![](/FE/2001/06/Files/Images/80099.jpg)
Suppose that you've engineered a processing line for a cheese and cracker snack that enjoys strong regional distribution through a major supermarket chain. Now suppose the chain phones one day to complain that a recent shipment of the product contains snacks with insufficient amounts of cheese. And because this isn't the first time the problem has occurred, it is threatening to discontinue sales of the product.
What do you do? Well, you're going to take a good, hard look at your filling operation - that's a given. But are you going to continue with the same random, manual inspections you currently employ to ensure that fill levels meet specifications?
Probably not. With machine imaging technology becoming more affordable and easier to operate, it's probably simpler - and more efficient - to perch a camera where you're line inspector is currently standing. Of course, machine imaging technology is bit more complex than that, typically comprising not only an imager and computer, but also software algorithms to provide an interpretation or classification of image data.
With most such installations, images are captured and sent by camera at the request of the hardware, and then stored and evaluated to determine whether they conform to the predetermined criteria. Serial or parallel data available after each inspection is typically used to integrate the vision system into the production line. For instance, digital outputs can be used to activate a reject mechanism, as well as provide simple statistics on the amount of good product versus bad the line has generated over a specified period of time. The serial alternative can communicate all checking results via a network to a host or other computer-based system, and transfer the results so that full statistical process control can be employed.
![](/FE/2001/06/Files/Images/80100.jpg)
A growing array of systems
Beyond these general attributes, however, machine vision systems collectively encompass a wide, even unwieldy, range of technologies and approaches. While some systems employ video cameras, others -- depending on the product or application -- rely on photoelectronic sensors, infrared technology or X-rays. Some systems are designed to evaluate products on the basis of color, others on the basis of size or shape, and still others on various combinations of the three. Some function as stand-alone products, others are incorporated into sorting machines equipped with conveyors, ejectors and other features. Some rely on complex software programming, others on comparatively simpler methods. For instance, supplier Cognex Corporation manufactures a stand-alone industrial vision sensor that features a spreadsheet allowing users to select tools and parameters through a series of simple menus and dialogue boxes, and then link cells together to perform functions such as locating objects or measuring parts. Engineers can thus avoid the "hassles of programming," according to Justin Testa, senior vice president of marketing for Cognex. Further, the system does not require a PC. In food packaging applications, the product is used to measure the position of product labels and read 2D and 1D bar codes on packages.On production lines, the vision sensor's digital camera is typically mounted at a camera-to-subject distance of 18 inches. The camera connects to an industrial-hardened vision processor that contains an onboard DSP chip for high-speed vision processing, and links directly to a standard VGA monitor. A hand-held control pad allows for rapid navigation through the worksheet's cells and menus, enabling the user to quickly configure an application and make on-the-fly changes during the run. For a job involving the evaluation of labels on high-end spice jars, the entire inspection sequence occurs in approximately 300 milliseconds, keeping pace with a line speed of 200 jars per minute set by the labeling machine. If a jar fails inspection, the vision sensor sends a reject signal via its on-board discrete I/O to a PLC, which then triggers a pneumatic gate to eject the jar into a reject bin.
Cognex developed the system -- known as In-Sight 2000 -- to fill the niche between "smart cameras," which are typically reserved for simple vision tasks, and general purpose vision machines, which tend to be more expensive and complex, said company spokesman Evan Lubofsky. The system's price: less than $5,000.
A more recent version of the product -- In-Sight 1000 -- integrates all vision processing, software and Ethernet communications into a camera-sized unit, thereby obviating the need for a separate vision camera and processor. According to Testa, the 1000 model responds to the "increased demand for adding low-cost vision at more points through the manufacturing process." He added that the system makes multi-point vision more "cost-justifiable" while allowing users to manage individual units as a system over the Ethernet network. Users can set up and modify vision applications from remote sites, monitor inspection activity from any plant location and share current production data with management. The 1000 model sells for less than $4,000.
Supplier Allen-Bradley's ColorSight 9000 Photoelectric Sensor provides a low-cost alternative for applications in which products are sorted by package color on conveyor lines. The sensor is equipped with precision adjustment capabilities that enable it to detect subtle differences in color shades (e.g. light blue versus dark blue), thus improving accuracy. According to Allen-Bradley, the sensor is available for a fraction of the price of a traditional vision system.
