Ootimurn seDaration of machine-d

aGd cross-dikction produ

The scanning measurement ch
Bmce F. Taylor

Fmtw scanning parallel measurement data procesing and adaptiveprofile trending

p m i t higher-frequency MD and CD meaurement without compromzking
the speed or accuracy of controlprofile determination.

Paper quality variability is measured and controlled in 1. Typical paper machine MD power spectrum
two dimensions: the machine direction (MD) and the cross
direction (CD). During the production process, the former
varies with respect to time ( I ) and the latter with respect
I I
plcal WID power spectrum
I

to both time and cross-machine position. The ultimate goal

of supervisory measurement and control systems is to I-
successfully control the entire spectrum of product quality
variations, from the very narrow and short-term to the power spectrum
very broad and long-term. While low-frequency variations
(with respect to both time and space) are relatively easy
to identify and control, high-frequency variations repres- R
W
ent a difficult challenge. a
Before high-frequency variations can be controlled and
0.0001 Hz 0.001 0.01 0.1 1 10 100
eliminated, they must be “seen”or measured by the system. 10,000 s 1,000 100 10 1 0.1 0.01
A scanning measurement system cannot possibly measure 167mln 16.7 1.7 0.2
2.8 h 0.3
100%of the sheet; only a portion of the sheet is “sampled”
WID VARIABILITY, frequency or period
and presumed to represent the entire product. As with any
sampling exercise, the more samples taken, the greater -Scan average Intrascan Ouality analysis+
the accuracy of the statistically-based estimate. Increasing measurement (to 200 Hz)
the sample “population” with a scanning measurement
system requires a n increase in the r a t e a t which
measurements are taken.
MD and CD measurement resolution has increased be erroneously perceived as CD variations unless they are
significantly in the last few years. Previously with scan- somehow isolated and removed from the final measured
average MD measurement, the highest measurable fre- profile.
quency was in the region of 0.017-0.025 Hz (a 40-60-s Until imported MD variability is eliminated from the
period, equal to approximately twice the scan time). As measured profile, “true” profile changes cannot be
illustrated in Fig. 1,intrascan measurement, and quality confidently identified and successfully controlled. Remo-
analysis systems, have increased process visibility to val of unwanted MD variability from the profile may be
include MD frequencies as high as 200 Hz. In addition, accomplished by weighting the measurement a t each
improved CD measurement techniques now enable profile cross-direction position to the long-term historical value
measurements as fine as 1cm wide. These improvements rather than the most recently measured value. This
are the combined result of faster, more precise sensor conventional procedure is referred to as exponential
signal processing, faster scan speeds, and innovative multiple-scan trending. In its basic form, exponential
computational methods. profile trending is quite simple and can be described by
When a sensor scans the sheet, it traces a diagonal path, a single mathematical parameter- the trend factor. As
measuring a “now” profile that includes both MD and CD noted in Fig. 3, when a trend factor of 0.2 is used, 20%
variations. As illustrated in Fig. 2, MD variations may of the most recent value will be added to 80% of the
historical value to derive the trended measurement.
As Fig. 3 illustrates, with a 0.2 trend factor it would
Taylor is senior marketing manager, profile controls, take approximately 10 scans to register 90%of a step-wise
Measurex Gorp., 1 Results Way, Cupertino, Calif. 95111. profile change. By comparison, if low MD variability
February 1991 TappiJournal
 2. Diagonal scan path introduces MD variations into "now" profile. 3. Exponential profile trending

Actual profile
Pure MD variation
.(no CD variability)
change at
position **x#*--l1-
c
= w1X0.8)+(2X0.2)=1.2

Measured "now" profile

Measured profile change
at positlon 'X, using a
0.2 trend factor

Unlt step change in

measured property at time 'T

- 4 - 2 0 2 4 6 8 10
SCAN NUMBER

permitted the use of a more aggressive 0.3 trend factor,

4. Scan speed affects "now" profile variability
about 90% of the unit profile change could be detected in
only 6 scans, to provide 40% faster profile measurement
response. I ~ --1
The ongoing development of higher-resolution measure-
ment technologies increases the importance of methods

