1 May 2005

Vision Precision

By Ellen Fussell

Speed, control see eye to eye in discrete processing.

In the last five years, vision has really blossomed in terms of discrete manufacturing, especially in packaging quality control. Vision systems have historically been prominent in the semiconductor wafer manufacturing realm, said National Instruments' Kyle Voosen, vision product manager in Austin, Texas. And even today, half of vision technology remains in the semiconductors and printed circuit board industry.

"It worked well for those applications, and they could afford it," Voosen said. But now that it's come down in cost and because the PCs have gotten faster, "it's really brought vision into some of these other areas that wouldn't think of it-packaging in general."

Vision systems can make sure a label is on correctly. It's a simple area where it's not vital, but if the cost is low enough, why not have vision make sure the package is aesthetically correct?

Take pencil inspection, Voosen said. "When pencils are flying down the assembly line, the vision system has to make sure the wood inside each pencil matches, the lead is in the proper place, and the color is correct. But in the past, they had 120 people inspecting every pencil. Now they have one camera and one computer. If one's bad, it shoots it with air to knock it offline."

Vision bus

Vision systems have gotten so fast you can do amazing algorithms, "things you couldn't do five years ago, with lower costs and faster processors," Voosen said. Years ago, they had the same algorithms in text books, but they took longer to run. So as processors keep speeding up, vision gets that much more powerful.

Voosen said the bottleneck has been the PCI bus in the computer. "Today, all the buzz is peripheral computer interface (PCI) technology. It's just a PC technology-the way you plug in cards to your PC," he said. The new PCI technology helps acquire data faster; it's a bus inside the PC, a way to get data from the camera to the processor, to and from a plug-in board to the processor in the PC itself. With a new PCI bus technology, you can acquire data a lot faster than you could before, he said. "A year ago, I could acquire images at 100 megabytes per second; that's not bad. A typical image is 2 megapixels, and I could acquire 50 per second. But now with PCI bus technology, I can acquire images at 2.8 gigabytes per second; that's 28 times faster."

One of Voosen's customers wanted to improve nozzle sprays for their aerosol cans, making sure the spray was evenly distributed. "When you hit an aerosol button, spray comes out fast. With the PCI bus technology and cameras plugged into the computer, you can acquire that data really fast, 1000 times a second," Voosen said. You can see every single particle in the spray. The application is called particle velocimetry. In testing, you can measure if it's the right plume, concentrated spray, or straight line.

Food and beverage

Vision systems have more to do with the speed of cameras and processes in the control world. "You can actually close the control loops with vision, monitoring flow or filling bottles, and canisters," Voosen said. "Anytime you need vision feedback or any motion control, say assembling cookies, the vision system guides the robot to put the top half on an Oreo."

While using vision to close a PID loop, the most important variable is time. "In PID, you have the three variables, proportion, integration, and derivative, run over and over," Voosen said. "If you can speed it up, it becomes better. In filling a bottle, if you can speed the image up, you can stop filling at exactly the right time. The faster you can acquire images, the more precise you can control a process."

Automotive testing

In an automotive crash test, you might want to coordinate the vision with all the different pieces of data, such as sensors inside the car, accelerometers, strain gauges, and gyroscopes.

Placing a strain gauge on the hand of a crash test dummy could show strain on one thumb to tell testers what was going on in that instant. "If you synchronize the data acquisition with the strain gauge and information coming from the camera, you can see when the strain on the thumb is giving a lot of information," he said. "From that image, you can tell the airbag got caught on the thumb. So now you can analyze that data on a whole new level." Maybe a button on the dash board is in the wrong place and causing the thumb to get caught, whereas before you had no idea because it happened too fast.

Voosen said while the crash test has been around forever, the process used expensive cameras. "You could acquire fast images, but you couldn't correlate the image with the data because there was no synchronization between the strain gage, the thumb, and the camera."

Two dimension vision

Direct part mark identification (DPMI) is another valuable vision application. It's been around for a while, reading letters and numbers on semiconductor wafers. But it's reemerging today in the form of machine readable data matrix codes, moving from human readable characters to machine readable characters.

A data matrix code is a two-dimensional (2D) symbol. One-dimensional (1D) code appears on most consumer products today. While a 1D barcode is typically an inch to an inch and a half long and contains eight characters of information, a data matrix code, or 2D code, could be a quarter inch square and carry 50 characters of information. "One of the key uses of 2D codes today are DPMI applications because they can hold more data in less space, they're more robust, and you can read them with much less contrast than traditional 1D barcodes," said Jamie Pearce, business development specialist for Cognex ID Products in Natick, Mass. "You can damage a 2D code and still read it because the code has built-in data redundancy. You could put a scratch through a data matrix code, and you could still read it. If you did that to a 1D code, it would very likely make the code unreadable."

Reading a 2D code in DPMI applications requires taking a picture of a code that's marked directly on a part and analyzing that image to extract data from the code. The data could tell you the part's production date, the lot number or batch, or the manufacturer's serial number. Within the data matrix code, you can have different information fields with ASCI characters that separate groups of information within the code.

