Many forces have conspired to create demand for machine vision — pressure on suppliers to ship fewer defective parts, the growing sophistication of robotics in assembly and packaging, consumer demand for consistent quality, lean initiatives to improve profits by reducing scrap, and, always, the push to automate repetitive processes.
The term machine vision applies to optical systems that include an industrial camera that captures an image, and a processor that analyzes specific features of the image itself. The processor can be in a separate PC or built into the camera. Depending on the camera, image resolution ranges from a few hundred thousand pixels to a couple million. That range, though less than consumer cameras, usually provides enough information for the processor to determine what the machine needs to.
Two factors are critical in capturing a good image: the lighting and the lens. Lighting must be sufficiently bright, the right color, and correctly angled to create contrast, which makes the feature of interest stand out from its background. To that end, a machine vision lighting sub-industry has evolved, offering specialized choices such as ring lights that surround the lens, low angle lights for revealing raised features or texture, or back lights to emphasize an object's silhouette. Lens quality determines how much the edges of the image are distorted, how uniformly bright the image is from the center to the edges, and how accurately the camera perceives color.
Machine vision has a wealth of industrial applications, including detecting flaws, verifying correct assembly, and counting, sorting, and measuring objects. Although machine vision can't match the acuteness of human sight and judgment, it's ideal for high-speed inspections and for performing repetitive tasks that quickly bore humans, such as making sure thousands of beverage bottles are filled and properly capped. While machine vision most commonly is used to detect bad finished products, it's increasingly being called upon to report small problems during processing — either to a human or, in a closed-loop system, to the machine — so errors can be corrected before they become big problems.
Among the many machine vision products on the market, multi-component vision systems are at the high end, both in price and expertise required. These custom systems include a camera to capture images and a PC to process them.
In the old school paradigm, vision systems used a relatively inexpensive analog camera, a PC capable of heavy-duty data processing, and a camera interface called a frame grabber built into the PC to digitize the images. Nowadays, more costly cameras are equipped with Gigabit-Ethernet (GigE) or FireWire (IEEE 1394) interface technology, eliminating the need for a frame grabber in all but the most high-performance applications, which require a special interface such as Camera Link.
Vision systems rely on custom or customized software designed for specific, difficult applications such as sizing tiny particles in pharmaceutical production, guiding robots as they perform complicated tasks, and inspecting objects with complex features. The software includes application-specific algorithms that extract key information from the data the camera captures — and depending on the camera's resolution, that could be hundreds of thousands, or even millions, of data pieces.
The sophistication of vision systems comes with a premium price tag, in the tens of thousands of dollars. The cost covers not only equipment, but also the expertise of engineers who design, install, and run the systems.
Machine vision for the masses
Vision sensors, also known as smart cameras, are simpler and significantly more affordable than multi-component vision systems. They do away with clunky PCs and custom software; instead, the processor and camera are in the same housing. And instead of programming, they only require configuration through a simple point-and-click software interface.
Vision sensors are versatile enough for a wide array of applications, some of which previously used a vision system because it was the only option available, and some that benefited from automation but could not justify the cost of vision. Vision sensors require far less expertise than vision systems do; a technician with a two-year degree or a computer-savvy floor worker who programs a PLC can become proficient in their use with only a few days of training.
A full-featured vision sensor starts at about $3,000, including a lens, light and cabling. Optional features, such as bar code reading, color recognition, or high-resolution imaging, increase the price.
When vision is overkill
Not every inspection application requires machine vision. In some cases, a much older optical technology — photoelectric sensors — is good enough. A photoelectric sensor detects a beam of light, either visible or infrared, and responds to the change in intensity of the received light beam.
The most familiar example is the two-piece photoelectric sensor that makes sure the automatic garage door doesn't close on the family cat. When Fluffy dashes under the descending door, she blocks the beam of light between the emitter and receiver for an instant. The receiver then sends a signal to the door controller, which stops the door and reverses its direction. Industrial photoelectrics are more durable than the garage-door variety and can respond to hundreds of beam breaks in a minute. But the underlying concept is the same: They detect a change in the light signal between an emitter and a receiver, and then output a signal to a device that responds in some way.
Several designs of photoelectrics are available, each with its advantages. For example, opposed mode photoelectrics are especially reliable because the beam travels between a separate emitter and receiver with little signal loss. Fixed-field mode sensors have the emitter and receiver in the same housing, so are easier to install; the emitter sends a beam that bounces off an object within its fixed sensing field and returns to the receiver.
Photoelectrics do a fine job of detecting an object's presence or absence, and with a little creativity can be used for a surprisingly large number of applications. For example, two or more can be combined to sort objects based on height. And they're inexpensive and simple to use because there's no fiddling with lighting or lenses. Like vision sensors and systems, their output can be used for machine control. For those reasons, photoelectrics are the first option to consider for automated inspections.
But photoelectric sensors can't collect and analyze thousands or millions of information pixels like a vision sensor can. And photoelectric sensors are, in effect, single-pixel devices, so the object to be sensed must be within the sensor's narrow beam. For that reason, they're typically used when the object's position is fixed and predictable.
Vision sensors and vision systems, on the other hand, capture a scene made up of thousands or millions of pixels, so they can detect an object even if it moves significantly. Motion is a normal part of many production processes, and being able to sense an object despite movement can save thousands of dollars on fixturing and allow production lines to operate at higher speeds.
With these three options — vision systems, vision sensors, and photoelectric sensors — manufacturers can successfully automate inspections at more places in their production processes, improving product quality and further driving down costs. These cost reductions can improve both corporate profitability and quality of life for employees and customers who rely on low-cost manufactured goods.
For more information, contact Banner Engineering at (800) 809-7043 or visit bannerengineering.com. To read more about machine vision systems, how they work, and how to select critical components, visit motionsystemdesign.com's Knowledge FAQtory and look for links that will connect you to related articles and information.
Vision's Holy Grail
Machine designers have long quested after the Holy Grail: An affordable way to integrate machine vision in a network to feed enough data to process controllers such as PLCs, human-machine interfaces (HMIs), and robotic controllers, to allow them to make decisions about complex traits such as color, size, and position. Until recently, their options were to settle for the small amount of information a photoelectric sensor could gather or the slightly more information a vision sensor could rapidly communicate from the tons of information it gathered, or to invest in a full-blown multi-component vision system.
Their quest has been answered with the marriage of low-cost vision sensors with communication protocols such as Modbus TCP and EtherNet/IP. These protocols have increased by a factor of 100 the amount of useful data vision sensors can communicate to a process controller fast enough for the controller to make realtime decisions. Tablet is baby blue instead of powder blue? Kick it off the line. Soda bottles on the fill line are empty? Alert the operator ASAP, because the fill tube probably broke off, and soda is spraying all over. Box is three inches to the left of the one before it? Tell the robotic arm so it knows where to pick up the carton.
While this kind of closed-loop feedback system is more technically challenging to set up than after-the-fact inspection, savings in reduced scrap and reworking makes it worth investigating. Thanks to recent advances in vision technology, today's machine designers can do just that — and get the level of optical sophistication ideally suited to the task.