See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/299497059

Model-based feedforward register control of roll-to-roll web printing

systems

Article  in  Control Engineering Practice · June 2016

DOI: 10.1016/j.conengprac.2016.03.009

CITATIONS READS

3 197

6 authors, including:

Zhihua Chen Ying Zheng

Huazhong University of Science and Technology Huazhong University of Science and Technology
45 PUBLICATIONS   332 CITATIONS    79 PUBLICATIONS   647 CITATIONS   

SEE PROFILE SEE PROFILE

David Shan Hill Wong

National Tsing Hua University
205 PUBLICATIONS   3,568 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Modeling and Register control of the Speed-up Phase in Roll-to-Roll Printing Systems View project

Process monitoring and fault diagnosis of high-mix batch process with run-to-run controller View project

All content following this page was uploaded by Ying Zheng on 17 October 2017.

The user has requested enhancement of the downloaded file.

 Control Engineering Practice 51 (2016) 58–68

Contents lists available at ScienceDirect

Control Engineering Practice

journal homepage: www.elsevier.com/locate/conengprac

Model-based feedforward register control of roll-to-roll web printing

systems
Zhihua Chen a, Ying Zheng a,n, Minjing Zhou a, David Shan-Hill Wong b, Lijuan Chen a,
Zhonghua Deng a
a
National Key Laboratory of Science and Technology on Multispectral Information Processing, School of Automation, Huazhong University of Science and
Technology, Wuhan, Hubei 430074, PR China
b
Department of Chemical Engineering, National Tsing-Hua University, Hsin-Chu, Taiwan

art ic l e i nf o a b s t r a c t

Article history: Roll-to-roll (R2R) printing press provides a continuous printing process to print the multi-color patterns
Received 20 May 2015 onto a web. To correctly print patterns on webs without the register error, it is necessary to regulate the
Received in revised form web tension and transport velocity. In this paper, a mathematical model is built for R2R web printing
15 September 2015
system, and a model-based feedforward PD (MFPD) control method is proposed to reduce the effects of
Accepted 11 March 2016
interaction in adjacent print units whose print cylinders are driven by electrical line shafts and is
compared with other control methods. The proposed control method is applied to an industrial 7-color
Keywords: gravure printing register system. The register error can be controlled in the range of 70.1 mm at a steady
Roll-to-roll printing system high speed, which meets the requirement of modern industrial application.
Mode-based feedforward PD
& 2016 Elsevier Ltd. All rights reserved.
Register error
7-color gravure printing register system

1. Introduction transport conditions, web tension, web material properties and

Roll-to-roll (R2R) printing press is a multi-axis system which
offers support, transport and control for elastic material in a
continuous, flexible strip form, called “web”, such as plastics, film,
belt, foil, and fabric. The R2R printing press has been widely ap-
plied in film processing, newspaper pressing, web printing and so
on since it allows mass production with high speed and low cost
(Dong, Lévine, Jo, & Choi, 2013). Print registration in a printing
process is to align successively print patterns on the material to
ensure the print quality through appropriately controlling various
web handling variables, such as web transport velocity, web ten-
sion, web strain, etc. (Seshadri & Pagilla, 2013; Seshadri, Pagilla, &
Lynch, 2013). The register error is the position misalignment be-
tween the two adjacent overlapped patterns. Product with ac-
ceptable register error shall be classified as conforming product,
otherwise it will be classified as nonconforming product. The ex-
ample of register error is given in Fig. 1 to illustrate a properly and
an improperly registered print pattern.
In order to produce an eligible product, the web requires suc-
cessive and appropriate spacial positioning through accurate an-
gular synchronization of print cylinders. The accurate angular
synchronization, namely register control, depends on the
n
Corresponding author.
E-mail address: zyhidy@mail.hust.edu.cn (Y. Zheng).
machine installation properties, such as web transport velocity,
roller size and span length between rollers. The register control
makes use of the regulation of web tension and web transport
velocity to decrease the register errors. In this paper, the transport
velocity and web tension is regulated by gravure cylinders.
Nowadays, traditional proportional derivative (PD) approach
and the operators’ experience are adopted in register control in
industrial application, which cannot achieve the satisfactory per-
formance (Koc, Knittel, De Mathelin, & Abba, 2002). With faster
printing speed and higher printing precision required in modern
industrial factories, it is necessary to develop a new control
method to solve the coupling, time delay and nonlinear problems
(Choi, Thanh, & Kim, 2009). Much research has been done to build
dynamic mathematical models of R2R web register systems and to
develop control methods on basis of the models. Whitworth and
Harrison proposed a dynamic model to predict the change of span
tension, which formed the basis for almost all the subsequent
work on span tension dynamics (Whitworth & Harrison, 1979).
Lynch et al. designed an online nonlinear tension observer for web
machines (Lynch, Bortoff, & Röbenack, 2004). Some web tension
control methods were proposed in Fletcher (2010), Lee, Kang, and
Shin (2012), and Kang, Lee, Shin, and Kim (2011). Pagilla et al. put
forward a dynamic model for R2R web processing lines, and de-
signed a decentralized state feedback controller for the regulation
of web velocity and tension (Pagilla, Siraskar, & Dwivedula, 2007).
It improved the regulation performance of web tension compared