Like many imaging systems, the Cognex and Allen-Bradley products are designed for a several industries in addition to food, including automobiles and medicine. However, a growing number of systems have been developed, or at least customized, for food processing applications. "The problem with food is that, unlike most other industries, your're often dealing with products that have natural and acceptable variability from one specimen to the next," said Craig Wyvill, division chief of the Food Processing Technology Division at the Georgia Tech Research Institute in Atlanta. "It's difficult to develop technology with the decision-making capability required to account for that variability and still meet the demands of a high-speed processing line." Several recent offerings have been developed with just that in mind. In 1999, for example, SRC Vision - a supplier later acquired by Key Technology, Inc. - adopted infrared technology for its Prism line of vision processors as a means of inspecting peel-bearing potatoes. The technology not only allows the machines to identify defects covered by skin, but to also differentiate between defects and peels. (The latter were previously deemed defects under standard inspection methods.)
Minding your peas and carrots
Parent company Key Technology has likewise tailored its Tegra line of inspection systems to very specific food applications, such as potato products, nuts, snacks and cereals, confections, fabricated foods and coffee beans. Members of the Tegra line typically feature a "Tilted X" camera viewing system that can simultaneously see the right and left sides, as well as the tops and bottoms of products. They also feature EISA/ISA/PCI bus hardware compatibility and Windows NT for flexible networking; a stainless-steel mesh catenary belt employing centrifugal stabilization for stable product feed and a tighter product launch trajectory; and a stainless-steel enclosure. One of the newer Tegra products -- named Operating System 2.0 -- features a divided belt infeed that allows for complex functions such as the simultaneous run of two similarly-sized products, or runs of two product streams with different sort criteria. Hence, processors can optimize production capacity by sorting products such as peas and diced carrots at the same time. Alternatively, the reject stream from the first sort pass can be automatically conveyed back to the Tegra onto the second half of the system for a second sort, providing true three-way sorting.
Key Technology has also customized the Tegra line to accommodate different-sized processing operations. The Tegra 7755, for example, is a compact optical inspection system designed for moderate production capacity requirements. The system features top and bottom trichromatic cameras that view product in-air from all sides as the product is launched from the metal mesh catenary belt. Like other members of the Tegra line, the 7755 employs object-specific technology that recognizes not only color and size, but also shape. The system's ejector, comprised of 256 air jets spaced only 6 mm apart, makes it easy to eliminate defects among small products like peas.
The problem with poultry
In the area of poultry, the agricultural branch of Atlanta-based Georgia Tech Research Institute is working to develop better machine vision technologies for detecting bone fragments and cartilage in deboned poultry products, as well as defects for the purpose of grading. The latter project is rooted in research that began in the 1980s, when members of the Georgia Tech team first began investigating machine vision technology as a means of sorting birds by size, combining their own programs with commercial equipment in order to do so. Next they developed a system to identify cut-up parts on a moving conveyor for sorting.After these applications became commercially available, the researchers began gathering and evaluating images of downgraded birds to develop programs that used color and texture information to automatically identify defects. By using the vision system for inputs emulating traditional plant operational modes, they started working with techniques capable of implementing flexible decision-making, such as fuzzy logic, which essentially empowers computers to make decisions like humans. Next, they conducted a 9-month trial at a Tyson Chicken plant in Cumming, Ga., where a sorter programmed to identify bruises, missing parts and other anomalies was located downstream of the plant's chill operation. "What we found was that the correlation between fuzzy logic-based evaluations and human evaluations wasn't that high,'' reported Wayne Daley, Georgia Tech's lead researcher on the project. "On the other hand, we found that the correlation from one human evaluation to the other wasn't that high either." In fact, the Tyson application yielded an impressive 80 percent correlation between the fuzzy logic and human-based evaluations. Now, the question is how to make the system adaptable to varying plant conditions, Daley said.
