Control Engineering

Link This | Email this | Blog This | Comments (1)

Do absolute encoders need calibration?
October 6, 2008

The short answer is that everything needs calibration!

The full text of this question is: “In the aerospace company I work in, all CNC systems are calibrated regularly, and adjusted as required. These have incremental encoders. We have robots in some of our thermal spraying cells where motion repeatability is not as critical. We are planning a new robot-based processing system where ensuring repeatability and accuracy remain the same. Time is important. The robot uses absolute encoders (and goes through self-checks on each start-up), therefore, the manufacturer states that no calibration is necessary. Is this your view? How then do we ensure that the robot is within specifications and retains that?”

What tests are needed?
Just this morning I was talking about high accuracy motion control with Kevin Kaufenberg at Heidenhain, who was saying that temperature variations are the most serious issue for the type of motion control we were discussing. We were talking about positioning the fiducials on the company’s most precise linear encoders, which are used to control movements of “steppers” that lay down integrated circuit patterns on semiconductor wafers. To make ICs with critical dimensions at the few tens of nanometers level, the steppers need to be controlled at the sub-nanometer level, which means that the encoders have to be made with pico-meter precision!

The moral of this story is that questions about calibration always have to be answered in context. Here the context is CNC machining at an aerospace company. Critical parts likely will be manufactured to tolerances in the range from a few ten thousandths to a few thousandths of an inch, which is a few tens of microns. That’s five or six orders of magnitude coarser than what Kaufenberg’s company deals with, and three orders of magnitude coarser than what his customers’ customers (semiconductor fabricators) deal with.

At the CNC level of precision, the absolute encoders providing feedback from the robot’s joints likely are perfectly stable. They’d be made of Zerodur, which is a “zero-temperature-coefficient” glass that makes Pyrex (which was the best low-coefficient glass available when I was dealing with the stuff) look absolutely sloppy. They’d also be rotary encoders, which are much less affected by temperature variations than linear encoders. My initial guess, therefore, is that traditional calibration — which looks at drifts that occur as components age — is unnecessary for the application.

Of course, the encoders had to be calibrated within a gnat’s eyelash to begin with, but that’s the encoder manufacturer’s lookout, not the users. Once an absolute encoder has been manufactured and deployed in the field, there is nothing a maintenance person at the user company can or should do. There’s no physical process known that can affect the uncertainty of its readings without doing obvious physical damage. 

There are three “danger” points (shown in red) on a robot axis where measurement uncertainty can be compromised.

Encoder measurement “Danger” points
Calibration in general, however, does not restrict itself to slow changes due to aging. Anything that can affect accuracy of the final tool position is suspect. So, it’s important to look for “danger points” in the structure of a typical robot motion axis where measurement errors can arise.

Optical rotary encoders consist of an encoder disk carrying etched fiducials whose movement past a readout sensor produces an output signal. Encoder disks often mount to the housings of the motors driving the axis they serve, with the sensor attached to the output shaft. Barring catastrophic destruction of the encoder disk or failure of the sensor, there are only three places where something can happen to degrade measurement accuracy.

One of the danger points is where the encoder attaches to the structure. Should the encoder disk slip with respect to the support structure, the measurement would be compromised. Similarly, should the sensor attachment slip with respect to the lever (bell crank, or whatever you want to call it) carrying the tool, the measurement will be compromised. Finally, if the lever arm itself flexes, bends, or changes dimension due to temperature changes, the CNC tool point will not be where the controller thinks it is based on encoder output.

Recovery strategy
I discussed these issues with Tom Wyatt, national sales & product manager in Heidenhain’s Automation Division, and Chad Henry, an application engineer at robot maker Stäubli Corp. They agreed that providing redundant encoder signals was the way to deal with these issues.

Robot controllers can use a redundant signal to keep track of the motion in addition to the absolute encoder signal. Examples of redundant signals are counting motor drive cycles, or output from a relative encoder scale that might be etched into the absolute encoder disk. Any event (collision with an errant forklift, for example) severe enough to cause the encoder to slip or arm to permanently deform will cause the redundant signal to miss counts. By constantly monitoring the position reported through redundant channels, the controller can identify encoder-slip events and send up an alarm. A significant event would be defined as a discrepancy in position signals of more than a predetermined amount.

When a significant event occurs, it would be necessary to manually bring the robot to a well-defined start position, such as all joints pointing straight up. The controller would then define the new encoder readouts as the zero points for each axis.

Note that it is unnecessary to “fix” the encoders. Relative encoders, of course, simply add or subtract fiducial crossings from the value in a computer memory register as the axis moves away from a starting point. Absolute encoders, on the other hand, simply use an offset from the encoder code’s nominal zero, anyway. An encoder slip would simply require finding a new offset value for the absolute encoder and zeroing the relative encoder’s position register.

Non-encoder measurement issues
Thermal expansions large enough to change mechanical component dimensions, on the other hand, can be problematical, but they are something that precision machining operations have always had to deal with. When machine tool and work materials are compatible, the problem largely goes away. Everything expands and contracts uniformly, so variations automatically compensate each other. Non-uniform temperatures must always be minimized, however. When machining incompatible materials, temperature becomes an important process variable that must be controlled.

Similarly, machine tool structures must be robust enough to keep stress-induced dimensional variations below acceptable levels. Again, this is an issue that machine shops have always dealt with. In general, feedback automation cannot compensate for rickety construction. Robots for use in automated CNC applications must conform to the same standards as the rest of the shop equipment.

In summary, providing the robots follow best practices for redundant feedback signals (including an absolute-value channel), shop-equipment quality, and procedures for recovering from incidents, there is no reason to put robots on an encoder-calibration regimen. The encoders won’t go out of whack. Everything else, however, needs to be monitored as part of a well-thought-out maintenance program.

Related reading:
Encoder calibration, safety audit
How Encoders Make Automated Motion Safe

Posted by Charlie Masi on October 6, 2008 | Comments (1)