http://dx.doi.org/10.1016/j.conengprac.2016.03.009
0967-0661/& 2016 Elsevier Ltd. All rights reserved.
 Z. Chen et al. / Control Engineering Practice 51 (2016) 58–68
Fig. 1. Example of a properly (up) and an improperly(down) registered print
pattern.
with existing industrial decentralized proportional-integral (PI)
strategy. A dynamic tension compensation method was proposed
to use in the unwind process (Shi & Zheng-Bing, 2011). Consider-
ing the existence of non-ideal rollers, a control algorithm of web
tension and velocity was proposed in Branca, Pagilla, and Reid
(2012), which could predict the periodic oscillation of the web
tension and velocity and help to observe the dynamics of the
system. Moreover, Hur et al. built a mechanical model for the R2R
plastic film system and proposed a model-based Cross-Directional
(CD) controller which improve both dynamic and static perfor-
mance of the system (Hur, Katebi, & Taylor, 2011). Koc et al. applied
a robust control and a linear parameter varying (LPV) strategy to
control the tension and velocity of elastic materials (Koc et al.,
2002). Abjadi et al. proposed a sliding-mode feedback lineariza-
tion method for a multi-motor web-unwinding system and a ro-
bust nonlinear tracking controller (Abjadi, Soltani, Askari, & Mar-
kadeh, 2009). Benlatreche et al. presented multivariable con-
trollers with one or two degrees of freedom (DOF) to improve the
tracking and anti-disturbance performance (Benlatreche, Knittel, &
Ostertag, 2008). However, register control is not researched di-
rectly in these works.
For multi-axis printing systems, since it involves the con-
tinuous multi-stage R2R printing, it is necessary to design long-
itudinal closed-loop control strategy to reduce register error with
high precision (Choi et al., 2010; Yoshida, Takagi, Muto, & Shen,
2011). Early work on mechanical models describing the long-
itudinal dynamic characteristics of the elastic materials was found
in Whitworth and Harrison (1979). Pagilla and Knittel gave an
overview of the longitudinal control technology (Pagilla & Knittel,
2005). To avoid complicated math equations, Xin and Hoang used
position and vision sensor to measure the longitudinal displace-
ment and adopted a fuzzy logic for the longitudinal closed-loop
control (Xin & Hoang, 2011).
However, the work mentioned above do not involve the reg-
ister control of the coupled R2R system. Jeon et al. proposed a
decoupling controller for bridle rolls of the steel mill drive system,
and it was proved to be valid in controlling the tension and speed
of metal strip without any coupling effects (Jeon, Kim, Jung, Sul, &
Choi, 1997). Since the dynamic characteristics of the speed and
tension of bridle rolls of steel mill drive system are quite different
from the R2R web printing systems, this decoupling controller is
not applicable in R2R systems. In Yoshida, Takagi, Muto, and Shen
(2008a), analyzed the dynamic characteristics of R2R register
system at low steady speed,developed mathematical models of
web tension and register error, and presented a crude decoupling
PD control approach. Although its register accuracy was somewhat
improved compared with PD approach, the register error could not
be completely compensated since only the coupling of the ad-
jacent printing units is considered in this approach, so it would
59
bring additional disturbance to subsequent print units. A nonlinear
control method was proposed in Yoshida, Takagi, Muto, and Shen
(2008b). And this method had some effect on reducing register
error caused by the coupling effect, but it required on-line tension
data, which could not be provided in practice because of high cost
and installation difficulties. Similar with Yoshida et al. (2008b),
Choi et al. presented a nonlinear feedback controller based on the
assumption of existence of a lot of tension sensors (Choi et al.,
2010). The method had good performance in fast computation and
register accuracy, but it was also difficult for industrial application
since the controller parameters searched by genetic algorithm
were sensitive to initial value and required high-performance
computing devices. Moreover, as the number of printing units
increases, the coupling complexity of this method increases
greatly.
In recent years, more and more R2R systems are driven by
electrical line shafts instead of traditional mechanical shafts,
especially for printed circuit production. In hybrid roll-to-roll
printing systems which combine gravure and inkjet printing, a
mathematical modeling of the register was derived in Kang and
Baumann (2014). For the implementation of multilayered flexible
printed circuitries through the R2R mass production, a high re-
solution of register control should be achieved to prevent electrical
leakages and short circuitries. The mathematical models of ma-
chine directional (MD) register and cross direction register and
control methods were suggested in Kang, Lee, and Shin,.. (2013)
and Kang, Lee, and Shin,..... (2010), the effect of tension was dis-
cussed in Kang and Lee (2015), and the dynamic model of thermal
characteristic of register was analyzed in Lee, Shin, and Lee (2015).
Because the motors of electrical line shafts are driven in-
dividually and synchronized electrically, the register control per-
formed upstream leads to register error and tension fluctuation
downstream. The position difference with respect to the down-
stream cylinders changes with upstream position difference. In
this paper, a Model-based Feedforward PD (MFPD) control algo-
rithm is proposed to reduce register fluctuations downstream. The
upstream control input is fed forward and register fluctuations is
compensated by the model-based feedforward control. A new
concept “decoupling depth” is introduced to quantify the number
of print units with register control performed upstream. The
proposed control method is applied to a 7-color gravure printing
press. According to the industrial application, the proposed
method can quickly eliminate the initial error and reduce register
fluctuations downstream efficiently.
2. Mechanical model
The technological nomenclature used in this section and fol-
lowing sections can be referred in Table 1.
The structure of R2R web printing system is shown in Fig. 2.
There are three sections in the R2R system, including the infeed
section, multi-color printing section and the outfeed section. Each
gravure cylinder corresponding to a color is driven by an in-
dividual motor and all motors are required to be electrically syn-
chronized. Driven by the unwind motor, the web enters the infeed
section and the tension is controlled. When the web goes into the
multi-color printing section, it is compressed between the im-
pression roll and gravure cylinder, a pattern and its color mark will
be printed on the web. Register errors are measured by a regis-
tration sensor. Then the web is dried in the bake oven before en-
tering the next printing unit. After all colors are printed, the web
enters the outfeed section.
The two adjacent printing units in R2R web system are shown
in Fig. 3, in which Ti and li represent web tension and web length
between the No. i and the No. (i + 1) gravure cylinder respectively,
60 Z. Chen et al. / Control Engineering Practice 51 (2016) 58–68

