Siemens SINUMERIK 840D sl Function Manual page 426

Hide thumbs Also See for SINUMERIK 840D sl:
Table of Contents

Advertisement

Compensations (K3)
Neural quadrant error compensation
QEC system variables
The data for parameterizing the neural network are defined as system variables that can be
written and read by an NC program. The following system variables are used for
parameterization of the neural network:
● $AA_QEC_COARSE_STEPS
● $AA_QEC_FINE_STEPS
● $AA_QEC_DIRECTIONAL
● $AA_QEC_LEARNING_RATE
426
"Coarse quantization of characteristic"
This parameter defined the coarse quantization of the input signal and is therefore the
resolution of the characteristic. The larger the value that is selected, the higher the
memory requirement and the greater the length of time required for the training phase.
See the end of this section for more information.
Range of values: 1 to 1024; recommended value: 49
"Fine quantization of characteristic"
This parameter defines the fine quantization of the input signal and is therefore the
resolution of the characteristic. The larger the value that is selected, the higher the
memory requirement.
Range of values: 1 ... 16; recommended value: 8
"Directionality"
This parameter defines whether the compensation is to be injected directionally or not. If
activated, a separate characteristic is determined and stored for each acceleration
direction. Because two characteristics are used, double the memory space must be
reserved in the non-volatile user memory.
Range of values: TRUE/FALSE; recommended value: FALSE
"Learning rate for active learning phase"
With the learning rate it is possible to determine how quickly the optimum characteristic is
to be learnt in the active learning phase of the neural QEC. This value is a weighting
factor with which it is possible to define to what extent the deviations affect the injection
amplitude. With higher values (>100%), the characteristic is learned more quickly but too
high learning rate values (weighting factors) can cause instability (two-step response).
A small learning rate is recommended for relearning processes during normal operation
(< 50%) otherwise the characteristic is changed on every little disturbance when the
speed passes through zero.
Range of values: > 0%; ≤ 500%; recommended value: 50%
Function Manual, 03/2009, 6FC5397-1BP10-4BA0
Extended Functions

Hide quick links:

Advertisement

Table of Contents
loading

This manual is also suitable for:

Sinumerik 840de sl

Table of Contents