Siemens SINUMERIK 840D sl Function Manual page 434

Compensations (K3)
Neural quadrant error compensation
Learning ON/OFF
The actual learning process of the neural network is then activated in the reference NC
program. This is done using the following high-level language command:
QECLRNON(axis name 1, ... 4) Learning ON (for specified axes)
Only during this phase are the characteristics changed.
After the learning motions of the required axes have been completed, the learning process is
deactivated for all axes. This is done with the high-level language command
QECLRNOF Deactivate learning (simultaneously for all axes)
After power-on reset, end of program (M02/M30) or operator panel front reset, learning is
also deactivated.
The current "Learning on/off" status is displayed in the service display "Axes" with "QEC
learning active" (1 = active; 0 = inactive).
Learning cycle call
Once learning has been activated, the reference NC program calls the learning cycle by
means of the following input parameters:
● Number of axes to which learning is to apply (up to four).
● Names of the learning axes
● Detailed learning active yes/no (TRUE/FALSE)
434
Prerequisites:
If more than one axis is to learn at the same time, all QEC system variables of the axes
involved must have the same values. These values are monitored and an error message
is output if they are not equal.
Initial number (same for all axes) Value always 0 (setpoint branch)
Learning mode (initial learning = 0; relearning = 1) 0: Initial learning active. All values of
the network are preset to 0 before learning.
1: Relearning active. Learning continues with the values already learnt in the defined step
width.
FALSE: "Detailed learning" is not active. The characteristic is therefore learnt in the step
width of the coarse quantization of the acceleration.
TRUE: "Detailed learning" is active. The characteristic is therefore learnt in the step width
of the fine quantization of the acceleration, i.e. with fine quantization of 10 steps per
coarse step, determination of the characteristic takes ten times longer. This parameter
must therefore only be used for extremely high precision requirements.
Note
If "Detailed learning" is selected, the number of learning passes can and must be reduced
in order to reduce the learning duration (recommended range: between 1 and 5).
Function Manual, 03/2009, 6FC5397-1BP10-4BA0
Extended Functions