Table 1 tension control in infeed section will maintain the tension around
Technology nomenclature. T⁎, and the control system in unwinding and winding section (see
Fig. 7) will maintain the angular velocity around ω⁎. While due to
Nomenclature
r Radius of gravure cylinder the external and internal disturbance, there exists the variation of
E Young's modulus the tension and the velocity in the multi-axis gravure printing
A Cross-sectional area of the web system. Then Ti and ωi can be described as:
K Elasticity coefficient
L Time period for the printed processing from the No. i to the No. (i + 1) ⎧ Ti (t ) = T⁎ + ΔTi (t )
unit
⎨
⎩ ωi (t ) = ω⁎ + Δωi (t ) (1)
ω⁎ Angular velocity of gravure cylinder at steady state
Tn Web tension at steady state
According to the principle of mass conservation, the mechan-
i, j Numbers of printing unit
li
ical model of R2R web system is shown as follows (Yoshida et al.,
Web length between the No. i and the No. (i + 1) gravure cylinder
Ti(t) Web tension between the No. i and the No. (i + 1) gravure cylinder
2008a):
ω i (t ) Angular velocity of the No. i gravure cylinder ⎧ dΔTi (t ) rω⁎ r (1 + KT⁎)
ei(t) Relative error between the No. (i + 1) and the No. i unit ⎪ = (ΔTi − 1 (t ) − ΔTi (t )) + (Δωi + 1 (t ) − Δωi (t ))
⎪ dt li Kli
ΔTi (t ) Tension variation in the No. i unit ⎨
⎪ dei + 1 (t ) Krω⁎
Δωi (t ) Variation of angular velocity of the No. i gravure cylinder ⎪ = (ΔTi (t ) − ΔTi − 1 (t − L ))
⎩ dt (1 + KT⁎) 2 (2)
ei(s) Laplace transform of ei(t)
ΔTi (s ) Laplace transform of ΔTi (t )
where K = 1/(EA) is the web tensile elongation; and E is Young's
Δωi (s ) Laplace transform of Δωi (t )
Gp(s) Laplace transfer function of the current printing unit modulus, A is the cross-sectional area of the web; L is the time
Gij(s) Laplace transfer function between the No. j and No. i unit period for the printed mark processing from the No. i printing unit
Δωf i (s ) Feedforward compensation of the No. i printing unit to the No. (i + 1) printing unit, which is defined as L = li/(rω⁎). The
Gf ij (s ) Laplace transfer function of the feedforward controller web length increases by a factor of (1 + KTi ) when the tension Ti is
added to the web. In this paper, the stability problem is not
considered.
As mentioned in Section 1, due to the cost and installation
difficulties, it is impossible to install high-precision tension sen-
sors in each print unit. Thus the web tension is unknown during
the printing process. The direct relationship between Δωi and ei is
necessary to be built up.
The printed mark of the first printing unit is regarded as the
second print unit's reference, thus it is unnecessary to control the
first printing unit, so Δω1 = 0. In particular, T0 represents the
tension in first unit, which is a constant T0 (t ) = T⁎ and ΔT0 (t ) = 0.
Laplace transforming mechanical model (Eq. (2)), we can obtain:
⎧ a b b
⎪
⎪ ΔTi (s ) = s + a ΔTi − 1 (s ) + s + a Δωi + 1 (s ) − s + a Δωi (s )
⎨
⎪ e (s ) = c ΔT (s ) − c e−Ls ΔT (s )
⎪
⎩ i+1 i i−1
Fig. 2. The diagram of multi-axis gravure printing system. s s (3)
rω⁎ r (1 + KT⁎) Krω⁎
where a = ,b= ,c= .
li Kli (1 + KT⁎)2
The second printing unit is discussed first. Because of
Δω1 = 0, ΔT0 = 0, we can obtain the following from Eq. (3):
⎧ b b
⎪
⎪ ΔT1 (s ) = s + a Δω2 (s ) − s + a Δω1 (s )
⎨
⎪ e (s ) = c ΔT (s )
⎪
⎩ 2 1
s (4)

