Computer Drive Installation Instructions

09.95

Siemens AG 2001 All Rights Reserved 6FC5197–jAA50
9–59
SINUMERIK 840C (IA)
Direction-specific Direction-specific injection can be selected via a function parameter under the
compensation Function parameters softkey. This is necessary in cases where the compensa-
tion is not equally effective in opposing quadrants when the injection is not direc-
tion-specific (see diagram below).
However, the following points must be noted in this respect:
S Twice the number of quantization intervals or memory locations must be pro-
vided for the characteristic, e.g. by doubling the coarse quantization.
S The number of learning process runs with the test signal (see “Learning pro-
cesses”) should likewise be increased because only every 2nd zero crossing
is executed again on the same input quantity (this time +/– a, last time |a|)!
S If the resolution is not altered, the start-up process takes longer.
S Changes to this parameter setting cause a re-initialization of the weight fac-
tors already learned.
Direction of
motion
Well
compensated
Poorly com-
pensated
y
x
Fig. 9.32
It is often possible to decrease the resolution while maintaining the same degree
of accuracy by reducing the maximum acceleration. If a higher acceleration rate
than the parameterized operating range is detected, then the injection amplitude
which was calculated for the parameterized maximum operating range is applied.
At high acceleration rates, this injection value remains relatively constant.
Influencing the period of As described under “Learning process”, the test signal for
“Detailed learning” the learning phase is derived from the parameterized acceleration range of the
neural QEC. In normal cases, the acceleration is varied in steps of approximately
one coarse quantization step when “Detailed learning” is set to “no” (e.g. from a1
to a2 – see diagram under “Learning process”). The duration of the learning pro-
cess is thus calculated from the function parameters according to the following
formula:
Learning period = (coarse quantization + 1) x number of learning process runs x
T
Per
The period T
Per
(see diagram “Learning process”) is 1 s. With the default set-
tings, therefore, the learning process period is approximately 12.5 min per axis.
As from SW4.4 the period of the learning signal can be parameterized (NC-
MD 1296*).
The learning period can be reduced by specifying a higher fine quantization set-
ting and a lower coarse quantization setting which will give rise to the effects of a
high fine quantization setting described under “Quantization of operating range”.
9 Drive Servo Start-Up Application (as from SW 3)
9.5.4 Neural quadrant error compensation (QEC – SW 4)
10.94