Siemens 611-D, 840C 9.5.4 Neural quadrant error compensation (QEC SW 4)

09.95

9–50

SINUMERIK 840C (IA)

9.5.4 Neural quadrant error compensation (QEC – SW 4)

Explanation/basic The quadrant error compensation function reduces the contour errors resulting

principles from friction, backlash or torsional stresses during reversal. Errors are compen-

sated through the injection of an additional speed setpoint pulse at the instant of

zero crossing of the speed setpoint (see diagram below on left).

With software versions up to and including SW 3, the intensity of the compensa-

tion pulse can be set according to a characteristic as a function of acceleration.

This characteristic must be determined and parameterized during start-up using

external measuring instruments (see diagram below on right), i.e. it is a relatively

complicated process requiring a certain amount of experience.

t

Dnmax

Dnmin

12 3 4

a3a’3

a1a2

maximum amplitude NC-MD 12320

minimum amplitude

NC-MD 12400

Acceleration

NC-MD 12480 NC-MD 12520NC-MD 12440

Fig. 9.26

With SW 4 and higher, the manually parameterized characteristic block used to

date can be replaced by a neural network of type CMAC which offers the follow-

ing advantages:

STo facilitate start-up, the characteristic need no longer be set by the start-up

engineer. Instead, it is automatically calculated during a learning phase. How-

ever, the characteristic can be calculated correctly only if errors occurring on

the workpiece during quadrant transition are actually detected by the measur-

ing system. This means that a direct measuring system, an indirect measuring

system with distinct load reactions on the motor (rigid mechanical construc-

tion, low backlash) or appropriate compensation systems must be available.

In the case of indirect measuring systems, the backlash compensation func-

tion should be applied to compensate any backlash.

SWith the conventional QEC, the characteristic is approximated by means of a

polygon with 4 straight lines. The neural network can simulate the actual char-

acteristic shape considerably better, ensuring a greater degree of accuracy.

The characteristic resolution can be adjusted to achieve the required accu-

racy and a directional dependency of the compensation amplitude can be

taken into account. Apart from the compensation amplitude, the decay time

can also be adapted to the acceleration rate in special cases.

During the learning phase, the neural network acquires a certain operating re-

sponse, i.e. it learns a certain correlation between its input and output quantities.

In the working phase, however, no further changes whatsoever are made to the

stored characteristic.

During the learning phase, the neuronal quadrant error compensation requires a

speed feedforward control of 100 %; a setpoint filter may be required for adapta-

tion of the dynamic response.

Notes The learning and working phases as well as the resultant neural quadrant error

compensation have a purely axis-specific action, i.e. there is no mutual influence

between axes.

9 Drive Servo Start-Up Application (as from SW 3)

9.5.4 Neural quadrant error compensation (QEC – SW 4)

10.94