Then we can get the relationship between the register error

and angular velocity as follows from Eq. (4):

Fig. 3. Simplified structure of two adjacent printing units. 1 a r

e2 (s ) = Δω2 (s )
1 + KT * s + a s (5)
ωi is the angular velocity of the No. i gravure cylinder, ei represents Similarly, the model of the No. i printing unit can be obtained
the relative error between the No. (i + 1) and the No. i printing according to Eq. (3) as follows:
unit, and r is the radius of gravure cylinder. As shown in Fig. 3,
i−1
each printing unit includes an impression roller, a gravure cylinder
and an error scanning sensor. The error scanning sensor samples
ei (s ) = G p (s )Δωi (s ) + ∑ Gij (s)Δωj (s) (i > 2)
j=2 (6)
the error when the gravure cylinder turns one circle. In particular,
T is composed of T⁎ which represents the controlled web tension in where Gp(s) is the transfer function of the current printing unit as
the infeed section and ΔTi which is the tension variation in the No. follows:
i print unit. ωi is composed of a controlled web feed speed ω⁎ and a r a
variable Δωi , which is the output of the register controller. ω1 is G p (s ) =
1 + KT⁎ s (s + a) (7)
the angular speed of the first cylinder, specifically, ω1 = ω⁎, because
the registration is unnecessary for the first pattern. When the and Gij(s) is the transfer function between the input of the No. j
multi-axis gravure printing system is operated normally, the unit and the output of the No. i unit as follows:
 Z. Chen et al. / Control Engineering Practice 51 (2016) 58–68 61

⎧ ⎛ s ⎞⎟
⎪ − G p (s ) ⎜ e−Ls + (i − j = 1, i ≥ 3)
⎪ ⎝ s + a⎠
⎪ ⎛
Gij (s ) = ⎨ a ⎞ s ⎛ a ⎞i − j − 2
⎪ G p (s ) ⎜ e − s + a ⎟ s + a ⎝⎜ s + a ⎠⎟
−Ls

⎪ ⎝ ⎠
⎪
⎩ (i − j > 1, i ≥ 3) (8)

3. Model-based feedforward PD control

3.1. Model-based feedforward algorithm

In order to derive the model-based feedforward algorithm, we

first transform the model (Eq. (6)). To compensate the register
fluctuations caused by the upstream print units, the following
equation is derived as follows:
i−1
G p (s )Δωf i (s ) + ∑ Gij (s)Δωj (s) = 0
j=2 (9)

where Δωf i (s ) is the feedforward compensation of the No. i

printing unit. It can be derived from (Eq. (9)) as follows:
i−1
Δωf i (s ) = ∑ Gf ij (s)Δωj (s)
j=2 (10)

