Singled OutNo matter what technology engineers use to feed parts to an assembly system, there are a number of different tasks that need to be executed. These include singulation, whereby a mass of parts is separated into individual parts; orientation, in which the parts are maneuvered so they are facing or sitting the same way; positioning, in which each correctly oriented part is placed where it can be transferred to the point where it can be worked on; and manipulation, whereby a gripper or some other device places the part in its final processing position.
Depending on the application and the parts involved, a robot can perform some or all of these tasks. For example, an engineer may elect to singulate and orient the parts in advance of their being manipulated by the robot, or have the robot execute the entire part-feeding operation from the beginning. Then again, an engineer may choose an arrangement that falls somewhere between these two extremes.
Ultimately, the type of part or assembly being handled will dictate the best approach. For example, if an engineer is processing hundreds of small, easily jumbled parts that lack good features for machine-vision recognition, it will likely be necessary to give the robot some help in sorting things out. On the other hand, if the part includes a wealth of recognizable features, cycle times are more generous and the part naturally achieves a stable, predictable orientation, the robot may be able to take on more tasks on its own.
“Are the parts round or otherwise not orientable? Is there a top side and bottom side, or is this applicable? Are the parts delicate or will they be otherwise damaged from being jostled around in a bin? These are all variables that need to be considered,” says Brian Jones, section manager for DENSO Robotics (Long Beach, CA).
Rohit Khanolkar, engineering manager at Applied Manufacturing Technologies Inc. (Orion, MI), notes that parts should also have sufficient features to allow a robot to pick them up reliably and efficiently. “Some parts that are light, but have no features, say a blank sheet of metal, can be picked up using vacuum, while others need proper clamping and locating to load to an assembly line,” Khanolkar says.
In terms of the actual mechanisms for presenting parts to a robot, the simplest-at least from the robot’s perspective-is to have them prearranged in a fixtured pallet. Another option is to have the parts supplied via a tape-and-reel feeder, like those used in the surface mount electronics industry.
In both cases, the parts are already singulated, oriented and positioned in three-dimensional space. All that remains is for the robot to grip each part and place it in a fixture for processing. The robot can be equipped with a machine-vision system to precisely locate and place the part exactly where it needs to go. The vision system can also check for defects, in which case the robot can shunt the part off to a reject bin.
To maximize efficiency, engineers can fixture multiple parts on a single tray. In some applications, especially those involving smaller parts with flat features, a single robot can use a large, vacuum-based end effector to pick up and load multiple parts simultaneously.
Engineers can also implement an automated pallet loading system, which makes it possible for a human operator to load multiple pallets that are then fed to the robot as needed. For example, Stäubli Corp. (Duncan, SC) recently helped build a fully automated cell phone assembly line employing multiple SpectraFlex Small Part Tray Handlers, in which dozens of trays are manually loaded into a set of vertically oriented belt slots. Each tray handler cycles full trays to a robotic parts-feeder, at the same time removing the empty trays.
In those applications where the robot is feeding parts to a workstation midway through an assembly process, the parts can also be presented to a robot using a rotary indexing table or pallet-based conveyor. For example, midway down the Stäubli cell-phone line, there are a number of stations where a pallet conveying work in progress stops in front of a robot, which then loads the assembly into a processing machine. When work is complete, the robot returns the part to its fixtured pallet, which continues on to the next station.
Picking at Semi-randomOf course, things would be much simpler if all an engineer had to do was plop a box with a pile of parts in front of the robot, and the robot just pulled out and loaded the parts as needed. However, this scenario, while appealing, remains problematic. In fact, it turns out that that equipping a robot to automatically pick parts from a jumble is remarkably difficult.
For example, there is the problem of overlapping parts and that bugaboo of so many machine-vision applications, variations in lighting. Then there is the challenge of creating a machine that can avoid collisions with things like the sides of a pallet and recognize the multiple geometries created by an object as it rotates through three-dimensional space. Not surprisingly, simply processing the data required to perform these kinds of calculations can be extremely daunting. It is no accident that many developers refer to random bin picking as the “Holy Grail” of robotics and machine vision.
Luckily, companies on both the robotics and machine-vision sides of the equation have made great strides in recent years, and true, robust and reliable random bin picking systems may soon be both widespread and readily available. A few systems are already at work in some plants in the automotive sector. Better still, a number of companies offer a range of “semi-random” systems that allow robots to take on a bigger portion of the parts-feeding job today.
For example, in those cases where engineers are working with parts that lie flat, they can implement a pick-and-place system in which a vision-equipped robot uses its end effector to pick and then load parts as they pass by on a conveyor. This process is not entirely random: Parts orientation is partially fixed by the fact that the parts are traveling down a flat surface. In addition, engineers need to singulate the parts, because overlapped parts will create geometries the robot can’t recognize. However, it is still much easier, and cheaper, than having a human operator manually prepare the parts by placing them in some kind of a fixture.
Ben Sagan, vice president of sales at KUKA Robotics Corp. (Clinton Township, MI), says his company has implemented a number of conveyor-based systems used in the assembly of light bulbs. According to Sagan, these systems incorporate a brush arrangement that ensures the unfinished light bulbs are all traveling down the belt in a single layer. The conveyor is also backlit so that anywhere from eight to 10 robots can locate and pick up the bulbs quickly and reliably. This backlighting also makes it easier for the vision system to identify cracks, chips or any other flaws, so that any defective bulbs can be allowed to pass unpicked and deposited in a rejection bin.