-
used to eliminate MD variability from cross-direction
profiles prior to trending. Without effective alternatives, speed

increasing measurement resolution will necessitate more Dalabox

width(w)- Scan path Typical
conservative exponential trending, resulting in slower "cut-off"
profile measurement and control response. Of course, this frequency
curve
would be undesirable. The remainder of this paper will \ I
Measurement
review various techniques designed to improve the lrequency (l/t)
Representative MD power
response of high-resolution measurement systems to "true" spectrum
profile changes.
speed is reduced

The scanning measurement challenge tPeriod Frequency +

MD VARIABILITY, frequency or period
The scanning measurement challenge is to ensure that
advances in measurement resolution do not compromise
the speed and accuracy of CD profile measurement and
control. New, effective ways to quickly remove MD
variations from the CD profile are required. One suggested 5. Serial measurement data processing
way to accomplish this is to scan more slowly to increase
the averaging time for each cross-direction measurement
zone (often referred to as a databox or minislice) (2). As
the scan speed is decreased, the "cut-off" frequency, shown
in Fig. 4, moves to the left, thereby reducing the amount
of high-frequency MD variability that may be introduced
into the "now" profile. Unfortunately, while fewer scans
ADC with
will be required to "trend-out" the remaining MD arithmetic averaging
variability, the time required to complete each scan
increases proportionally. The net effect is a trade-off
between measurement speed and accuracy. In addition, I "Now" databox p r o i e 1
decreasing scan speed reduces MD measurement resolu-
tion directly, opposing new high-speed measurement
efforts such as intrascan MD measurement.
As illustrated in Fig. 5, this slow-scanning method
typically employs serial measurement data processing. An High-resolution Actuator resolution
analog filter first attenuates the sensor signal (to reduce display profile control profile
sensor signal noise), then an analog-to-digital converter
(ADC) samples the residual signal approximately every
10 ms. Finally, n samples are averaged within a databox
88 February 1991 TappiJournal
6. Alternative signal processing techniques: signal sampling 7. Alternative signal processing techniques: signal integration

+
t
Minislice transformation time
averaging time
c- A!-
\
Integration
time
Sampling
time m r
Faster scan path

Faster scan path

Minislice
width (w)
-
Slower scan path

8. Parallel measurement data processing

to produce the databox measurement. As shown in Fig.
6, slower scanning improves the accuracy of this sampling
technique by increasing the number of samples per
Q Sensor voltage databox.
An alternative method uses faster scanning to trend-out
MD variability more quickly by completing multiple scans
in less time. This fast-scanning technique employs a 512-
kHz voltage-to-frequency converter (VFC) and a pulse
counter to integrate pulses every 6.25 ms. By integrating
Scan Interpolatlo Transformation +
100% of the sensor signal (Fig.7), much of the spurious
variability (“noise”) generated by sampling techniques is
eliminated to provide a more accurate measurement. As
a result, faster scan speeds can be used without compro-
mising accuracy.

From a profile control standpoint, the measurement

system’s goal is to develop accurate actuator resolution
profiles as quickly as possible, allowing faster response to
process transients following breaks or grade changes., By
comparison, analytic exercises like streak analysis, which
use higher-resolution profiles (down to 1 cm), are not
~~

9. Faster scanning enables measurement of higher-frequency MD

variability. particularly time-critical. As shown in Fig. 8, the fast-
scanning method satisfies each specific requirement by
processing “now” minislice data (equivalent to databoxes)
through different channels for different end uses.
Faster scanning enables higher-frequency MD mea-
surement (Fig. 9). While this is desirable for MD
measurement and control purposes, it will increase the
“now” minislice or databox profile variability.
To compensate for the increased MD variability
imported into the “now” minislice profile, parallel
measurement data processing transforms it into the
appropriate actuator resolution control profile prior to
trending (Fig. 10). This simple exercise immediately
removes much of the unwanted MD variability from the
2.8 h 0.3
“now” control profile. Consequently, less trending, fewer
MD VARIABILITY, frequency or perlod
scans, and less time is required to accurately measure
control profile changes.
By comparison, because high-resolution ‘‘now”measure-
ment profiles must contain more MD variability, trending
from “now” databoxes into databox profiles (as shown in
February 1991 TappiJournal
d