where Gf ij (s ) is the transfer function of the feedforward controller

for compensating for the coupling Δωj and ei of the No. j unit on
Fig. 4. The diagram of R2R feedforward decoupling PD register control system.
the No. i (i < j ) unit. It can be obtained from (Eqs. (6)–(8)) and (Eq.
(10)) as follows:
⎧ −Ls s simulation analysis is given below to show the different perfor-
⎪ e + s + a (i − j = 1, i ≥ 3) mance when different “decoupling depth” is adopted. According to
⎪
Gf ij (s ) = ⎨ ⎛ ⎞ Eq. (6), the register errors can be obtained as follows:
⎪ ⎜ a − e−Ls ⎟ s ⎜⎛ a ⎟⎞
i−j−2
(i − j > 1, i ≥ 3)
⎪
⎩⎝ s + a ⎠ s + a ⎝ s + a⎠ ⎧ e2 = G p (s )Δω2 (s )
(11)
⎪
Based on Eq. (11), we can observe if parameter variations are ⎪ ⎛ −Ls s ⎞
⎪ e3 = G p (s )Δω3 (s ) − ⎜⎝ e + s + a ⎟⎠ G p (s )Δω2 (s )
caused by reasonable measure error, it can be regulated by the PD ⎪
controller. If there are technical parameter variations, they will ⎪ e = G (s )Δω (s ) − (e−Ls + s ) G (s )Δω (s )
cause the compensation variation in MFPD controller. The com- ⎪ 4 p 4
s+a
p 3
⎪
pensation variation will reduce the effect of the parameter varia- ⎪ ⎛ a ⎟⎞ s
+ ⎜ e−Ls − G p (s )Δω2 (s )
⁎ l
tion. For example, if r , ω⁎, li change, the parameters a = rωl , L = rωi ⁎ ⎪ ⎝ s + a⎠ s + a
i ⎪
in Eq. (11) will vary, thus Gf ij (s ) and Δωf i (s ) will be updated. Since ⎪ e5 = G p (s )Δω5 (s ) − ⎜⎛ e−Ls + s ⎟⎞ G p (s )Δω4 (s )
⎪ ⎝ s + a⎠
the feedforward compensation is open-loop, it will not cause ⎪
stability problem. Furthermore, if the web is not changed, the ⎪ ⎛ a ⎟ s⎞
⎪ + ⎜ e−Ls − G p (s )Δω3 (s )
parameter variation is trivial, the variation of Gf ij (s ) and Δωf i (s ) is ⎨ ⎝ s + a⎠ s + a
⎪
trivial too. So the MFPD controller is robust enough to the para- ⎪ + ⎜⎛ e−Ls − a ⎟⎞ s a
G p (s )Δω2 (s )
meter variation in practical R2R web system. ⎪ ⎝ s + a⎠ s + a s + a
⎪
⎪ ei = G p (s )Δωi (s ) − ⎜⎛ e−Ls + s ⎟⎞ G p (s )Δωi − 1 (s )
The diagram of R2R web system with the proposed controller is
shown in Fig. 4. ⎪ ⎝ s + a⎠
⎪
⎛
⎪ + ⎜ e−Ls − a ⎞ s
3.2. Decoupling depth ⎟ G p (s )Δωi − 2 (s )
⎪ ⎝ s + a⎠ s + a
⎪
i−3 ⎛ ⎞
Although Eq. (9) relieves the coupling between Δωj and ei ⎪ a ⎟ s ⎛⎜ a ⎞⎟i − j − 2
⎪ + ∑ ⎜⎜ e−Ls − G p (s )Δωj (s )
theoretically, it is difficult to put it into practical application. ⎪ j=2 ⎝ s + a ⎟⎠ s + a ⎝ s + a ⎠
Firstly, full model-based feedforward algorithms are too compu- ⎪
⎪
⎩ (i ≥ 6)
tationally intensive. Secondly, accuracy loss brought by the dis- (12)
cretization of the algorithm may cause the negative effect on the
According to Eq. (12), the velocity variation upstream will affect
stability. A variable “decoupling depth” is introduced to quantify
all the downstream register error at the same time, and the cur-
the number of upstream printing units which is considered for
feedforward control. That is, when “decoupling depth”¼0, no rent register error is affected by the current and upstream velocity
upstream printing unit is considered in the control algorithm; variations with different coefficients. The feedforward compensa-
when “decoupling depth”¼ 1, only one adjacent upstream printing tion is done on register error instead of Δω to achieve ei ¼0. Thus,
unit is considered; when “decoupling depth”¼ 2, two adjacent although the current register error is compensated based on its
upstream print units are considered, and so on. Next, the upstream velocity variations, nothing is done on the compensation
62 Z. Chen et al. / Control Engineering Practice 51 (2016) 58–68
of the register error downstream. And the feedforward compen-
sation of the downstream unit still requires all the upstream ve-
locity variations. For example, the compensation in the 5th and the
6th unit uses the velocity variations of upstream unit including
Δω4 to achieve e5 = 0 and e6 = 0. But at this time Δω4 still exists, so
for the 7th unit, Δω4 is still required according to Eq. (12) to
compute the compensation to achieve e7 = 0.
Without loss of generality, we take the 2nd printing unit for
example to investigate the register fluctuation. Assume that the
inputs of other units except the 2nd unit are 0, and we get the
register errors as follows:
⎧ e2 = G p (s)Δω 2 (s)
⎪ here to regulate the register errors.
⎪ ⎛ −Ls s ⎞
⎪ e3 = − ⎜⎝ e + s + a ⎟⎠ G p (s)Δω 2 (s)
⎪ up a register control system that can realize 10-color printing units
⎪ ⎛ a ⎞ s
⎪ e4 = ⎜ e −Ls − ⎟ G p (s)Δω 2 (s)
⎨ ⎝ s + a⎠ s + a
⎪ ⎛ ⎞
⎪ e5 = ⎜ e −Ls − a ⎟ s a
G p (s)Δω 2 (s)
⎪ ⎝ s + a⎠ s + a s + a
⎪ ⎛
⎪ a ⎞ s ⎛ a ⎞i − 4
⎜ −Ls ⎟
⎪ ei = ⎜ e − s + a ⎟ s + a ⎝ s + a ⎠ G p (s)Δω 2 (s) (i ≥ 6 )
⎜ ⎟
⎩ ⎝ ⎠ (13)
It can be concluded from Eq. (13) that the influence of register
control performed by 2nd print unit propagates downstream and
causes register fluctuation downstream with a time delay. How-
ever, the influence decreases gradually. When the distance be-
tween two print units reaches 3 print units, the registration fluc-
tuation downstream will decrease rapidly. Thus in practice, it is
not necessary to set the decoupling depth too large. Moreover, a
larger decoupling depth will lead to more complicated transfer
function, larger discretization error, more sensitivity to dis-
turbances. On the other hand, if the decoupling depth is too small,
the decoupling and the compensation are incomplete, which will
fail to reduce register fluctuations downstream. So it is very im-
portant to choose an appropriate decoupling depth. Generally, we
can combine the analysis of the transfer functions such as Eq. (13)
and the simulation result to determine the decoupling depth,
which is discussed in detail in Section 4.1.
4. Industrial examples
An industrial 7-color gravure printing system is taken as an
example in this section. The gravure printing system in a Chinese
enterprise is shown in Fig. 5. Its line sketch is shown in Fig. 6.
Fig. 5. An industrial 7-color gravure printing system.
The rotogravure printing press in this experiment is made in a
machinery manufacture plant located in Shantou, China, which
can be divided into 5 sections: the unwinding section, the infeed
section, the multi-color printing section, the outfeed section and
the winding section which is shown in Fig. 7. In the unwinding
section and the infeed section, the web is spread out and the
tension of web is controlled within a steady level by a tension
control system. Then the web is fed into the multi-color printing
section at a steady speed. In the winding section and the outfeed
section, the web is continuously and smoothly rolled up. In the
multi-color printing section, each printing unit is controlled by a
register controller and the proposed control method is applied
Using industrial PC and movement controller combined, we set
at most. The register control system is consisted of the controller,
servo motor, gravure cylinder and register error sensor, which
measure the deviation by photoelectric detector and photoelectric
encoder. The structure chart is shown in Fig. 8. The register error is
feedback to the register controller and the control signal is cal-
culated according to the proposed algorithm and sent to the servo
motor to regulate the speed of gravure cylinder. The change in the
speed of the gravure cylinder can cause the change in the web
tension, then leads to the change of webs longitudinal length,
ultimately results in the decrease of the register error.
The technical parameters are shown in Table 2.
Using the technical parameters in Table 2, we can set up the
mechanical model of this R2R web printing press according to Eq.
(3):
⎧ 1
⎪ ΔTi (s ) = ΔTi − 1 (s )
⎪ 4.235s + 1
⎪ 1
⎪ + Δωi + 1 (s )
⎪ 0.0147s + 0.00346
⎨
⎪ 1
− Δωi (s )
⎪ 0.0147s + 0.00346
⎪
⎪ e (s ) = 0.000366 ΔT (s ) − 0.000366 e−4.235s ΔT (s )
⎪
⎩ i+1 i i−1
s s (14)
The simulation starts with an initial register error 3 mm on the
2nd printing unit while those of other printing units' are zero. A
traditional PD controller is adopted in the factory currently. A lot of
experiments are required to identify the PD parameters before it is
applied to a new R2R printing press. In the experiments, a lot of
testing web is generally used to tune the PD controller. When the
traditional PD controller is adopted, the register errors of the 2nd
 Z. Chen et al. / Control Engineering Practice 51 (2016) 58–68 63

