Siemens SINUMERIK 840C Installation Instructions Manual page 672

09.95
10.94
The operating procedure from SW 3 can still be used if, for example, the condi-
tions listed under facilitation of start-up cannot be met or if there is insufficient
computing time available for the neuronal network.
Quantization of The input quantity (setpoint acceleration) is quantized before it is processed by
tof operating range the CMAC network. The entire operating range is broken down into intervals, the
output value remaining constant in each interval (see diagram below). A memory
location is assigned to each interval. The interval width increases with the accel-
eration rate in order to optimize memory requirement and learning period.
Finer quantization – involving higher memory space requirements – allows the
resolution of the characteristic to be increased (see diagram below). However,
this also prolongs the learning phase, i.e. start-up takes longer.
Compensation amplitude
Fig. 9.27
Quantization of the input quantity is set by means of the two variables fine quan-
tization c and coarse quantization q. The amount of memory space required or
the total number of quantization intervals is calculated according to the following
formula:
Amount of memory space = c x (q + 1)
A maximum of 1000 memory locations are reserved for each axis, affording suffi-
cient resolution even when a high degree of accuracy is required.
Two quantities are used to determine quantization since fine quantization c has a
further function in addition to its role in determining the interval sizes. A high fine
quantization setting results in a "similar" output value being calculated for adja-
cent intervals of the input quantity, making it possible to identify, for example,
measuring errors which only occur at a certain acceleration rate. In contrast, a
low fine quantization setting allows characteristics with sharp changes in shape
to be simulated better. For the purpose of neuronal friction compensation, the
greater error tolerance afforded by a high fine quantization setting (c setting in
the range of approx. 10) should be used. Only when the coarse quantization
value is no longer significantly higher than the fine quantization value may the
latter be reduced.
The lower the acceleration rate, the finer the quantization of the input quantity. In
the low acceleration range, a particularly high resolution is required to ensure that
the widely varying compensation values in this range can be simulated.
The following diagram shows the correlation between the quantization interval
width and the setpoint acceleration (storage utilization by means of variable node
distance).
Siemens AG 2001 All Rights Reserved 6FC5197–jAA50
SINUMERIK 840C (IA)
9.5.4 Neural quadrant error compensation (QEC – SW 4)
Coarse quantization
Setpoint acceleration
9 Drive Servo Start-Up Application (as from SW 3)
Compensation amplitude
Fine quantization
Setpoint acceleration
9–51