Specifying a vision system

READ OTHER APRIL ARTICLES:  Contents, April 2007 Additional information to help you specify cameras, lenses, lighting, and software for machine-vision systems.

Vision systems can be configured in various ways, and image data can be processed by a PC, a programmable logic controller (PLC), a smart camera, or a vision system itself (Figure 1). In addition to choosing a camera and lens with the necessary resolution, you also need to light your setup properly and select an interface for transferring image data. The best path to choosing the right components begins with an evaluation of your product and your production line.

Figure 1. Engineers can put together vision systems in a variety of ways, four of which appear here in simplified form. A complete system may combine these approaches to inspect products at several points along a manufacturing line.

If a vision system must make basic gauging measurements or determine the presence or absence of components, engineers can quickly agree on what dimensions or characteristics distinguish a pass or a fail condition. “But when vision systems must find small defects, engineers require a database of accurate images that show good, bad, and marginal products,” explained Steve Cruickshank, principal product marketing manager for PC vision products at Cognex. “From an inspection standpoint, that’s half the battle—getting an agreement about what good or bad products look like and what defects make a product bad. If engineers plan to identify marginal defects, they must specify which of these 'defects’ they can accept and which ones they must reject.”

“Most of the time, people start with a few sample images, but the more images they can provide, the better they can identify problem products,” said Cruickshank. “We ask a customer’s engineers to review the images and visually separate the good products from the bad. Before we can assist customers, we need a fixed set of test images.”

Agapakis of Siemens noted that images provide a good start. Often, though, vision-equipment vendors and vision-system integrators must work closely with human inspectors to quantify their knowledge. That type of information, gained from years of experience, can prove critical for distinguishing good, bad, and marginal products.

Most vendors of vision equipment can perform feasibility studies that help engineers determine whether they can inspect for each characteristic or defect they specify. These studies usually involve sending a vendor samples of good, bad, and marginal products. Depending on the vision products or systems a vendor offers, the report from a study may recommend specific camera types, light sources, controllers, mounting configurations, and so on. Vendors also may loan equipment to engineers so they can perform tests in their own lab or on their production line.

A vision system’s camera and lights must fit within existing space on a production line. In this instance, the system supplies a ring light around the camera to provide uniform direct illumination. Courtesy of Cognex.

“The cost for an integrator’s study can run from $5000 to $15,000,” said Greg Hollows, vision integration partners coordinator at Edmund Optics. “That’s a lot to spend, particularly if the results show you cannot achieve the results you expect. But you must ask yourself, 'Should I pay for a complete feasibility study, or should I buy vision equipment that may not fit on my production line or that may not work?’” Even if initial tests by an integrator fail, the integrator’s experience may let it suggest alternative approaches.

Hollows recommends engineers start vision-system planning concurrent with the design of production equipment. “I get calls from people who want to put a vision system on an existing production line or in production machinery well along in design. In some cases, the machinery has only small openings or constrained mounting arrangements that make it difficult or impossible to include a camera, lens, and lights.”

Inspection of a PCB, for example, may require special lights that illuminate a PCB from several angles and require considerable space. “Engineers must leave space in their production machinery to accommodate vision components,” said Hollows. “You just can’t add them as an afterthought.”

According to Cruickshank at Cognex, one customer defined an inspection task that required a camera mounted a significant distance from products undergoing inspection. The new vision system detected failures that operators could not trace to a specific defect, so they needed to find the cause of the false rejects. “They looked at the images and saw during the inspections someone had put their hand between the camera and the products. No matter how good the vision system, software tools can’t perform an inspection if someone blocks a part.”

As engineers evaluate the requirements of their vision system, they must consider the type of camera and lens an application demands. “Often engineers want to purchase a camera right away,” said Matt Slaughter, vision product engineer at National Instruments. “But camera selection should come later in the development process. Your system requirements should determine the type of camera you choose, not the other way around.”

a)
b)
a) A Data Matrix symbol formed on a textured surface appears as a gray blur when illuminated with a ring light. b) The same symbol produced a high-contrast image when illuminated with a diffuse on-axis light. Courtesy of Siemens Energy & Automation.

Resolution defines how well you can see the smallest objects in an image. Say you have a PCB on which you must inspect 0402 (0.04x0.02-in.) surface-mount components. Your camera should capture at least two or three pixels along the smallest edge to create enough information for your vision software. (Some imaging experts would argue for subpixel techniques that extract edge information from a number of nearby pixels, but I’ll stick to basic concepts here.)