Fig. 6. A line sketch of 7-color gravure printing system.

Fig. 7. The rotogravure printing press.

printing units is offset by the feedforward compensation. So the

demand of the testing web is greatly reduced. The appropriate
“decoupling depth” has been analyzed in Section 3.2. The simu-
lations with the model-based feedforward algorithm are carried
out with the “decoupling depth” from 1 to 5. The register errors
are shown in Fig. 10. Fig. 10(a) shows that the register errors of the
Fig. 8. Industrial PC and movement controller combined system. 3rd and the 7th unit with the “decoupling depth” ¼1. As shown in
Fig. 10(a), the register error of the 3rd unit decreases greatly and
Table 2 maintains in the range of 70.1 mm, while the register error of the
Technical parameters. 7th unit still has large register error fluctuations which is out of
the required register error range. This indicates that the “decou-
Parameters Values Measurement unit
pling depth”¼1 has improved the performance of the 3rd unit, but
Span length (li ) 7.06 m has no enough positive influence on the performance of the up-
Radius of gravure cylinder (r ) 0.68 m stream units. If the “decoupling depth” is changed from 2 to 5, the
Angular velocity of gravure cylinder at steady 100 m/mim response register errors of the 7th unit are shown in Fig. 10
state (ω⁎)
(b) respectively. When the “decoupling depth” increases from 1 to
Web tension at steady state (T⁎) 100 N
Elasticity coefficient (K ) 1/N 3, the system performance has been obviously improved and
2.3 × 10−4
register errors can reach and remain in the range of 70.1 mm
quickly. When the “decoupling depth” is increased from 3 to 5, the
to 7th printing units are shown in Fig. 9. The register error is re- performance is improved slightly, while the calculation amount
quired to be maintained in the range of 70.1 mm. It is shown in increases greatly. Compromising on the performance and im-
Fig. 9(a) that the register error of the 2nd printing unit remains in plementation limitations, we choose “decoupling depth”¼ 3. So
the range of 70.1 mm after 27 periods. The register errors of the Eq. (10) can be rewritten as follows:
3rd to 7th printing units shown in Fig. 9(b) indicate that there are
⎧ i−1
large oscillations out of the range of 7 0.1 mm in 0–110 periods, ⎪ ∑ Gf (s )Δωj (s ) (i = 3, 4)
which will result in a huge waste of printing materials. ⎪
⎪ j=2
ij

Δωf i (s ) = ⎨
⎪ i−1
4.1. The determination of “decoupling depth” ⎪ ∑ Gf ij (s )Δωj (s ) (i ≥ 5)
⎪
⎩ j=i−3 (15)
The proposed MFPD controller is applied in this subsection.
With this controller, the PD parameters can be used in most R2R where Gf ij (s ) is derived by Eq. (11). According to the parameters in
printing press, since the tension change caused by upstream Table 2, Gf ij (s ) can be obtained as follows:
64 Z. Chen et al. / Control Engineering Practice 51 (2016) 58–68

300 100
e2 e3
e4
80 e5
250
e6
e7
60
200
Error(0.01mm)

Error(0.01mm)
40
150
20

100
0

50
−20

0 −40
0 20 40 60 80 100 120 140 160 0 20 40 60 80 100 120 140 160
Periods(n) Periods(n)

(a) Output of the 2nd unit (b) Outputs of the 3rd to 7th unit
Fig. 9. The simulations of the 2nd to 7th printing units with PD controller: (a) output of the 2nd unit and (b) outputs of the 3rd to 7th unit.