10. Transforming "now" minislice profiles into "now" control profiles 11. Transforming into "now" control zones before trending reduces
the effect of hiqh-freauencv MD variabilitv.
3 Typical 30-Hz w

r High-resolution "now"
minislice Drofile
MD frequency

Minislice width-
(e.g., 1 cm)
it Y
4
H
3
a
Example
Typlcal control zone wldth:
6 cm (half sllce) to
12 cm (full sllce)
- -
Typlcal2.5 5-HZ
fast-scan "cut-off
frequency range affer
transformlng Into
Typlcal "cut-off' "now" control profile
frequency range:
e (30112) to (3016)
= 2.5 - 5 HZ
Typlcal 16-Hz
slow-scan
"cut-off
frequency

w
n ,
0 . m 1 Hr
10,Ooo8
167min
2.8 h
0.001

'i?
0.3
0.01
100
1.7
0.1
10
0.2
1
1
10
0.1 1 100
ani

M
Control zone width Typical 3aHz fast-scan
M D VARIABILITY, '*cut+r frequency
'*now'*control profile frequency or period beforetransformln Into
-
Typical 2.5 5-HZ
MD frequency range
- "now" control profie

12. Reducing scan speed in the presence of low-frequency M D 13. Faster scanning attenuates low-frequency MD variability.
variations will increasethe apparent cross-directionalvariability of the
"now" profile.

Instantaneous
scan
"Now" profiles Frequency H- "Cut-oft" frequency shift
1w -
90- $
80-
70-
60-
Typical "fast" scan
50- speed (30cmls)
40-
30-
20-

0
10-
0 . m Hz 0.001 0.01 0.1 1 10 100
10,000. l.m loo 10 1 0.1 0.01
167mln 16.7 1.7 0.2
variation 2.8h 0.3 'Based on 600-cm
(236 In.) trim width

MD VARIABILITY, frequency o r period

Fig. 4) requires more scans, and hence more time, to

determine "true" profile changes. The relative benefit of
me influence oflow-fiequency MD
trending before and after control-zone transformation is The foregoing discussion focused on the relative merits of
illustrated in Fig. 11. Transforming into control zones various scanning measurement techniques used to
moves the "cut-off" frequency far to the left, significantly eliminate high-frequency MD variability from CD
reducing the amount of MD variability that may be profiles. This "high-frequency focus" has been the topic
introduced into the ''now" control profile. of many recent technical papers (2-4). However, few
90 February 1991 TappiJournal
14. Scan speed affects low-frequency variability of “now” profiles. directional variability of the “now” profile, making it more
difficult to quickly determine accurate CD profiles.
This phenomenon is illustrated mathematically in Fig.
~ ~ _
13._As the scan speed is reduced, more of an MD variation

3
Typical 30-s
slow-scan
”cut-Off”
period 1 will be measured within a single scan. For example, with
a 20-cm/s scan rate on a 600-em-wide sheet, 1.5 cycles of
a 0.05-Hz MD variation would be imported into each ‘‘now’’
profile, whereas when the scan speed is increased to 30
3
0
I-
cm/s only one complete cycle would be imported into each
U “now” profile.
As the scan speed is increased, the “cut-off” period shown
W in Fig. 1 4 moves to the right, thereby reducing the amount
P of low-frequency MD variability introduced into the “now”

-
W
n. profile. Faster scanning essentially attenuates the spacial
0.
1 frequency of imported low-frequency MD variations so
167min 16.7 1.7 0.2 that less trending is required, resulting in faster, more
2.6 h 0.3
MD VARIABILITY, frequency or period accurate control profile determination.