So, assume you need three pixels along a 0.02-in. component edge, or 150 pixels/in. Thus, if you must examine 0402 components across a 6x6-in. PCB, you will need a camera with at least a 900x900-pixel CCD sensor—well within the capabilities of an SXGA-resolution camera with a 1240x1024-pixel sensor. But if you have a larger PCB, say 12x18 in., you may need to use several area cameras, a higher-resolution megapixel area camera, or a motion system that moves the large PCB across one camera’s field of view. If you plan to examine the quality of solder joints rather than just locate component edges, you will need much higher resolution to bring out joint details.

“Some people may need one image every few seconds, while others may need to inspect 100 to 1000 images per second,” said NI’s Slaughter. The total amount of image data, obtained by multiplying the number of pixels per image by the number of images per second, determines a minimum bandwidth.” (Keep in mind one pixel “creates” from 8 to 12 bits of monochrome information or 24 bits or more of color information.)

Bandwidth requirements determine the interface your system will use to transfer image information to a computer; options include RS-170, USB, FireWire, Camera Link, Ethernet, and Gigabit Ethernet (GigE). If bandwidth requirements exceed the capability of an interface or of a host PC, consider using smart cameras or vision sensors that can perform analysis and metrology functions prior to sending image information to a host computer. Also, you could give some inspection tasks to a small vision PLC that will manage cameras, process images, and report results to a larger computer.

When engineers evaluate a camera interface, they must balance the cost of the cables against performance requirements. Although Camera Link cables offer a high bandwidth—5.44 Gbps—they are expensive and span only about a 10-m distance. (Fiber-optic links can extend the distance: A 100-ft link costs about $1800.) In contrast, a camera that operates over a GigE connection can use standard Cat-5e or Cat-6 cable that costs under $0.20/ft, and a GigE connection can run at least 100 ft.

“Due to the nature of GigE communications, remote triggering presents a synchronization problem,” noted Slaughter. “So, engineers might decide to trigger and synchronize a remote camera via the camera cables. Typically, though, a remote GigE camera would rely on a small nearby proximity sensor to sense an object and trigger the camera and the lights.” It proves almost impossible to synchronize a network of GigE cameras over a nondeterministic GigE network.

Like cameras and their interfaces, light sources require thorough evaluation, too. Mike Romano, laboratory manager at Advanced Illumination, commented that engineers must run feasibility studies on lights just as they do for cameras and lenses. “It’s a huge oversight to design lights into a vision system without ensuring they work for the camera you expect to use and the product you must inspect,” he said. “Suppose you have purchased a red-LED light source and you plan to inspect red printing on a label. The spectral response of a CCD camera may not let it detect the red characters. You don’t want to discover that at the last minute.” And improper lighting can decrease contrast so much that engineers can’t extract useful information from an image.

Most companies that supply lights for vision systems offer engineers evaluation or demo units—and advice. “If engineers tells us they want to examine PCBs for missing components, based on our experience we can suggest several types of suitable lights,” said Romano. “Or the engineers can send us good, bad, and marginal PCBs so we can make a preliminary evaluation, report our findings, and send back test images. The engineers can run the images through their software to ensure it finds what they want—characters, components, colors, and so on.” To start the test process, engineers fill out a questionnaire on the Advanced Illumination Web site and send it to the company along with sample products.

Local conditions also affect light choices. “Some engineers use infrared [IR] illumination in places where bright or flashing lights might bother people,” said Romano. But IR light has a drawback: It can diminish color contrast to the point that a monochrome camera “sees” red and green or red and yellow as the same color.

Figure 2. A ring light attached to a camera illuminates the bottom of a bottle so an inspection system can check for cracks or other defects. Courtesy of Advanced Illumination.

Ambient light may play an unexpected role in inspections. Cruickshank of Cognex explained, “In one application, sunlight from a skylight interfered with the inspections as the sun moved throughout the day and the inspection system tried to deal with the effect of changing light conditions.” Engineers can establish a constant light intensity, perhaps in an enclosure, to overcome problems with changing ambient light.

Some applications require engineers to reproduce a vision system for use on several production lines. They can work out any problems or bugs on one line and then transfer their knowledge and experience to the others. “Engineers must ensure their system is robust and not fragile,” remarked Cruickshank. “You don’t want a system to 'just work’ on the first production and then fail on the next line, even though you bought the same camera, the same lens, and so on. Subtle differences can cause a duplicate system to fail.”

Cruickshank recommends engineers slightly vary the lighting, defocus the camera, move the camera, change exposure time, and so on to determine if and how the changes affect performance. Use good engineering techniques to ensure your system doesn’t operate on the “edge” of equipment tolerances. In the end, engineers may have to adjust their inspection priorities. He explained, “You want a vision system that reliably performs a few tasks all the time rather than a system that works half the time and tries to do too much.”