But how do you know that the correct badge has been placed on each truck? That's a question that assembly line workers at Ford's Louisville, KY, plant must ask themselves several Arial a day. Those workers assemble Ford's Heavy-Duty pickups--the F-250 through F-550. Recently, they have become more confident in their ability to match badges with vehicles, thanks to the installation of a vision-equipped robot that verifies that each truck has the correct badge before it leaves the final assembly area.
"There are at least 12 different badges that can go on a truck, and the vehicles all look pretty much the same when they are coming down the assembly line," says Frank Maslar, a Ford automation technology specialist. "No manual verification process is 100% effective," he adds. "Studies have shown that when people are asked to sort black and white ping pong balls by color that they only get that right 85% of the time."
Numbers like that also explain why machine vision systems are becoming common fixtures in all types of manufacturing facilities. The Automated Imaging Association (AIA, Ann Arbor, MI) says that worldwide machine vision equipment sales grew by about 20% in 2000 to $6 billion, with roughly 34% of those purchases made by North American companies.
Jeffrey Burnstein, AIA executive director, says that sales were slow during the first half of 2001. He believes the slowdown is temporary, and attributes the decrease more to a general decline in spending on capital equipment than to a decline in demand for vision systems. In fact, the AIA stands by its projections that worldwide spending on vision systems will reach $12 billion by 2005.
Reasons for that optimism include technological advances that are making vision systems less expensive even as they become more powerful. "This is creating opportunities for the use of machine vision in a wider range of industries," Burnstein says.
Automakers and their suppliers have long been regular users of machine vision for tasks ranging from guiding robots for welding and pick-and-place assembly tasks to online inspection of components and subassemblies. And as Ford's badge-verification system shows, the auto industry continues to find new uses for this technology. (See "Pickup Badge Verification at Ford," on p. 52.) But Burnstein and other industry experts say companies that make semiconductors, medical equipment and other products that use electronic components are now the driving forces in the vision system market.
"Many of these companies are finding that machine vision allows them to do things they simply couldn't do before," says Larry Chin, an executive vice president with Intelligent Automation Systems (Cambridge, MA), a machine-vision systems integration firm. "We have clients in the telecommunications industry using machine vision to move optical fibers a distance of 1 micron or less. You can't do that by hand."
No more bowls
Peterson Manufacturing (Kansas City, MO), which makes lights for tractor-trailers and emergency vehicles, is able to do without some expensive factory equipment by using a vision-equipped robot to load components onto its assembly lines. The robot, which Peterson purchased from Seiko Instruments (Torrance, CA), picks up parts from a bin and shows them to the vision system, which determines whether the parts are facing the right direction for placement on the assembly line. If the part is not facing the right direction, the vision system tells the robot which way to turn it before placing it on a conveyor belt leading to the assembly line. "This eliminates the need to have bowl feeders, which are expensive pieces of equipment designed for handling specific components," says Steven Ham, Peterson Manufacturing's plant manager.
Philips Oral Health Care (Snoqualmie, WA), manufacturer of the Sonicare electric toothbrush, uses Seiko's vision-equipped robots in half of the dozen steps involved in assembling the 50,000 brush heads it builds daily. The vision systems handle tasks ranging from ensuring that parts are turned the right way for insertion into a production machine to checking that specific components have been inserted as the brush heads move along the assembly line. The makeup of the brush head--a 5-inch long package of wires, microchips and magnets--would make this tedious work for a human.
"Some of these processes previously were not done at all," says Devin Nelson, a Philips automation engineer. "Others were prone to problems because they were being done manually, and people are subject to fatigue. We installed the vision systems because we needed repeatability in those processes."
In general, machine vision systems are used in assembly for three types of tasks--guidance, inspection and process verification. In a guidance application, the vision system serves as the eyes for a production machine that has to find a particular part and place it in a specific location. Examples of this are pick-and-place operations such as placing electronic components on printed circuit boards.
Inspection operations, which typically take place at the end of an assembly process, consist of using machine vision to scan for defects. Process verification involves making sure that a particular step, such as the insertion of a component, has been performed before a product moves to the next production step.
Machine vision is based on the same technology that is used in optical character readers or the scanners that convert paper documents into electronic files. The technology relies on a camera to take a picture of a designated object and then converts that image into a stream of digital information that can be read by a computer. Machine vision was born when scientists began connecting these character-reading devices to machine tools.
