Put vision sensors on the line

Not every automated inspection task requires a complete PC-based machine-vision system. In many cases, a small, stand-alone vision sensor will do the job. But if you think the words "vision sensor" refer to a simple light sensor such as a photodiode, banish that thought. Today's vision sensors do a great deal, from reading encoded information to making quantitative measurements on a product. For the most part, these inexpensive sensors take little time to set up and require little or no programming, although many come with high-level programming tools, should you choose to use them. Even when you think only a large PC-based machine-vision system will solve an inspection problem, consider using vision sensors. Often, you can break down a complex inspection task into several simpler tasks, many of which can use a vision sensor. Initially, the hardware for several vision sensors might cost more than hardware for a single machine-vision system. But in the long run, development, maintenance, and setup of individual sensors may cost less. Sensors require little maintenance, demand no programming, and don't rely on a ruggedized and expensive PC.

Figure 1. Vision sensors include a camera and image-processing hardware and firmware in a small package that easily fits on a production line. Courtesy of DVT.

You can place vision sensors on a production line to acquire and process images of products. Although vision sensors won't measure solder-paste volumes or look at hundreds of PCB component pads to find missing SMT parts, they will match patterns, look for missing connectors, determine distances, detect edges and rotated angles of parts, locate fiducial marks for robotic placement systems, read encoded markings, and so on.

In most cases, an inspection result will produce a simple pass/fail signal that tells a nearby programmable logic controller (PLC) to pass the product or to push it into a reject bin. More sophisticated sensors can transmit data read from a bar-code label, the dimensions measured between two targets on a PCB, and the angle between two components.

Vision sensors also perform an operation called pixel counting. After a sensor acquires a gray-scale image, it applies a series of thresholds to divide the pixels into two categories, black or white. The images in Figure 2 show an acquired gray-scale image and the results of applying a threshold.

Figure 2. Applying thresholds to a gray-scale image produces a binary image in which a processor can count pixels. A black—or white—pixel count outside a fixed range indicates a defective product. Courtesy of Banner Engineering.

By counting the black (or white) pixels, the sensor's internal processor can determine whether an image—and thus a product—should pass or fail. The pixel-counting operation works well to detect the absence of components or the presence of extra components. In either case, the pixel counts from the images would not correspond to those for a good product.

You can buy vision sensors with different image sensors, ranging from standard arrays of 640x480 pixels up to arrays of 1024x1240 pixels. The larger image sensors can resolve small changes in pixel counts.

Because the mechanical and electrical hookup of a vision sensor doesn't present a great challenge, applying a vision sensor to an inspection task can look deceptively simple. But before you buy a box full of sensors, you must decide what you want a sensor to inspect for and what you want the inspection results to tell you. For example, if you want the sensor to measure the angle between connector A and the main PCB, you'll also need to include the range of angles the sensor should accept or reject.

Once you can define your needs clearly, you'll find that using vision sensors instead of a central system can save time. It can take six months or longer to completely specify and order a large machine-vision system. But because inexpensive vision sensors take little time to set up and use, the period from calling a salesperson to making a purchase can require only a few days to several weeks.

In most cases, purchasing and installing small vision sensors won't need an IT department's blessings. Small vision sensors add no "foreign" hardware or software to your company's computers or networks. (Some sensors do provide a network connection, but it's not necessary to network such sensors to use them.)

Although vision systems look appealing, don't try to use every feature they provide. Engineers enjoy a technical challenge, so if a feature exists, they'll try to use it. This philosophy can lead to trouble, as engineers expand the tasks assigned to a vision sensor in an attempt to squeeze out of it all its capabilities. An overloaded sensor may fail to live up to its promise, or worse, it may eventually "crash" a production line. Practice will show what you can and can't do with a sensor, but start with a simple task and ensure a sensor performs it well before you try to expand what the sensor does on a production line.

To learn more about the qualitative and quantitative capabilities of sensors, read the manuals for several of the sensor types you might use. These short manuals show what sensor setup involves, what it takes to program a sensor, and the various options a manufacturer supplies. You can ask manufacturers for printed manuals, or you can download data from manufacturers' Web sites. (See "Manufacturers of vision sensors," p. 50.) The 32-page manual for Banner Engineering's Banner Presence Plus pixel-counting sensors, for example, provides product data and helpful tutorials on general lighting and how to choose a lens.

If you're familiar with machine-vision systems, then manufacturers' data sheets and specifications should give you enough information to decide whether to use a vision sensor. But if you're new to machine-vision—and it's not your primary job responsibility—you may need more assistance.

Manufacturers offer different levels of support. Some sell their sensors through distributors or sales representatives, and people at these businesses can offer advice about using the systems they sell. These people sell many products, so don't expect them to have detailed knowledge about every one. When you reach the limits of distributors' or reps' knowledge, they should refer you to an application engineer at the sensor manufacturer.

Suppliers that work directly with end users usually have their own field application engineers (FAEs) who can offer guidance. Because the prices for sensors start below $4000, don't expect a great deal of support unless you plan to automate many production lines. A small sale won't get you a great deal of assistance. On the other hand, you're investigating vision sensors because they're easy to use; by definition, they shouldn't require a lot of support and applications assistance.

If you do run into problems or decide an application requires more expertise than you have, vendors can recommend system integrators (Ref. 1). These integrators work with the vendor's products, they have used them in other applications, and they may have experience working with other companies that faced challenges similar to yours. An integrator can work with you as a consultant or can deliver a complete inspection system. Working with an integrator will cost more than working on your own, but an integrator may do the job faster and should guarantee a specific level of success. Some integrators also will contract with you for continuing support. Keep in mind that integrators require detailed specifications, and working with an integrator can lengthen the time needed to complete a project.

Vision-sensor hardware has no value on its own. Someone must set it up so it "knows" what to inspect for and how to evaluate the images it acquires. This setup step can involve simply pressing a button, connecting inspection-tasks icons on a monitor screen, using a small handheld control box, or entering information in a spreadsheet-like table.

Figure 3. The HawkEye 15 has a single function: Its sensor reads 1-D and 2-D codes for production control and product tracking. Courtesy of RVSI Acuity CiMatrix.

Vision sensors all "communicate" with some sort of external device, but communication capabilities vary from model to model. Today, few sensors require more than a few I/O lines that communicate with a PLC. So, sensors provide transistor outputs that can sink or source current to indicate an on/off condition. Sensors also accept a trigger input. Sensors that read codes, recognize printed characters, count pixels, measure dimensions, and provide other quantitative data provide a communication link such as an RS-232 port or an Ethernet port. An attached computer can acquire the transmitted data and store it for later use or pass it on to other equipment.

Most applications today don't use sophisticated communication technologies, even though those capabilities exist. Cognex's In-Sight sensors, for example, can communicate using TCP/IP and can transfer images to a server using the standard Internet file-transport protocol (ftp). Expect to see vision-system developers use existing communication capabilities—and ask for expanded capabilities—as managers request more quality-control data. You'll also see developers tie communications into production equipment to form a feedback loop that can adjust a process to reduce defects. Engineers will also find other uses for inspection data. After all, with an Ethernet connection available on a sensor, what engineer could resist attaching it to a network, just to see what it could do?