Evan Lubofsky, Cognex Corporation

During the past several years, there has been an increasing trend in mobile phone manufacturing toward the use of machine vision technology for 100% automated inspection. Machine vision is used to perform a number of operations including inspecting LCD displays for single pixel defects, verifying the correct placement of keypads, gaskets, and other components, automatically reading bar code labels for part tracking, checking the consistency of LCD backlighting, and inspecting product closures for scratches and other defects.

While the mobile phone manufacturing processes could benefit from automated inspection at several key process points, the traditionally high price of machine vision systems has often proved a deterrent to using vision at any stage other than the end of the line. End-of-the-line inspection systems can do a great job at final part checking, but waiting until after a phone has undergone numerous value-adding stages of production can often mean time consuming and costly rework of a faulty phone. At the end of the line, a device typically has been fully assembled and enclosed, so it then has to be reopened so that the problem can be investigated and fixed. Many mobile phone housings, however, are not meant to be reclosed after opening. As a result, the housings and any other associated parts such as gaskets, rivets, and shock pads have to be scrapped. The other problem with end-of-the-line only inspection is that since defects are not detected at the point of occurrence, little information can be generated about how and why defects occurred in the first place.

The multi-camera approach
Manufacturers could, however, expand an end-of-the-line system to other points on the production line by multiplexing vision cameras from a single processor. The benefit here was that the hardware cost associated with each inspection point was minimal. The downside was that, linking off of a single processor, vision processing had to be divided up among multiple cameras, resulting in reduced vision performance at each point. And then there was the issue of scalability, which in technological terms can be defined as how well a solution to some problem will work when the size of the problem increases. With a multiplexed camera configuration, scalability is inherently poor since only a finite number of cameras can be used no matter how many additional uses a manufacturer might find for vision.

Finally, there was a great deal of software complexity and integration costs involved in managing and maintaining multiplexed cameras. Special programming was typically required to synchronize camera triggers and outputs, and a problem at one station often meant the entire system had to be taken offline.

Distributed vision sensors
Recent advances in machine vision sensor technology have enabled mobile phone manufacturers to distribute vision throughout the manufacturing process without the limitations associated with a multiplexed camera approach. Because of the compact, affordable nature of many vision sensors available today and improvements in processor technology, it has become feasible to have one processor per camera, resulting in maximum vision performance at every point on the production line where a sensor is placed, and greater assurance of uptime. And, because each sensor is an independent device, the complexity associated with synchronizing multiple cameras off of one processor is eliminated.

Perhaps more importantly, it is now cost-justifiable for many manufacturers to add vision in enough places so defects are caught at the source. A company that had only used vision at the end of the line to check display functionality may now be able to justify the addition of dedicated vision sensors downstream to check the placement of gaskets, buttons, and printed circuit boards as they are robotically applied. That way, if components are positioned precisely, corrective action can be taken on the spot. With a multiplexed camera approach, there are only so many vision cameras one could use without having to absorb the cost of another vision system. With vision sensors, having one camera per processor means having the scalability to add vision to these points, and others in the future.

The Move Towards Ethernet
Because machine vision can now distributed very cost-effectively throughout the mobile phone manufacturing processes, vision is finding its way into more and more places throughout the factory. As a result, many manufacturers now need a centralized way to maintain and manage the ever-growing number of vision sensors running on the floor. Running 60 vision sensors across 10 production lines is one thing; setting up applications on each, and then modifying them during product changeover, is another.

Over the past decade, Ethernet networking has become a more integral part of the manufacturing process, transcending down from corporate IS level networks to the plant floor. Replacing what have traditionally been costly, complex proprietary network connections, Ethernet provides higher-level computing systems access to the plant floor, allows intelligent, high-speed control devices to share information required for tasks such as work-cell coordination, and offers high speed access to plant floor data from a broad range of plant floor devices.

Now, Ethernet is becoming a key ingredient in the way people use vision on the factory floor. Today's most advanced vision sensors offer built-in Ethernet networking capabilities that enable users to link multiple vision sensors across the factory, manage vision activity remotely, and share vision results data with all levels of the organization.

Networked vision can be implemented in two primary ways. First, two or more vision sensors can be linked over Ethernet to form a dedicated vision area network. In a vision area network, vision sensors can exchange data, and can be managed by some type of host, whether it is a PC or another vision sensor. This type of configuration offers a number of key benefits.

Consider the example mentioned earlier, where vision may need to be implemented at numerous assembly stages. To enable data exchange between several conventional vision systems on this line, one would need to establish a serial communications link by linking several serial cables between each system. Then, to view inspection results, one would either need to have a separate VGA monitor at each inspection point to view inspection results, or have the results sent over serial lines into a single workstation, where an HMI package would be required to consolidate and present the results. In contrast, a network of vision sensors linked over Ethernet enables direct peer-to-peer communications between each sensor over a single line, so there are no complicated cabling schemes to deal with. And, because the network of sensors is managed by a host, vision results data from all four sensors can be collected at a central point, and viewed on a single VGA monitor. The host may also be used, for example, to archive failed images from each vision sensor, which can be used to better determine why certain failures occurred.

The second way of implementing networked vision is to uplink a vision area network to existing plant and enterprise networks. This can provide a number of benefits. First, it enables users to manage vision activity from remote locations. For example, one could set up and modify vision applications, share applications with other plant sites, and remotely troubleshoot problems with technicians, all without ever leaving the office. Additionally, uplinking to plant and enterprise networks enables manufacturers to gain direct access to the data related to the quality of their products directly from the vision sensors from anywhere in the plant, the enterprise, or anywhere within their global organizations. Quality engineers may want to view SPC data, while management may want to keep an eye on production output. All it takes is a workstation with TCP/IP capability.

Instead of being reserved for end-of-the-line part checking of mobile phones, machine vision can now be cost-effectively deployed at various key process points along the way, and centrally managed over Ethernet. By distributing vision at more points in the process, manufactures can respond more quickly to manufacturing problems and achieve better process control. Cost savings can also be substantial. Manufacturers have reported that catching defects on a device at the board level, before final assembly, can save up to 90 minutes of a technician's time per defect in the repair loop. This includes debugging, disassembling, reassembling, retesting, process logging, and reinserting the device on the assembly line. Additionally, with Ethernet connectivity, the ability for manufacturers to centrally manage multiple vision sensors running on the floor, and make vision results data more accessible to all levels of the enterprise, is greatly improved.