement. Then came Intel’s Pentium processor with MMX, providing power to process geometrical pattern recognition at a lower cost than custom chips.
Imaging Technology Inc. (ITI, Bedford, Mass.) offers PC slot image acquisition boards and PC vision analysis software. SmartSearch added geometrical search capability to NGC software, greatly improving both search time and ability to perform inspections in areas with varying light levels. ITI also uses technology from artificial intelligence and neural networking to enhance analysis. Recently introduced is Netsight—a Microsoft Windows CE-based remote vision engine. Windows CE allows developers to use standard tools when adding to the system. The product also uses Ethernet that is built into CE for fast communication to a host.
Cognex Corp. (Natick, Mass.) says its PatMax software technology is the only one solely using geometric object location with no NGC. This geometric system doesn’t look at the pixel level at all, it looks for lines and arcs. A square, for example, is four line segments; a football is two arcs. If the object grows, shrinks, or rotates, relationships still exist and can be found quickly. PatMax can run on a host PC or Cognex’s embedded PCI, Compact PCI, or VME vision systems.
A source of problems in vision applications is electrical noise induced into analog camera cables. Digital cameras reduce that problem. Cognex, in conjunction with Sony, has introduced a digital camera specifically designed for the industrial environment. CVC-1000 combines high-speed acquisition and compact size. It digitizes image data synchronously without filtering or sampling to reduce distortion. Now Cognex, a Rockwell Automation partner, has released software expediting communi-cations with Rockwell’s Allen-Bradley control products.
Adept Technology’s (San Jose, Calif.) ObjectFinder uses geometric modeling to find parts primarily for guiding its robots. The algorithm first uses edge detection tools to define features matched into “feature pairs.” Multiple proposals are tested to match the image to the model at the feature pair level. Best match proposals are prequalified prior to final model matching. Final verification matches image features, along portions of their arcs and lines, against the model to provide accurate information.
Vision does bar codes
Processing bar-code information, including two-dimensional data matrix codes, is a central strategy for RVSI Acuity CiMatrix (Canton, Mass.) according to John Agapakis, chief technology officer. Mr. Agapakis notes, “Higher performance is now achieved through custom processors becoming more economical. Software that is easier to use than the low-level programming of the past is expanding the market. We were a pioneer in processing data matrix code as part of a vision system.” RVSI’s Visionscape is a complete system on a PCI card that provides data gathering and part tracking along with traditional inspection and verification routines.
National Instruments (Austin, Tex.) is well known in the measurement and instrumentation field. According to product manager John Hanks, vision technology is just a logical extension of data acquisition applications. He notes two advances to vision technology expanding potential applications—software and cameras. National Instruments’ Vision Builder is an interactive tool that helps users configure a vision application. Digital cameras are providing faster images at greater resolution than older analog CCD cameras. Getting image data from camera to processor is the current challenge. Mr. Hanks sees IEEE 1394 (sometimes called “Firewire”) technology providing a solution.
Support for CompactPCI and economical PC/104 along with standard PC platforms now is available from Matrox Imaging (Dorval, Quebec, Canada). Matrox Imaging Library includes processing capabilities for FFT, pattern matching, geometric transforms, bar and matrix code reading, and OCR (optical character recognition).
The vision technology theme for Siemens Energy & Automation (Alpharetta, Ga.) is integration with its control platform. As detailed in the “On Line Extra” application story, Siemens’ Videomat system integrates with its PLC to control quality in anti-lock brake systems at Continental/Teves (Morganton, N.C.). senior manufacturing engineer, Hans Gruhn, reports no downtime in his system while increasing process uptime. VS 710 is a digital camera with processing on board. Built-in Profibus port enhances communications ability.
PPT Vision (Eden Prairie, Minn.) has a high-speed (90 frames per sec.) digital camera with built-in processing. The system uses hubs to buffer data from multiple cameras and communicate to automation. All images are processed simultaneously eliminating delays caused by a single processor handling all input.
The latest technology from PPT Vision is 3D. Larry Paulson, chief technology officer, says that PPT owns the patent for Scanning Moire Interferometry (SMI). Traditional Moire interferometry has been used for 3D applications for many years. It basically involves illuminating the object with a Moire light pattern (projected by placing a fine grating in front of a light source) and capturing the image with a high-resolution CCD camera. PPT brings this technology to manufacturing environments by changing the acquisition procedure. Instead of the high resolution 2D CCD array, SMI uses a tri-linear CCD (triple-line scan). As an object moves by the sensor, three intensity points are acquired for each scanned point. 3D topography is calculated using these intensity readings. The target application is ball gate array inspection in semiconductor manufacturing. Motion applications also lend themselves to this technology.
From fuzzy logic to neural nets
Art Gaffin, ceo of SighTech Vision Systems (San Jose, Calif.), says its Eyebot is the first neural network application used in manufacturing. Eyebot is a vision system that doesn’t require programming—it teaches itself. The process is simple. The user places Eyebot in learn mode and passes several good parts by it. The system processes features and stores them in memory. It watches the rate of learning of new features. When the rate drops to a low level, then training is done, and the system is set for run. If a bad or wrong part passes before the scanner, the rate of learning of new features goes up—a signal to the system that a nonstandard part is present. Eyebot processes 13 million features per second which is fast enough to eliminate the need for strobe lighting in many cases.
BrainTron from BrainTech Inc. (Vancouver, British Columbia, Canada) is a “fuzzy adaptive classifier.” Technical director Charles Hooge says the system learns “on the fly” even from partial or noisy data. Unlike neural net systems that need all data first, fuzzy logic systems continue learning new classes. BrainTech has also developed Odysee Development Studio, an object-oriented approach to the design, testing, and construction of applications using machine vision or other sensor recognition systems. It uses a functional vision system model dividing the system into components. Developed subcomponents are “objects” and can be reused in future applications.
Vision as smart sensor
Several companies focus on what can be called the “low-end” market. This term means low-priced, easy-to-use products, not cheap and featureless. The emphasis here is treating vision systems like other types of sensors in a manufacturing setting.
Omron Electronics (Schaumburg, Ill.) says a dominant industry trend is ease-of-use. Its systems require only a keypad setup. Typical applications for its family of products include inspecting assemblies for missing or improperly aligned parts, fiducial and registration mark alignment, presence sensing, and pattern matching.
Mike Schreiber, director of applied engineering at DVT (Norcross, Ga.), says “We make it as easy as possible to use at the plant level. By making vision seem more like a sensor, it becomes acceptable with manufacturing engineers. In fact, our product is called a ‘Smart Image Sensor.'” Series 600 has added Ethernet communications capability. Anything in the data table, not just digital I/O points, can be communicated. It is shown using Opto-22’s (Temecula, Calif.) Ethernet I/O products for control. DVT’s products not only process the usual image data but also interpret bar code information including data matrix.
Keyence Corp. (Woodcliff Lake, N.J.) has a low-cost color vision system. An operator trains the system with a hand-held controller containing a thumb-operated joystick by placing the cursor on the object to be inspected and clicking a pushbutton. The detection range of the hue can be fine-tuned by the operator.
Vision Components (Cambridge, Mass.) has a smart camera with 1,280 x 1,024 pixels. The on-board processor uses NGC to do applications like OCR and data matrix codes.
Advances in other electronics industries such as more powerful microprocessors and visual, object-oriented software are making vision systems more powerful yet easier to use. These systems will become much more prevalent on the factory floor in the not-to-distant future.
Do you have experience and expertise with the topics mentioned in this content? You should consider contributing to our WTWH Media editorial team and getting the recognition you and your company deserve. Click here to start this process.