Sell machine vision to your boss

Before you can convince your company's management to invest in a machine-vision system, you need to complete an evaluation process. This article presents six planning steps that serve as starting points for an analysis of your requirements, vendor capabilities, and cost implications.

Whether you begin your evaluation on your own initiative or at the suggestion of your boss, keep your boss up to date about what you're doing. A steady flow of information makes it easier to gauge the boss's interest and commitment, and if you eventually recommend the purchase of a system, the recommendation won't come as a surprise.

Before you can decide whether a vision system will help you improve quality, you must evaluate every step in your manufacturing process. Note where defects occur, the types of defects the inspectors see, any trends they observe, and why the defects occur.

Your goal is to determine whether a vision system can help you prevent future defects, not just help you remove defective products. As you evaluate the manufacturing process, determine what catching the various types of defects sooner might do to improve productivity. What if a system could catch missing components right after solder reflow, rather than at the end of the printed-circuit-board (PCB) manufacturing line? How would that affect production, rework, functional test, and other operations on the production line?

If you want quantitative data, list the types of defects you encounter and gather defect data over several production runs. You then can produce a histogram of defect type vs. defect occurrences. This graph, called a Pareto chart, will help you set priorities for the types of defects you decide to inspect for. Keep in mind that just because a defect occurs many times, it doesn't necessarily become a high-priority inspection target. Defects that occur less frequently may cost more in the long run.

After you evaluate your process and locate defect sources, start to assign the costs that defects add to production. If you catch a PCB with missing components at the end of a production line and send it to a rework station, include all the associated costs to produce the defective board to that point; then add the repair costs, which include parts, rework labor, additional testing steps, and restocking charges. Then, estimate the costs associated with discovering the missing parts before reflow—by catching defects sooner, you eliminate rework and repair costs and reduce in-process inventory.

Also keep in mind that if you have a 5% failure rate at the end of your production line, you'll have to start 105 PCBs down the production line just to get 100 good ones. Always account for the cost of manufacturing those five bad boards. A vision system may reduce bad boards to a 1% failure rate, which means you start with fewer "bad" boards down the line. In effect, you increase the production capacity of the line by 4%. Figure that increased productivity in your estimates of the savings a machine-vision system can yield.

Vision systems can also uncover subtle problems that get past electrical tests but that can cause a product to fail in the field. For example, a solder-print problem could prevent a pad from receiving enough solder paste, yet the component's contact could provide just enough solder to melt and form a weak connection. This connection could pass electrical tests, but it has a high probability of failing in the field under normal use. Using a vision system after the solder printer would catch boards that lack sufficient solder, saving the costs of field returns and damage to your company's reputation.

Once you know what types of defects you need to detect, you must "estimate" what you can do to improve quality. You may decide an inspection system that provides information about solder printing by measuring a 2-D solder-paste area would be preferable to a system that makes more extensive (and expensive) 3-D volumetric measurements. Such assumptions may undergo changes as you progress through your planning, but they give you a starting point to work from.

You can make informed guesses about the priorities you want to place on inspections. The information on your Pareto chart may show SMT-device placement causes the most defects. So, inspecting a PCB right after parts placement may take priority over checking a process that seems to work well almost all of the time.

Your assumptions might consider whether you need color or black-and-white cameras, the spatial resolutions you need to examine on SMT components, the speed of your production line, and the variety of PCBs you plan to inspect on the line. Consider whether you might expand a vision system to include production-line control, LAN communications (often standard), bar-code readers, and other capabilities.

Your estimates of costs and capabilities—even though imprecise—give you criteria to keep in mind as you seek information about machine-vision systems, hardware, software, and applications. You can learn about machine-vision techniques and technologies in resources such as the Inspection technology section on T&MW's Web site (www.tmworld.com/ins). You should also check the Web sites of manufacturers of vision hardware and software for application notes, data sheets, white papers, course notes, and tutorials. You can also download trial versions of machine-vision software so you can see how it operates.

In this step, don't try to select specific products. Instead, look for the types of products you think will solve your problems. If you think you need 2-D solder-paste inspection systems, obtain information about the types of measurements they make, what they do with the data, and how they integrate with statistical process-control (SPC) software.

After you learn more about the capabilities of vision-systems hardware, software, and applications, start contacting vendors. Ask vendor representatives if they have had experience with applications such as yours, if they have delivered a system that inspects similar types of products, and if they will perform a feasibility study on your PCB.

