Vision technology heats up

When most people consider trends in technology, they think of faster, cheaper, smaller, and easier-to-use products. Those attributes apply to machine-vision products, too. But both the need to track products and the advent of inexpensive infrared sensors are signaling other trends in the use of vision technologies.

Many vision-system suppliers have seen increasing uses of machine-readable information, such as bar codes and Data Matrix marks, to track components and packages. Industries that manufacture semiconductors, pharmaceuticals, medical equipment, and aerospace components all rely heavily on the ability to track products, starting with incoming parts and ending with delivery to customers. Various bar-code formats have met this need, but emphasis now focuses on the checkerboard-like Data Matrix mark that can carry as many as 2335 bytes.

Figure 1. A laser system can etch Data Matrix marks on flat circuit boards or on curved surfaces for product or assembly tracking. One mark can encode as many as 2335 bytes. Courtesy of Trumpf.

Manufacturers also want to trace defects back to their sources so they can better evaluate and control production processes. "Take a semiconductor manufacturer, for example," says Gilbert Chiang, a product-marketing manager at Cognex. "They have valuable dies on a wafer, so they want to track each one. And tracking specific defects back to positions on a wafer can help fab operators reduce processing errors."

There's another reason to implement a tracking system based on machine vision: A system that "reads" information about a component or an assembly will make fewer manufacturing errors. "When building a motherboard, a computer must match the right processor with the proper PCB," says John Agapakis, a senior VP at RVSI Acuity CiMatrix. "Put in the wrong processor, and you get trouble." If a manufacturer uses a Data Matrix mark on chips, boards, and assemblies, a vision system can distinguish between product types and revisions.

The need for traceability goes beyond manufacturers. Recently, the US Department of Defense (DoD) mandated the use of unique Data Matrix marks on all mission-critical or serialized equipment, as well as on all parts or equipment that cost more than $5000. Vision-system manufacturers should experience increasing demand for hardware that reads Data Matrix marks as DoD suppliers strive to meet the department's requirements. (Most vision software already include capabilities to read bar codes and Data Matrix marks.) "We invented the Data Matrix code and we put it in the public domain," says RVSI's Agapakis. "It's great to see its progress."

An analysis of machine-vision trends requires a look at the infrared (IR) end of the spectrum. Companies have sold high-end IR cameras for years, but at $50,000 to $60,000 per camera, the price has discouraged all but those with deep pockets. Now, though, engineers can buy off-the-shelf IR imaging equipment through suppliers such as Edmund Industrial Optics. The cost for a modest IR camera with a resolution of 160x120 pixels starts at about $10,000, and the cost of a small system starts at $15,000. (Higher-end cameras offer arrays of 320x240 pixels.)

Figure 2 . Infrared images, or thermographs, of a) a normally operating motor and b) a motor running too hot can help designers determine the cause and amount of overheating. Courtesy of Indigo Systems.

A 160x120-pixel IR camera takes data from more than 19,000 points, so it can view large areas and create a thermograph—a contour map of temperatures. If a camera's resolution isn't fine enough for an entire PCB packed with tiny components, a technician can concentrate on sections of the board that generate heat or contain heat-sensitive components.

Some engineers have already successfully applied IR vision systems. Jason Mulliner, vision marketing manager at National Instruments, has seen auto manufacturers using IR imaging during production to check heating systems, heated seats, and IrDA communication ports. Other engineers should also start to embrace low-cost IR imaging systems, because new designs usually require careful analysis of heat dissipation and the effect of heat on components and assemblies. But first, these engineers must understand what an IR system can do.

Tom Scanlon, a VP of sales and marketing at FLIR Systems says, "When engineers visit us at shows, they're amazed by our demonstrations. They have no idea what IR inspection can do, or they think of it as an esoteric technique. They didn't know they could buy a small, inexpensive system."

