Conceptually, machine vision isn't new, but advances in technology are magnifying its benefits by changing what it is and how it's implemented. Considered by some as the ultimate computer peripheral, its influences in manufacturing are spreading rapidly with each advance in processing power. Machine vision emerged with computers in the late 1950s and 1960s as researchers at the Massachusetts Institute of Technology, Carnegie Mellon University, Stanford University, and other organizations became fascinated with what they considered to be a subset of artificial intelligence -- the idea of getting a computer to see and interact with its environment. The basic components include a TV camera, a computer, and software designed to perform the desired task. In the 1980s the early adopters in manufacturing profitably explored such uses as inspection, quality control, and machine guidance. However, the benefits of the early implementations also came with penalties in terms of the technical expertise needed to install a machine and use a vision solution. To profit from the technology, companies had to have either highly qualified in-house experts or a project involving special system integrators. A common result was a relatively high cost of implementation. For example, in the mid-1980s, a Cincinnati Milacron flexible manufacturing system went on line sporting an elaborate $900,000 three-dimensional robotic guidance system. Several forces are removing those penalties. Benefiting from advances in computers, computer interfaces, cameras, and other components, vision systems have not only grown more powerful, they also have become easier to use while costing less. "For example, in the last year prices have fallen 17%, and the average cost of an implementation in 1998 was only $40,000," notes Nello Zuech, a researcher for the Automated Imaging Assn. (AIA), Ann Arbor, Mich., and president of Vision Systems International Inc., a 16-year-old Yardley, Pa.-based vision consultancy. He notes that the average was drawn from a range that extended as high as $2 million. The price decline becomes even more dramatic when viewed over the last decade, notes Gary Wagner, president of Imaging Technology Inc., Bedford, Mass., a supplier of hardware and software for industrial inspection, machine control, and scientific image analyses. He estimates the dollar cost of a system installed in 1999 at 25% of what it would have been in 1989. Reminiscing about the early days before the advent of powerful PCs and Windows, he observes that machine-vision vendors are in transition from being proprietary-hardware suppliers to a new role as software specialists. "And ordinary engineering personnel, not technology mavericks, can now implement and use vision systems using the same PCs and the same point-and-click approach common to word processing." As prices continue to fall, vendors have been able to extend the reach of vision solutions beyond the early adopters in semiconductors, electronics, and automotive. (In those industries the pressures of rapidly advancing technology, rugged competition, and rising labor costs encourage the aggressive acceptance of advanced manufacturing techniques such as machine vision.) Today, the semiconductors and electronics market segments have a dual significance for machine-vision vendors. Not only do those segments account for half of the vision market, they also head the list of products that would be technologically impossible to make today without machine vision. The reasons include increasing miniaturization, increased use of robotics in assembly operations, and the development of highly automated volume manufacturing equipment. But products with far less sophisticated images are rapidly joining that list, at least from an economic point of view, says William Silver, vice president at Cognex Corp., Natick, Mass., a leading supplier of vision systems. His examples include disposable diapers "where manufacturers have decided that it is absolutely intolerable to have defective products reach the customer. Without machine vision to do two jobs -- one, to guide the robots that place the diaper pieces in the right place and, two, to reject the bad stuff before it gets to the customer, it would not be economically feasible to produce a disposable diaper." Another example: "A manufacturer of eyeliner pencils found machine vision indispensable for ensuring the quality and safety of the product. First, the company needed to inspect for the smearing of color material around the tip of the pencils. Then, they had to verify that the pencils were sharpened at the correct end, since a pencil sharpened at the wrong end would expose a wooden point -- a potential hazard to consumers. Finally, they needed to gauge the correct shape of the sharpened tip. The solution was provided by a Cognex Checkpoint 900C color vision system." Related systems the company has installed include inspecting sunglasses for blemishes, verifying the presence of ink ball tips on pens, reading characters on contact-lens molds, and verifying that logos on razor blades are of high quality. In automotive applications, machine vision is proliferating beyond the usual focus on productivity and quality issues on production lines. Drawing on increased functionality and ease of use, one automaker is adapting vision systems to tasks such as part identification. One implementation at a General Motors Corp. assembly plant solved the need for an automated way to verify that tires routed by conveyor belt are the correct tires for the target vehicle. Because various tire models can be similar in appearance, relying on human inspection to accurately differentiate between them was unreliable. If the wrong tire is routed, the vehicle is prevented from moving to the next stage on the production line until a correct tire is rerouted. The need to avoid costly downtime is only one part of the problem. Because new tire models are constantly added to the line, ease of use was an important factor. The assembly plant uses a Cognex system that compares the tire tread signatures with a trained image. When a new model is added to the line, production engineers are able to create a trained image of the tire's tread pattern simply by using a point and click interface, eliminating the need for assistance from a vision expert. Market researchers at Frost & Sullivan position the automotive sector, estimated to be 28.2% of the 1999 market, to become the largest user in 2004 with 31.7% of the market. Not all of the cost reduction evident in today's vision systems comes from the ever-decreasing cost of computer power. Some comes from conceptual innovation in packaging the vision solution attributes for a specific type of customer. One example is the new and simpler breed of automated inspection systems offered by such companies as Omron Electronics Inc., Schaumburg, Ill., and Imaging Technology. They are pioneering a new and simpler approach to automated