From time to time we get into discussions on accuracy in motion systems. These discussions usually end up in two general camps; higher accuracy based on the notion of zero backlash mechanisms or higher accuracy feedback devices to measure error and eliminate it electronically. These are interesting conversations because both trails are full of assumptions that have to be dealt with properly or they will lead to the poorly performing systems.
The first thread is based on a purely mechanical perspective. Good mechanical designs lead to good motion control solutions. Understanding the limitations of the mechanics can lead to increased accuracy. So when mechanical components are being selected for a high precision task, things like angular error are a primary concern from the power source to the load.
Mechanical accuracy is most often based on angular error in a motor and the motor’s feedback device or in a gearbox. This makes motors with high resolution feedback preferred. 3 phase servos are commonly available with 17 bit absolute and 20 bit serial encoders. That means feed back of 0.0027 degrees of angular error or less. Depending on your requirement this may, for all practical purposes, be zero lost motion. But it is not actually zero and this presents some problems.
If the motor is turning at 6000 rpm with 17 bit accuracy, it will hit 7.8 Mhz frequency of quadrature data. So you have to make sure that the controller has the bandwidth to read all the data without dropping any bits. The second element of this problem is that the error is non-zero. So a machine that is running at very high speeds, electronic part placements for example at 15,000 per hour, might accumulate 120 billion tiny errors during a year if no means of controlling error is implemented.
Fortunately, high resolution feedback devices can be used as sensors to actually measure the error in the mechanical system. This is crucial to understand and embrace as part of the control system design for higher reliability. You cannot control what you cannot measure. On the other hand, if you can characterize the mechanical error in the system electronically, you can use software to keep track of it.
There is one final consideration on the software side of lost motion. Binary systems are not good at dealing with numerical results with a rollover fraction. Since the sensor creates position information as a binary power of 2, and the real world mechanism is usually not evenly divisible, there are rollover fractions that will accumulate in the control system software. I actually remember an application where they forgot to deal with this in software and when the controller finally faulted out due to a numerical overflow in the position error register, it took days to figure out what went wrong.
More on this next week.