40 15
e3 depth=2
e7 depth=3
20 depth=4
10
depth=5

0
5
Error(0.01mm)
Error(0.01mm)

−20
0
−40

−5
−60

−10
−80

−100 −15
0 50 100 150 200 0 50 100 150 200
Periods(n) Periods(n)

(a) Outputs of the 3rd and 7th unit with depth of 1 (b) Outputs of the 7th unit with depth of 2 to 5
Fig. 10. The register error by MFPD controller with different “decoupling depth”: (a) outputs of the 3rd and 7th unit with depth of 1 and (b) outputs of the 7th unit with
depth of 2 to 5.

8 Table 3
e3 Comparison of the absolute maximum error of the 3rd to 7th unit with traditional
e4 PD and MFPD with depth¼ 3 (in mm).
6
e5
e6 Printing unit number 3 4 5 6 7
4 e7
Traditional PD control 0.9052 0.7685 0.6978 0.6605 0.6268
MFPD control of depth¼ 3 0.0506 0.0427 0.0404 0.0761 0.0764
2
Error(0.01mm)

0 Table 4
Comparison of the settling time of the 2nd to 7th unit with traditional PD and
−2 MFPD with depth¼ 3 (in periods).

Printing unit number 2 3 4 5 6 7

−4
PD control 27 42 59 74 87 99
−6 Feedforward decoupling PD control of depth¼ 3 23 1 1 1 1 1

−8
⎧ 4.235s
0 50 100 150 200 ⎪ e−4.235s + (i − j = 1 )
⎪ 4.235s + 1
Periods(n) G f ij (s) = ⎨
⎪ 4.235 s 4.235 s
− e−4.235s (i − j > 1 )
⎪
Fig. 11. The register errors of the 3rd to 7th printing units with “decoupling ⎩ (4.235s + 1)i − j (4.235s + )i − j − 1 (16)
depth”¼ 3.

where i and j are defined as: i ≥ 3, j = 2, 3, … , i − 1.

The register errors of the 3rd to 7th printing units with
 Z. Chen et al. / Control Engineering Practice 51 (2016) 58–68 65

350 8
e2 e3
6 e5
300
e7
4
250

2
200
Error(0.01mm)

Error(0.01mm)
0
150
−2
100
−4

50
−6

0 −8

−50 −10
0 50 100 150 200 0 50 100 150 200
Periods(n) Periods(n)

(a) Register error of the 2nd printing unit (b) Register errors of the 3rd, 5th and
7th printing units
Fig. 12. The register errors of the R2R printing system with white noise: (a) register error of the 2nd printing unit and (b) register errors of the 3rd, 5th and 7th printing units.

Table 5 4.2. Anti-disturbance ability

The absolute maximum error and settling time of the 2nd to 7th unit with white
noise.
Printing unit number 2 3
The absolute maximum error 3 0.0833 0.0661 0.0613 0.0735 0.0862
(in mm)
Settling time (in period) 24 1
“decoupling depth” ¼3 are shown in Fig. 11. Compared with Fig. 9,
the absolute value of the register errors in Fig. 11 is much smaller
and the errors of all printing prints can reach and remain in the
required region of 70.1 mm. The comparison of the absolute va-
lue of the maximum register errors and the settling time of the 3rd
to 7th printing units using traditional PD and MFPD are shown in
Tables 3 and 4 respectively. It is indicated in Table 3 that the
maximum register errors controlled by MFPD controller are almost
decreased to one-tenth of those used by the traditional PD con-
troller. Moreover, Table 4 shows that the settling time of all
printing units controlled by MFPD controller is much smaller than
those by traditional PD controller, which demonstrates that MFPD
controller has effectively suppressed the register fluctuations
downstream.
The noise sensitivity simulations of the proposed MFPD con-
4 5 6 7 troller are carried out in this subsection and a band-limited white
noise with the power 0.1 is added to the R2R printing system.
It is shown in Fig. 12 that the register errors are almost in-
sensitive to the white noise. Comparing Table 5 to Tables 3 and 4,
1 1 1 1
the absolute value of the maximum errors and settling time of the
3rd to 7th unit are very close, and all register errors maintain in
the range of 70.1 mm. This indicates that the proposed control
algorithm has a certain anti-disturbance ability.
4.3. Comparison with other control methods
On the register control of sectional drive rotogravure printing
press, two different control methods, i.e. Model Predictive Control
(MPC) (Zhang, Dubay, & Charest, 2015) and Nonlinear Control (NC)
(Yoshida et al., 2008b), are compared in this subsection.
As shown in Fig. 13, the MPC can achieve a better performance
in the 2nd printing unit, but the performance of the following
printing units is worse. On the other hand, MPC needs to real-time
calculate the matrix inversion. When there are more than
7 printing units in the R2R system, the dimension of transition

x 10
x 10 e2MPC
3
e2MPC e3MPC
4 e4MPC
2.5 e3r
e4r e5r
e2MFPD
Register Error (m)