inspection systems that, in the case of Omron, start as low as $1,300. Known as vision sensors because they marry vision and sensor technology, these systems fit in a compact box roughly the size of a 35-mm camera and offer the potential to be very cost-effective in reducing the number of flawed products that reach customers. Especially notable is the simpler setup of these systems. While a conventional machine-vision system consists of separate digital cameras, lighting, computers, and sophisticated software, this new breed obviates the need for setup and component coordination expertise. For example, Omron's vision sensors consist of a camera, light source, and processor integrated in a single unit. That has earned them the nickname "smart cameras." They cut costs and simplify operation in two important ways, says Omron's Peter McHugh, manager of machine vision. "First, they eliminate the need for software and associated programming. Second, they do away with the time-consuming process of adjusting lighting, because the light source is built right into the units. No complicated configuration is required. Operators simply place a good sample of the object to be inspected beneath the sensor box, and the system compares the stored image to what it is 'seeing' on the line. Items that fail to match the master image are rejected, and it all happens at speeds of up to 1,200 inspections per minute." McHugh says the systems can be used to automate basic inspection chores such as detecting missing or displaced parts, determining if a label is facing the right way, and ensuring that instruction sheets are included in a package. This helps to improve quality control in virtually any kind of manufacturing environment. One company benefiting from Omron's newest inspection technology is RXI Plastics Inc., a container manufacturer in Triadelphia, W. Va. It is applying an Omron vision sensor to ensure the proper cutting of the foil safety-seal liners -- part of a tamper-proof feature. Running too fast for a visual check by the human operator, the equipment is now teamed with Omron's F30 vision sensor. RXI says it could have used conventional photoelectric sensors, but it values the vision-sensor benefit of being able to determine whether the liner is the right size and shape to cover the cap -- the vision sensor actually counts the number of pixels on each liner. If the pixel count falls below a certain threshold, indicating a hole or other liner anomaly, the system sends a signal to stop the machine for a manual check. Omron's strategy in competing for what is traditionally a photoelectric-sensor application is to emphasize the new flexibility and easy product changeovers that customers gain. It points out that users of photoelectric sensors often believe it cost-effective to configure several photoelectric sensors for applications requiring two-dimensional inspection. The drawback, it points out, is that some reliability is at risk, as is the considerable time required for configuring the sensors. Omron also says the conventional approach does not allow for simple product changeovers and that users may find out that setting up banks of photoelectric sensors can become nearly as costly as setting up a machine-vision system. Imaging Technology's Wagner, who is targeting the same entry-level market with a new Windows CE-based vision solution, offers another rationale -- user education. He notes that "new users of vision systems need a simple low-cost way of beginning to have success in machine vision." Omron's McHugh concurs: "Justifying a purchase order of $15,000 or more for a high-end vision system is a steep learning expense." McHugh and Wagner agree that once new users feel comfortable with the basic concepts of machine vision, they will be more likely and more eager to try the high end. The scope of machine-vision solutions also is being extended by combining them with X-rays. That solved an airbag inspection problem for Autoliv Automotive Safety Products Inc., Ogden, Utah. Controls engineer David A. Wilson explains: "Since the components being inspected are sealed within a metal canister before final inspection, we must use X-ray technology to view the sealed components. A further complexity is the need to use the system on multiple products. We wanted a machine that could be retooled for new inflator products and could be easily maintained by internal resources." Autoliv's solution integrated a vision system (Powervision from Acuity Imaging LCC, Nashua, N.H.) with the X-ray cabinet using programmable logic controllers and a data-collection system. Printed circuit boards are another application for systems using X-rays to locate hidden problems. "Many of our customers now require in-house X-ray capability," says Peiman Amoukhteh, president of COMTEL Electronics Inc., a Tustin, Calif.-based contract manufacturer. COMTEL's X-ray system is from CR Technology Inc., also in Laguna Niguel. In February that company announced a combination X-ray/machine vision system. Most industry analysts see a rosy future for machine vision despite a market hiccup last year from a lag in the semiconductor sector. Zuech's research for the AIA shows that 1998 increased only 1.5% from 1997 with sales of $1.252 billion. Units were up 22.7%, reflecting lower system cost. In contrast, 1997 registered an increase of 18.7% from 1996 with sales of $1.234 billion. Is machine vision for everyone? Industry's quest for quality, especially quality tied to process control and productivity, would indicate so. But Cognex's Silver raises an exception. He says it works best for volume production of highly repetitive objects of at least moderate value. That may be why in France, vision systems on the production lines of Moet & Chandon Group inspect 8,000 to 13,000 bottles of the bubbly per hour for an annual total of 24.2 million bottles. ITMI, a subsidiary of the French company Cap Sesa, provides Moet & Chandon's vision systems.
| What Vision Systems Can Do |
| Robot guidance -- Find the part to be picked up and the location where it is to be placed and transmit the information to the robot. Dimensional gauging -- Measure diameters, arcs, distances, and angles on parts such as a brake cable. Flaw detection -- Inspect parts to make sure they are not chipped, scratched, dented, or missing material. Print verification -- Verify that laser markings on buttons and controls are correct and of high quality. Code reading -- Read optical-character-recognition or 2-D matrix codes for inventory control and part traceability. Systems can handle etched codes such as those on engine blocks and windshields. Assembly verification -- Find all of the components of an assembly such as a test kit, and verify that they are present, in the correct position, and are undamaged. Sorting -- Differentiate between various products or containers by size, shape, or color to ensure the correct products are placed in the correct containers. |