Cognex Corp. (Natick, MA), was a pioneer in the machine vision field. It sold its first system in the early 1980s to a typewriter manufacturer that used it to check that the keys on its typewriters were in the correct positions. Today, more than 100 companies sell various types of machine vision systems. In addition to Cognex, leading vision system vendors include Coreco Imaging Inc. (Bedford, MA), PPT Vision Inc. (Minneapolis), and DVT Corp. (Norcross, GA). Robotics manufacturers--including Adept Technology Inc. (San Jose, CA), Fanuc Robotics (Rochester, MI) and Seiko Instruments--also have joined the field by integrating vision systems with their robots.
Most contemporary machine vision systems contain the same basic components, starting with a camera, a light source, a power supply and a personal computer. Unless the system employs a digital camera, the PC must have a special board called a frame grabber.
The frame grabber converts the images captured by the camera to digital signals that can be read by the computer. The other system components include a set of cables to connect the camera to the power supply and the PC, and software for running the entire system.
To deploy most vision systems in a manufacturing setting, the PC must be connected to the machine that performs the process that the vision system will either guide or monitor, as well as to the plant's local area network (LAN).
Watching the process
Machine vision systems have historically been used for inspection applications, largely because these were the easiest kind of tasks to set up. But some industry observers say that is changing.
"The potential uses for machine vision literally are unlimited," says John Little, a director with Vision 1 (Bozeman, MT), a systems integrator. "But we are seeing a shift away from using vision systems to check things that are already made. More companies are now using them to monitor products as they are making them."
The growth of in-process vision applications has coincided with technological advances that include more user-friendly vision software and the advent of the small, relatively inexpensive vision systems referred to as smart cameras or vision sensors, which come with built-in computers.
"Almost all vision software is now Windows-based," says Little. "The solutions have become more elegant, as the developers like to say. They offer graphical interfaces that make it easy for users to configure the systems to perform their particular applications without having to understand the underlying technology."
In the early days of machine vision, almost all systems were custom configured by system integrators who typically bought standard cameras and computers from regular retail outlets and modified them to work with a particular manu-facturing process. These early systems also required custom software programs written by highly skilled programmers.
Today, Cognex offers software with its In-Sight line of vision sensors that allows users to configure an application by entering data into a spreadsheet. Fanuc Robotics sells a software package called visLOC that makes setting up a vision system much like installing a software program with a Microsoft Windows wizard. Ford uses this package to run its badge-verification system. Maslar says it took three days to install this system, which also employs a Fanuc robot and vision system hardware--specifically a camera and frame-grabber--from Cognex.
In the past, it could have taken three months or longer to custom-build a similar system. Most users say the most time-consuming part of implementing contemporary vision systems is designing the process that you want the system to perform; setting up and programming the vision system is now the easy part.
Industry sources point out that continuing price declines and the decreased complexity of today's machine vision systems are making the technology a practical option for a growing number of assembly line applications. Traditional systems configured from multiple components with connections to a corporate LAN have declined in cost from $40,000 or more a decade ago to around $20,000 today. The cost of installing systems has dropped significantly as well, because it no longer requires modifying hardware and writing custom software programs.
For some applications, the development of low-cost smart cameras and vision sensors can lower machine vision costs even further, while giving users more flexibility in the way they use the technology. These systems, which typically cost about $5,000, can be placed almost anywhere on a production line. They normally are used for a single task, such as checking for the presence of a part before a product moves to the next station.
"Placing a few smart cameras at various points in a production line is a cost-effective way of creating a distributed intelligence system in your plant," says Nello Zeuch, president of the consulting firm Vision Systems International (Yardley, PA). "The ultimate value is an overall improvement in quality because you can spot problems as they are happening, and implement process improvements. For instance, you can establish a rule that if you see failures in a process, you stop that process and take some corrective action."
Currently, most users track the number of failures that their vision systems detect, but few have gotten to the point of using that information to trigger immediate changes to their production processes.
"Most companies use vision systems as a 'Saint Peter' device," says Little. "They let the good parts pass through and they send the bad ones to another place."
In reality, most of parts rejected by vision systems during an assembly operation are not bad at all. They simply have a component missing or in the wrong position. The normal remedy is to send the part back to the station where that component is inserted, where the problem is fixed, and the part is again checked by the vision system.
That is exactly what happens in the Ogden, UT, plant where Autoliv Automotive Safety Products assembles airbags. Autoliv employs a Cognex vision system to verify that two key components of the part that inflates the airbag are installed correctly.