In the meantime, researchers at Georgia Tech have met with greater frustration on the bone and cartilage project, which focuses on refining X-ray technology in order to improve the accuracy of screening deboned product. The problem is that many current X-ray systems produce too many false positives to be cost effective in deboning applications, while modified systems often require special product handling (e.g. compression and water immersion) that can compromise product quality. The false readings are due in part to the varying thickness of chicken. Hence, systems calibrated to scan 1-in.-thick pieces may barely penetrate thicker portions of the sample and provide inaccurate readings of thinner portions. Factor in the motion of the poultry conveyor, which can result in blurred or distorted images, and the potential for false readings is even greater. In addition, most X-ray systems tend to miss cartilage altogether.
According to Wyvill, researchers at Georgia Tech have focused on using X-ray scatter to improve the accuracy of readings. The underlying principle is that items denser than meat block the penetration of X-rays more than meat does, causing some of the blocked energy to scatter. Since scatter patterns vary according to the make-up of the blocking element, researchers have tried identifying elements present in an object by studying unique scatter patterns. The thinking is that a bone fragment may elude the X-ray screen, but still produce a scatter response indicating its presence. So far, so good. However researchers have hit a brick wall in the last year. According to Wyvill, interference in the direct X-ray field has made it difficult to achieve consistent readings in the indirect, or scatter fields. "It has to do with the limitations of sensing technology, which has to be sensitive, directional and angular all at once" he explained. As a result, the project is currently on hold. "But we'll be back to pick it up," Wyvill said.
In the meantime, at least one supplier has begun marketing an X-ray inspection system specifically intended for inspection of deboned chicken and fish. "We're the only ones with machines located behind deboning lines," said Ken Libby, vice president of sales with Spectral Fusion Technologies, Ltd., whose Bonescan system employs fuzzy logic software to help distinguish actual bone matter from folds and other anomalies that register as dark areas in a scanned image. According to Libby, the system has up to a 98 percent detection rate with only a 3 percent false detection rate. By comparision, conventional X-ray inspections have up to a 40 percent false detection rate in bone detection applications involving poultry, Libby observed. But he also acknowledged that superior detection capabilities come at a price and that, accordingly, the Bonescan sells for as much as three times that of more conventional X-ray imaging systems.
Sidebar
Processor peppers its line with color sorting technology
San Martin, Calif.-based Nature Quality, a processor of bell peppers, recently improved the efficiency of its sorting operation by incorporating a color sorting machine into its line.Nature Quality processes bell peppers from all over the Western states and sells them as an ingredient to industrial food processors. Their lines include washing, coring and slicing of whole peppers. The peppers are then blanched and frozen.
In the past, the Nature Quality employed as many as 22 hand operators to sort and select product following the blanching operation. "The product at this stage is very wet and sticky," said Karen Ash, general manager at San Martin. As a result, she explained, hand operators had a difficult time working efficiently. Further, it was difficult for the company to manage its labor for such a seasonal and irregular crop. "The cost factor was too high," Ash recalled.
Hence, Nature Quality decided to see if a color sorting machine would reduce its hand selection costs. Although the company was familiar with camera technology for vegetable inspection, it was concerned about how the equipment would fare in the plant's wet and steamy environment.
The company eventually hooked up with, Sortex Inc, of Newark, Calif, whose Niagra Colour Sorter had met with considerable success in European fruit and vegetable applications. The product employs Trichromatic -- two visible and one infrared sensor -- optical technology, allowing it to sort according to visible and near infrared color wavelengths. The product is equipped with a conveyor inclined at 60 degrees to the horizontal, so that bottom cameras can view a product from either side without becoming covered in product. "The problem of lower cameras on traditional color sorters being blinded by product splatter is virtually eliminated," said Sortex spokesman Mike Evans. The sorter also features "open hygienic construction" that allows for easy cleaning in both wet and dry processing environments. The system's "SmartEject" feature consists of ejectors located every _ inch across the width of the machine.
As a result of these features, Nature Quality elected to install a single Niagra unit on a performance trial basis. "We worked with Niagra for a period of five to six months," Ash said. "There were a number of issues [we needed to address] to achieve our standards without [incurring] too much product loss. Above all, we were able to substantially reduce our inspection costs."
Nature Quality eventually purchased the sorter, and the processor and Sortex continue to work together. New simultaneous shape sorting technology available from Sortex may have some applications for peppers.