Whether or not you’ve considered using machine vision in motion control, there’s something you should know. Many engineers are using vision technology in a wide range of applications, and with encouraging results.
From both a hardware and software perspective, the obstacles that hindered vision in the past are disappearing. Today’s solutions are inexpensive, easy to implement, and flexible enough for almost any application you can think up.
Why wait? Find out now if machine vision has something to offer you.
Let's get practical
Q: Where does machine vision make sense?
A: Vision sensors are commonly employed to inspect labels on bottling and packaging lines (presence, absence, orientation); check assembled parts for shape, height, and rotation; find and inspect features on printed circuit boards; detect surface flaws in plastic molded parts; perform optical character recognition and verification on labels for date and lot codes; and capture high-speed position and tooling wear data for robot controllers.
Q: How do I implement a vision system?
A: Distributed machine vision is the easiest and most economical approach. The first step is to identify places where you want to inspect or measure your process. Look for points where you can reject faulty parts before adding significant value or stop a botched process before incurring material loss or scrap.
Q: Where do I start?
A: Define what you want to inspect or measure and what you want to do with the results. Next, determine what type of lighting you need to give the image sufficient contrast to expose the features of interest. If you’re having trouble, you might want to contact a company that deals specifically with machine vision illumination.
Q: What about software development?
A: With newer vision sensors, setup is entirely menu driven or accomplished through Windows-based, drag-and-drop software. The procedure, either way, is to establish your best image, set up your inspections and measurements, then configure the results. Vision information is usually available in raw form (digital output) or as data transmitted from a standard communication port to another system, such as motion controller, PLC, or data collection device.
Q: How fast is a typical vision system?
A: Throughput is determined by the type of sensor and the nature (and number) of inspections being performed. The fastest vision sensors can capture and analyze an image in 3.5 msec or so. Typical inspection rates are from 1,500 to 5,000 parts per minute.
Q: Is vision feedback used in motion control?
A: Yes. Some sensors link directly to motion controllers through standard RS-232C/RS-422 or Ethernet communication ports. Others send their output to PLCs, where it may be employed to actuate motors and drives. Edge detection, rotation, and X-Y position information are common motion control feedback signals.
Answers by Mark Sippel, vision product manager, Omron Electronics LLC, Schaumburg, Ill. For more information, contact Mark at (847) 843-7900 or firstname.lastname@example.org.
A taste of success
Machine vision isn’t always the answer, as one candy manufacturer learned through a bittersweet lesson.
The maker of premium chocolates had been using a general-purpose vision system to guide a robot that picks randomly oriented candies off a conveyor and places them in individual wrappers. From the get-go, the solution was expensive and temperamental, and it had difficulty locating chocolates in close proximity with each other.
The problem wasn’t the use of vision, but the way it was implemented in hardware and software. Ironically, for a fraction of the cost, a simpler solution provided the fix. Now, with a Cognex vision sensor, not only can the robot pull randomly oriented chocolates off the moving line, it knows to ignore chocolate smudges on the conveyor surface, a pitfall for the prevision system.
CCD lets servos see
Researchers at the University of Strasbourg in France have developed a new motion control technique based on machine vision that gives robots a more humanlike ability to do work. The method, called “visual servo,” relies on feedback indicating the position and motion of a target relative to the robot itself. As the target moves, its new position is captured by camera then delivered to a “visual loop” controller. The image signals are then applied to the robot, making it move it with the appropriate dynamics to achieve the intended task.
“We are hopeful that visual servo technology will have wide applications in the manufacturing industry, from welding cars on automated assembly lines to picking cookies off a conveyer belt,” says Jacques Gangloff, primary investigator at the university.
The key to visual servo is its ability to track objects, researchers claim. In the average manufacturing environment, controllers don’t always know the precise location of every object. But if an object’s movement can be tracked, say with a digital camera, a visual guidance system can adjust for variations in speed and trajectory, permitting automated tasks that would be impossible with ordinary blind placement.
Strasbourg’s experimental setup consists of a high-speed CCD camera, a six degree of freedom industrial manipulator, a PC with a frame grabber, and a visual loop controller. The camera, mounted on the robot’s end effector, records the moving target at the rate of 120 images per second with a resolution of 640 x 240 pixels. The frame grabber acquires and digitizes the image, then sends it to host memory for processing and control.
Basically, the controller issues a move command and compares the actual trajectory with what was expected. The process is repeated every 8.33 msec, at the 120 Hz sampling rate, each time resulting in six new joint commands to keep the robot on the desired path at the proper speed. To avoid delays, the controller uses a rather simple math model, a linear system of equations based on a sample trajectory, as its predictive element.
Easy as Ethernet
Once relegated to end-of-the-line inspection, machine vision is moving into all types of manufacturing operations to catch and correct defects at the point of occurrence. Compact, inexpensive vision sensors with built-in processors are one reason; Ethernet connectivity is the other.
Over the past decade, Ethernet has migrated from the corporate IS level to the factory floor, displacing many proprietary network connections. As a result, a wide variety of intelligent high-speed control devices, including vision sensors, are now sharing information with each other as well as higher-level networks.
By making vision data accessible, Ethernet has given manufacturers a new tool for improving quality, consistency, and throughput. Ethernet-based vision area networks let machines “see” what they are doing, so they can make adjustments on-the-fly to optimize their output. By contrast, end-of-the-line vision systems can only provide information about process problems after the fact.
By uplinking vision area networks to existing plant and enterprise networks, manufacturers can manage vision activity from remote locations — setting up and modifying applications, sharing applications with other sites, and troubleshooting machines and plants from anywhere at anytime. Uplinking to plant and enterprise networks also gives manufacturers access to quality related data throughout the plant, enterprise, and global organization.
Comments by Evan Lubofsky, Cognex Corp., Natick, Mass. For more information, contact Evan at (508) 650-3140 or email@example.com.