Clearly state what you want to inspect and what you want the inspection results to tell you. You might tell the vendor that you need to inspect x solder joints on a rectangular PCB with dimensions of a × b in. This inspection should measure the area of the solder deposit at each pad and determine whether it falls within acceptable limits. Or, you might want a vision system to ensure an SMT device exists at each of x component positions on a PCB. (Let the vendor know the sizes of the components you want to inspect.) Based on such information, a vendor can help you determine whether a vision system can handle your application.

Determining whether a vendor has tackled applications similar to yours will help calm fears that the vendor plans to use your funds to explore a new market for inspection systems. Ask the vendor to put you in contact with a company that has a similar application. By talking with vendors' customers who have solved a similar vision-system problem, you'll get a better idea about what a vendor's equipment can—or can't—do. And be sure to ask about reliability, support, upgrade policies, and training.

As you investigate ways to solve your inspection problems, also consider working with a system integrator. An integrator can supply parts of a system or a complete working system, including regular updates and maintenance (Ref. 1).

Figure 1 In some cases, you can save money by training someone who worked as a production-line inspector to maintain new machine-vision systems. Courtesy of PPT Vision.

Next, narrow your search to two or three vendors. These vendors may suggest specific hardware and software for you to examine in detail. Ask them to perform a feasibility study in which each sets up its equipment and runs several tests using your product as a test subject. You must supply good, bad, and marginal products to get valid test results. (Several sample feasibility-study reports, courtesy of Data Translation, are available in the box at the top of this page.)

When you talk with a vendor, be sure to discuss whether its system will be compatible with future upgrades. Don't buy a system that will force you to scrap hardware and software in order to take advantage of new offerings from your vendor. Also, discuss your anticipated future needs. If, for instance, you'll need to inspect PCBs at a faster rate or to read data-matrix codes, tell the vendor. Otherwise, you may get locked into a system that will not work in more demanding applications.

Understand that vendors perform feasibility studies because they want to sell hardware, software, or services. They do a lot of free work to show how specific products will solve your inspection problems. Don't think, "Why should I hire a consultant when this vendor will work for free?" Abuse the relationship, and the vendor may put you "on the clock" and charge for development work.

At the end of a feasibility study, the vendor will provide a ballpark cost for either a complete machine-vision system or for key parts. You need to include these costs in the plan you prepare for your boss. Remember to include costs to install and maintain the equipment and costs to train operators. Perhaps you can save money by training someone who now performs visual inspections to take over a computer-based machine-vision system (Figure 1). When you include the cost of people, have your accounting people provide the cost of benefits, which can add from 20% to 30% to a person's base salary.

At this point, combine the costs and savings estimates you determined in the Estimate step with the costs you calculated in the Demonstrate step. You can set up a spreadsheet to track the financial information, which will make it easy to adjust amounts and observe their effect on the total costs or savings.

Based on your initial results, you may need to revise some of the estimates you made in the Estimate and Demonstrate steps. Perhaps you can slightly reduce the capabilities of a machine-vision system and decrease the system's cost, but without adversely affecting quality. On the other hand, you might add inspection capabilities—with little effect on the overall cost of the vision system—to enhance the quality of products.

A spreadsheet model will help you determine how long it will take for the savings brought about by quality improvements to pay back the cost of the vision systems. The "payback period" shouldn't exceed a year, and most managers expect a payback period of six months or less.

You should finally be ready to make a report to your boss. Remember, you've kept the boss up to date, so your recommendation shouldn't come as a surprise.

You can reverse the six steps presented here to develop an outline for your report. Present your recommendation first—then show the calculations, describe how you found a vendor or vendors, explain what you investigated, and describe the quality problems you solved. You may not need the information from the Estimate step. Finally, polish your writing and produce overhead slides for your formal presentation.

For further reading

"Machine Vision & Image Processing Applications Worksheet," Executive Automation Systems, Palm Harbor, FL. www.scx2.com/fourth_level/workshee.html.

Scheiber, Stephen F., Building an Intelligent Manufacturing Line, ConsuLogic Consulting Services, Slingerlands, NY, 2001.

Scheiber, Stephen F., "The economics of x-rays," Test & Measurement World, February 2001. p. 10. www.tmworld.com/archives .