me “regionof
By combining the low- and high-frequency focuses, it is
possible to define a “region of influence” bounded by the
low and high “cut-off” frequencies or periods shown in
Figs. 9, 11, and 14. The size of this region of influence
is indicative of the amount of MD variability that must
be removed from the “now” profile by multiple-scan
exponential trending. Consequently, the larger the region
of influence, the more scans and time will be required to
15. The overall ”region of influence” is reduced by fast scanning and develop accurate control profiles. As illustrated in Fig. 15,
parallel measurement data processing. the combined advantages of faster scanning and parallel
measurement data processing should result in a smaller
region of influence and hence the ability to more quickly
W
0
and accurately measure profile changes.
z
4:
E
s Adaptive profde trendhg
J

f After the “now” minislice profile is transformed into the

z
U
“now” control profile, residual MD variations must be
“trended out” to derive the “true” CD profile. The ideal
0
ii amount of trending will be proportional to the bounded
W
0 area under the MD power spectrum curve, defined as the
f5n region of influence in Fig. 15. At times, if the paper
machine is quite stable, the region of influence might be
0.0001 M 0.001 0.01 0.1 1 10 100
10,000s 1,000 100 10 1 0.1 0.01 small, while a t other times, such as after a break or grade
167min 16.7 1.7 0.2
change, the region of influence might be quite large (Fig.
2.6 h 0.3
MD VARIABILITY, frequency or perlod 16). Consequently, the ideal amount of trending for any
paper machine will vary with process conditions.
In recognition of this, rather than using a fixed trend
factor (Fig. 3), adaptive trending uses a proprietary,
patented ( 5 ) procedure to automatically match the trend
factor to the size of the region of influence, on a zone-by-
zone basis, a t the end of every scan. This approach enables
faster recognition of “true” profile changes and heavier
trending of transient product variations that are too short-
term for the control system to correct.

published discussions have dealt with the influence of low- CO~dU§igP!dTS

frequency MD variability on profile measurement
response and accuracy. To effectively measure and control cross-direction paper
As illustrated in Fig. 12, slower scanning increases the quality variations, MD variability must be quickly and
apparent CD spacial frequency of imported MD variations. thoroughly removed from control profiles. A wide range
Consequently, reducing scan speed in the presence of low- of MD variations exists on the paper machine, so
frequency MD variations will increase the apparent cross- measurement techniques must be designed to remove both
February 1991 Tappi joutnd 91
M

i
8
16. The size of the reqion of influence varies with process conditions. Faster scanning and parallel measurement data
processing have been shown to reduce the region of
influence, allowing faster, more accurate profile determi-
8 MD power spectrum:
1 nation.
unstable operation As higher-resolution measurement technologies evolve,
additional innovations designed to further reduce the
region of influence may be required to maintain and
0
b- . MD power spectrum:
stable operation improve profile measurement and control effectiveness.0
8 -
5
W Literature cited
0
wPE l /‘Region of influence”\\- 1. Bialkowski, W. L., Control Systems’90 Preprints, International
o.oooi nz 0.001 0.01 0.1 1 10 100 Conference on Instrumentation and Automation in Pulp and
10,0009 1,000 100 10 1 0.1 0.01 Paper Industry, 1990, Helsinki, p. 220.
167min 16.7 1.7 0.2
2. Adams, W. L., 1990 CPPA Annual Meeting Proceedings,
2.8 h 0.3
Canadian Pulp and Paper Association, Montreal, p. A289.
MD VARIABILITY, frequency or period 3. Wallace, B., PIMA, 72(2): 56(1990).
4. Cutshall, K. A., Tappi J. 73(6): 81(1990).
5. Hu, H. T., U.S. pat. 4,707,779 (Nov. 17, 1987).
Received for review Nov. 12. 1990.
low- and high-frequency MD variability from control Accepted N ~ 13,~1990
,
profiles.
The conceptof a of influence” bounded by low Presented at the TAPPI 1991 Process Control Conference.
and high “cut-off’’ frequencies characteristic of the
measurement system design has been introduced to aid
in the value analysis of alternative cross-direction
measurement technologies.

92 February 1991 Tappi Journal