There are a couple of over-arching drivers for why this is a hot topic now, Pearce said. "First, manufacturers want better process control by tracking their products throughout the manufacturing process and their life cycle. Data matrix codes give manufacturers the capability to incorporate a large amount of information about a product in a small space with a code they can read even with directly marked parts with low contrast. Second, they want to comply with traceability requirements because of new regulations in the aerospace industry and automotive industry," Pearce said.

The Department of Defense (DOD) has implemented an initiative to identify any part that's either valued at $5,000 or more or is uniquely tracked. "That means every part the military receives meeting that criterion will need a mark with a data matrix code and every DOD supplier that change affects," Pearce said. Quite a few auto manufacturers now track and trace parts from the beginning of life through end use as well.

"On the technology side, from a machine vision standpoint, we've been able to read very low contrast and difficult-to-read data matrix codes with PC-based vision systems for some time," Pearce said. "Users would need a computer with an integrated frame grabber board and camera to get the results they wanted." Because of advances in the speed of today's microprocessors, and the corresponding decrease in cost, you can now read data matrix codes with handheld and fixed-mount readers that can fit in the palm of your hand without needing a PC.

By tracking components and assemblies, manufacturers can keep tighter control of their inventory. If there's a warranty or recall on a specific batch of products manufactured during a specific time frame at a specific plant, they can contain the recall better. "So instead of recalling 100,000 cars, they may have to only recall 10,000," he said. Within the process, manufacturers can identify a problem before they've made 1000 parts with the same error. "They can quickly detect problems and stop the line to contain the problem and reduce scrap. DPMI also helps error-proof the process," he said. DPMI applications especially use data matrix codes today since they allow manufacturers to mark a machine-readable code directly with human-readable tracked information. From a crank shaft, transmission, or pinion for an automobile, to an aircraft turbine blade, you can now track machined parts as long as you create a contrast for the code with different lighting. Different types of lighting create a contrast between the mark and the background surface based on the way it reflects off the mark.

Some marking technologies will create their own contrast, for instance a black mark on a white label. "In direct part marking applications, however, in many cases you need to use different angles or types of lighting to produce the contrast," Pearce said. The contrast allows you to differentiate the printed versus the unprinted cells or the light cells from the dark cells. "When Ford is manufacturing transmissions, they want to track each of their products throughout the manufacturing process to control the flow of products through their plant," he said. "Because of the harsh manufacturing environment and the desire to put a large amount of information on the part, in most cases, a 1D barcode label is not an acceptable alternative."

Automotive manufacturers will etch or dot peen (into metal) each of their components with a 2D data matrix code, which they can read with reading and imaging systems throughout the manufacturing line. Other industries are starting to use DPMI as well, Pearce said. The medical industry uses the technology for joint replacements or surgical instruments. Manufacturers can mark data matrix codes on nearly any type of material. "Once you mark a part, if you can apply the right lighting solution to create contrast for the codes and apply the latest in machine vision object location algorithms to find the code in the image, you can read a data matrix code on almost any part."

Another example of a 2D datamatrix code application is Boston Scientific's use of the codes to identify plastic surgical devices and then verify the code on the device matches the corresponding bar code on the package. "They need to be sure the right device goes in the right bag, so when the surgeon asks for an item, he gets the right one," Pearce said. Manufacturers mark the parts with 2D codes during packaging operation and compare those codes with marks on the bag to prevent errors.

Even 1D codes took years before they reached wide acceptance in the world of tracking products. "Today, however, we're 10 years into the general industry adoption of 2D matrix codes," Pearce said. "So we have now moved past the stage of early adoption and are quickly moving toward general industry adoption."

Direct part marking methods

Dot peening: Automotive and aerospace industries use dot peening by pneumatically or electromechanically striking a carbide-tipped or diamond-tipped stylus against a material's surface. Code readability depends on the quality of the indented dots. Manufacturers control dot size, shape, and spacing by prescribed maintenance of the dot peen marker to monitor the quality of the stylus. In some cases, machining or cleaning are processes manufacturers use to further prepare and process the part surface to improve code readability.

Laser marking gives high speed, consistency, and high precision in the semiconductor, electronics, and medical device industries. The process involves applying heat to a part's surface, causing it to melt, vaporize, or change in some way to produce a mark. A laser can produce round and square modules-usually a square module-and continuous finder pattern for higher density (large data capacity) codes.

Electro-chemical etching (ECE) is a good process to use on components of round surfaces and stress-sensitive parts, such as jet engine components. The process produces a mark caused by oxidation through a stencil impression on surface metal. This happens by sandwiching a stencil between the surface to mark and an electrolyte-soaked pad and passing a low-volt current between the two.

Ink-jet printing provides fast marking and good contrast on moving parts. It works by precisely propelling ink drops to the part surface. Then the fluid forming the ink dot evaporates and leaves a colored die on the part's surface. Manufacturers might need to prepare the part surface because the chemical interaction of the ink to the surface determines the mark's contrast and permanence.