Commissioning Of Neural Qec - Siemens SINUMERIK 840D sl Function Manual

K3: Compensation
5.6 Friction compensation (quadrant error compensation)
5.6.3.4

Commissioning of neural QEC

General information
Commissioning the QEC function with neural networks is described in brief below. As we
have already mentioned, the compensation characteristics during the learning phase are
determined automatically.
The axis involved must perform reversals with acceleration values constant section by
section. The QEC system variables for parameterization of the neural network must also be
preset to meet the requirements.
To simplify commissioning as much as possible, NC programs are provided to serve as
reference examples (see Section "Learning the neural network").
In the learning process, a distinction is made between "initial learning" (especially for first
commissioning) and "relearning" (especially for re-optimization of characteristics already
learnt). The procedures of "initial learning" and "relearning" are described below.
If the compensation characteristics for the machine are to be learned for the first time, we
recommend use of the reference NC programs specified in Section "Learning the neural
network".
"Initial learning" sequence
"Initial learning" -> cycle parameters "Learning mode" = 0
1. a) Activate QEC with neural networks for the required axes with machine data setting:
2. Activate QEC system variables:
3. Create the NC program that moves the machine axes to the required positions and
434
MD32490 $MA_FRICT_COMP_MODE = 2 (friction compensation method)
Note
QEC with neural networks is an option!
b) Reserve memory space for the compensation points with machine data
MD38010 $MA_MM_QEC_MAX_POINTS
(number of values for quadrant-error compensation with neural network)
If the required number is not yet known, a generous amount of memory must be reserved
initially (see also item 12).
c) Parameterize and optimize the speed feedforward control (required for the learning
phase)
d) Perform a hardware reset (because of the re-allocation of the non-volatile user
memory)
Adapt the reference NC program QECDAT.MPF for assigning the QEC system variables
for all axes concerned (if necessary use the recommended values) and start the NC
program. If error messages are output, correct the values and restart the NC program.
parameterizes and calls the reference learning cycle QECLRN.SPF (as in the example
program QECSTART.MPF). The feedrate override switch must be set to 100% of the
learning phase so that the parameters can take effect in accordance with the defaults.
Function Manual, 09/2009, 6FC5397-1BP20-0BA0
Extended Functions