Each bag requires two of these parts, called squibs. They look slightly different, and they must be inserted in specific chambers inside the bag. Production workers insert the squibs into their respective chambers. A machine then completes the process by pressing the squibs into place. But the press cannot do its job until it gets clearance from the vision system.
The vision system goes into action when a worker flips a switch, sending a signal that the squibs are in place. The vision system's camera, which is located just above the assembly line, takes a picture of the squibs and transmits it to the system's computer, which checks to see if both squibs are present and in the correct chambers. If everything checks out, the system sends a pass sig-nal to a programmable logic controller (PLC) connected to the press, and the PLC tells the press to punch the squibs into place.
Where's that squib?
If a squib is not in place, the vision system tells the PLC not to activate the press. As this happens, the worker views the process on a monitor. When the system detects a problem, the problem area is highlighted on the worker's screen. Once the worker addresses the problem, he or she flips the switch to start the verification process again.
"We previously used pass-through sensors for this type of application, but the vision system is more accurate," says Jonathan Sommer, an Autoliv process engineer. Pass-through sensors rely on two pieces of optical fiber that pass light to one another across an assembly line. As the product being inspected passes the sensor, a break in the beam of light indicates that the parts the sensor is looking for are present.
"Once a squib is inserted into a chamber, it can move about a sixteenth of an inch before it is pressed into place," Sommer says. "We had to constantly align the pass-through sensor to accommodate for that movement. The camera in the Cognex system can actually track the part as it moves and still capture the correct image."
Marlow Industries, a Dallas-based manufacturer of thermoelectric cooling devices, is considering expanding its use of machine vision as a means of improving quality control. Marlow's devices, which consist of two boards containing electronic circuits that are mated together in precise fashion, control the temperature in a number of products, from refrigerated picnic boxes to telecommunications networking equipment.
Marlow currently uses vision-equipped Seiko robots to mate its circuit boards. But Michael Gilley, a production engineer, envisions having vision systems verify every step in Marlow's manufac-turing process, which would, in effect, mean the return of a 100% product inspection program.
"The pressure to produce quality products throughout the economy is bringing 100% inspection back," Gilley says. "With machine vision, we could inspect products throughout the assembly process. That would mean each product had been 100% inspected once it reached the end of the line. Then we could have people pull samples for an end-of-process inspection.
"That would give us the best of both worlds," Gilley concludes. "The sample inspection would catch some of the subjective quality measures that are beyond a machine's comprehension. Meanwhile, the machine inspections would allow us to catch problems early in the process, when they are easier, and less costly, to fix."
PICKUP BADGE VERIFICATION AT FORD
One of the newest applications of machine vision at Ford Motor Co. is a system used at Ford's Louisville, KY, plant to verify that the correct identification "badge" is placed on the fender panels of pickup trucks moving down the assembly line.
The truck badge verification process starts with workers looking at electronic sign boards that hang above their workcells to identify which truck model and trim series--the XL, XLT or XLT Lariat--is coming down the line at any given moment. This information comes from Ford's production management software system, which tracks each vehicle as it moves through various stages of assembly.
The badges are stored in bins beside the assembly line. When a truck enters the final assembly area, a red light on top of the bin containing the appropriate badges for that truck comes on. A worker then retrieves a badge from that bin and places it on the truck.
The truck then moves to the robotic verification station. As the vehicle enters this station, a bar code containing information about the truck's configuration is scanned, and that information is transferred to the software program that workers use to operate the robot workcell. The cell-control program then instructs the robot to verify that the truck has the correct badge. To do this, the robot moves the vision system's camera above the badge. The camera takes a picture of the badge and transfers it to the vision system's software program, which compares the picture to an already-verified image of the correct badge that is stored in a database.
If it is not the correct badge, a message is sent to the worker operating the cell. The worker enters this information into the plant's quality system, and sends the truck back to the beginning of the final assembly area, where the verification process starts again. If the truck has the correct badge, the worker releases it from final assembly and enters that fact into the quality system.
This system was initially installed for a six-month pilot program that involved checking badges on only one side of the truck. The system is now ready to begin inspecting all badges on all trucks, says Frank Maslar, a Ford automation technology specialist.
Masler says that the system will save Ford money by eliminating the need to send trucks back through the line for new badges, but he declines to be specific on the savings. Suffice it to say, Masler notes, "if it wasn't cost-effective, we wouldn't be doing it."