Wayne Ruddock, director of Advanced Infrared Resources and a consultant to Indigo Systems, says, "Many people don't understand that an IR imaging system doesn't look at temperature. It looks at radiated energy, and that energy can come from an object, or it can come from heat reflected from the object's surface."

To capture the emerging market, IR-system suppliers must redouble their efforts to educate users. Users may not understand, for example, why they can't get an accurate measurement from a shiny metal surface. "If they put some black tape on that metal and gather data from the tape's surface, they'll get good readings," adds FLIR's Scanlon. "But they need to know tricks like that, and the physics behind them, before they can effectively employ IR-vision technology."

Any review of trends should include a look at price and performance, which still rate high in importance with suppliers and customers. The recent dismal market for vision systems in the electronics industry has forced many suppliers to re-evaluate their markets and their plans.

"Everyone's coming out of a slow economy," says Mike Roberts, the general manager for machine vision and imaging at Data Translation. "For now, there's no pent-up demand and no reason for manufacturers to quickly ramp up or add capacity. They just want to keep costs down." If vision systems cost too much, people won't buy them. Sometimes it's less expensive to accept some defects than to add a machine-vision system.

Figure 3. Algorithms that process images froma  high-speed production line must complete tasks quickly to prevent stalling the line. Deterministic algorithms guarantee execution with a preset time limit. Courtesy of National Instruments.

Vendors also face increasingly complex technology challenges. Gilbert Chiang of Cognex says, "You'd think images would get better as people learn more about vision, but instead, images get worse, because users have more demanding applications. We continually see images in which it's difficult to locate a pattern due to noise or varying light conditions, or it's difficult to find tiny defects people want to detect. Suppliers must continually develop and refine tools to keep up with customers' latest imaging needs."

Some vision-system users also demand faster processing speeds and higher image resolutions. Chiang sees these demands coming mainly from semiconductor capital-equipment suppliers who aim to reduce production cycle times. They also want faster image acquisition to help speed production and higher image resolution to improve part-alignment accuracy.

DVT recently released a vision sensor that incorporates a digital signal processor (DSP). According to Dr. Phil Heil, the applied engineering manager at DVT, the DSP's pipeline architecture provides a 3X to 8X speed advantage over earlier DVT products. The higher-speed sensors find use in new applications, such as monitoring components on a fast-moving production line.

Users also demand faster setup times, so many vendors now deliver equipment that includes lights, a camera, lenses, and software, all in a ready-to-use system. "Users inspecting solder balls, for example, shouldn't have to worry about concepts such as blob analysis or morphology," says RVSI's Agapakis. "Instead, they want a system that lets them specify what they want to measure or detect—missing balls, extra balls, pitch, shape, and so on. That way, they can get useful results right away." But he adds, "Users need to know that developing a complete vision system takes more than choosing five menus and filling in a few dialog boxes."

The algorithms behind those menus and dialog boxes play a starring role, although they often operate behind the scenes. But even the best machine-vision algorithms can stretch processing times unreasonably when given an unusual image to process. That can occur when, say, a badly damaged PCB undergoes inspection. The PCB's image doesn't come close to anything the vision system saw during training. So, a pattern-matching operation can seem to go on forever.

If image-processing times stretch out, a vision system may not keep up with a fast-moving production line. Jason Mulliner of National Instruments says, "Vision systems need deterministic algorithms so developers know the maximum processing time each one requires. We set a time limit for each vision algorithm, so if it doesn't finish its task in time, it generates an interrupt."

In effect, the algorithm says, "I didn't execute in the right amount of time." Then the software can quickly fail the product so an operator can check it further. In a case like that, says Mulliner, even though the algorithm didn't finish its task, the computer still rejected the part and went on to the next inspection without breaking its stride.

No matter what trends seem important now, customers' shifting needs and new technologies will, as always, set the course for the future. There's no guarantee that today's trends will survive beyond the next market surge or dip.

For more information on machine vision, visit www.tmworld.com/ins.