Sagan notes that in addition to freeing literally dozens of people to work in other parts of the plant, the robots have markedly improved product quality and scrap rates. For one thing, Sagan says, the machine-vision system is much more reliable when weeding out bad bulbs. For another, the robots handle the delicate parts more gently than their human counterparts.
“If you ask a robot to perform an inspection task, the robot will always perform the task and take action-based results,” agrees Dick Johnson, general manager of material handling at FANUC Robotics (Rochester Hills, MI). “This is important as users strive toward aggressive goals like Six Sigma…. If I were asked to look for a defect and the first 10,000 parts did not have the defect, I might not be as attentive on the next 10,000 parts.”
“Machine vision has certainly revolutionized automated assembly,” says DENSO’s Jones. “Advances such as low ambient light sensitivity and color recognition have made automated assembly faster and more reliable than ever before.”
Johnson notes that, in addition to being reliable, robot-based systems also offer the advantage of being able to handle a range of different components when manufacturing multiple products on a single line. For example, his company’s Flex Feeder system, which “floods” parts to the robot on a linear or circular surface, can differentiate between multiple product types, so that the robot picks only what it needs, even when two different component models are passing by at the same time.
Similarly, Sagan notes that his company’s light bulb systems handle a variety of different bulb geometries.
In addition, John Clark, national sales manager for EPSON Robots (Carson, CA), points to the reduction in “wear and tear” on human operators as yet another benefit of taking them out of the parts-feeding equation.
“A simple ergonomic example is this: Imagine asking an operator to load and unload a 2-pound part every 30 seconds,” Clark says. “Seems very reasonable. [But] run the numbers. That operator is moving close to one ton of material every day.”
Finally, with regard to the robots themselves, Bob Rochelle, sales manager at Kawasaki Robotics USA Inc. (Wixom, MI), notes that, like vision systems, motion technology is also evolving, opening up new part-feeding possibilities as machines become faster and more precise. “We just introduced a small higher-speed six-axis robot that can almost compete with the standard pick-and-place units and some SCARAS,” Rochelle says, as an example of the industry’s growing capabilities.
Increasingly RandomMoving up a level in terms of randomness, engineers today also have the option of implementing a system capable of picking parts directly from a bin, when those parts are in a somewhat “structured” arrangement. Examples include parts that can be laid flat without overlapping, either on a tray or in a pallet, and offer sufficient physical features that allow the machine vision system to differentiate them without difficulty.
For example, FANUC Robotics offers a system that can unload multiple layers of transmission gears from a single bin. Not surprisingly, the company had to overcome a number of challenges to make this system work. For example, the robot’s vision system has to be able to disregard things like oil rings left behind on the plastic sheets that separate the different layers.
Nonetheless, because the gears offer a wealth of easily definable geometric shapes and do a good job of laying flat without overlapping, the robot can still pick them efficiently even though the parts are not individually fixtured.
In terms of truly random bin picking, a robust system that can be readily installed in a wide range of applications is still in the future. However, a number of robotics companies are hard at work on the problem, and random bin picking systems have begun to go to work in some real-world applications.
For example, Stäubli has installed a number of systems in the automotive industry, unloading parts like valves and gears. Similarly, FANUC Robotics has installed systems that pick parts like transmission gears and covers. Motoman Inc. (West Carrollton, OH), ABB Inc. (Auburn Hills, MI) and EPSON are also working on systems.
“There’s no doubt, 100 percent randomness is difficult to do. I think we have a ways to go. But, by no means are we at the beginning, and we’re making progress,” says David Arceneaux, business development manager for Stäubli. “The trend is going in that direction. A lot of people are asking questions about [bin picking].”
FANUC Robotics’ Johnson agrees. According to Johnson, the systems his company has installed picking random parts have shown that consistent cycle times can be difficult to obtain as the robot and machine-vision system work to home in on a single part-which may, in turn, require part queuing to allow uninterrupted downstream operation. However, newer visual techniques, and search and identification algorithms are increasingly bringing down these cycle times.
Finally, robot suppliers emphasize that engineers can often make big gains not so much by implementing the latest in vision technology, but by simply rethinking the way they go about organizing their assembly processes.
For example, if an assembly includes an injection-molded component, why not have a robot unload the molded parts directly to a nearby assembly station, as opposed to dropping them into a bin for processing it elsewhere? Similarly, if a manufacturer is working with machined parts, why not allow a robot to handle the entire transition from the machining to the assembly stations?
According to Johnson, it only makes sense to exploit a robot’s ability to multitask as much as possible, once it has a product safely within its grasp.
“The ability to perform multiple tasks depends on the amount of time available to the robot once it has performed the primary task. It also is related to the envelope of the robot, since it may need to reach another operation, and the articulation needed to present the part properly for the operation,” Johnson says. “Gauging, inspection, gluing, labeling, marking, deburring, assembly and palletizing are all good examples of the types of tasks that can be performed.”
“We’ve only scratched the surface,” Arceneaux says, regarding the degree to which manufacturers are taking advantage of these kinds of efficiencies. “While [a robot] is waiting for a part to be processed, it can be inspecting or palletizing or any number of things.... The possibilities are endless.”