2 e5r 2
Register Error (m)

e2MFPD e3MFPD
1.5 e3 e4MFPD
0 e5
e4
1
−2
0.5
−4
0

−0.5
0 20 40 60 80 100 120 0 20 40 60 80 100 120
Sample Time (Peroid) Sample Time (Peroid)

(a) Comparison with MPC in Printing Unit 2 (b) Comparison with MPC in Printing Unit 3
and 4
Fig. 13. Comparison with model predictive control: (a) comparison with MPC in printing unit 2 and (b) comparison with MPC in printing unit 3 and 4.
66 Z. Chen et al. / Control Engineering Practice 51 (2016) 58–68

x 10
3 x 10
0.5
e2MFPD
2.5 e3MFPD
0
e4MFPD
e5MFPD e2MFPD
2
Register Error (m)

−0.5

Register Error (m)

e2NC e3MFPD
1.5 e3NC e4MFPD
−1
e4NC e5MFPD
1 e5
·Ç −1.5 e2NC
e3NC
0.5 −2 e4NC
e5
0 −2.5 ·Ç

−0.5 −3
0 20 40 60 80 100 120 0 20 40 60 80 100 120
Sample Time (Period) Sample Time (Period)

(a) Comparison with NC in Printing Unit 2 (b) Comparison with NC in Printing Unit 3 and
4
Fig. 14. Comparison with nonlinear control.

300 20 25
model output model output model output
plant output 15 plant output plant output
20
250
10
15
200 5
10
Error(0.01mm)

Error(0.01mm)
Error(0.01mm)

0
150 5
−5
100 0
−10
−5
50 −15

−20 −10
0
−25 −15

−50 −30 −20

0 10 20 30 40 50 60 70 0 10 20 30 40 50 60 70 0 10 20 30 40 50 60 70

Periods(n) Periods(n) Periods(n)

(a) Register error of the 2nd unit (b) Register error of the 5th unit (c) Register error of the 7th unit
Fig. 15. The simulation and practical register error of 2nd, 5th and 7th unit.

Table 6 between units. So it is difficult for practical implementation of this

The register errors of the 2nd to 7th printing units of the industrial printing press method. The comparison result is shown in Fig. 14, which indicates
(in mm).
that the register errors by NC controller are much larger than
Periods Unit no. those by the proposed MFPD controller.

2 3 4 5 6 7 4.4. The industrial application

2 2.989 0.053  0.074  0.021  0.056 0.190
7 2.179 0.021  0.105  0.032  0.077 0.105
The proposed model-based MFPD controller is applied to a
12 1.063  0.011  0.063 0.032 0.039 0.168 7-units of the industrial printing press shown in Fig. 5 with the
17 0.305  0.032  0.021 0.168 0.197 0.242 technical parameters shown in Table 2. In this subsection, the
22  0.063 0.032 0 0.053 0.081 0.084 register error of the 2nd printing unit also starts with an initial
27  0.116 0.042 0.011  0.200  0.014  0.032
error of 3 mm, while other printing units' initial errors are zero.
32  0.126 0.042  0.021  0.284  0.087  0.137
37  0.074 0.021  0.021  0.168  0.224  0.158 The comparison of register errors of the 2nd, 5th and 7th units of
42  0.063  0.021 0.021  0.021  0.151  0.074 the simulation and practical output is shown in Fig. 15. The com-
47  0.053  0.063 0.032 0.105 0.074  0.158 parison indicates that the simulation has a good match with the
52 0  0.042 0.032 0.126 0.102  0.147 practical system, which proves the accuracy of the proposed
57 0.032 0 0.011 0.053 0.165  0.042
62 0.042 0  0.011 0.042 0.207 0.021
model (Eq. (14)) and the validity of the simulation analysis in
67 0  0.011  0.053 0.053 0.165 0.042 Sections 4.1 and 4.2. However, there is a little mismatch between
model and real plant in Fig. 15. Firstly, due to the limitation of the
manufacturing process, many parameters of the product are un-
matrix will reach 240  18. It is very difficult to exploit the MPC certain. Secondly, linear approximation is performed in the model,
online because of too heavy calculation. which will cause deviations in high frequency part. Thirdly, the
The Nonlinear Control (NC) method is referenced from Yoshida gravure printing press in the paper is a new machine, so different
et al. (2008b). The control method proposed in this study requires parts need running-in. Moreover, many disturbances such as ma-
real-time tension data. However, current industrial printing terial characteristics, air humidity, and the temperature in bake
presses are not furnished with sensors to measure the tension oven will also affect the accordance between the model and real
 Z. Chen et al. / Control Engineering Practice 51 (2016) 58–68 67

100 40
Model−based feedforward PD Model−based feedforward PD
Empirical decoupling PD Empirical decoupling PD
80
30
60

40 20

Error(0.01mm)
Error(0.01mm)

20
10
0

−20 0

−40
−10
−60

−80 −20
0 20 40 60 80 100 120 140 160 0 20 40 60 80 100 120 140 160
Periods(n) Periods(n)

(a) Output of the 3rd unit (b) Output of the 4th unit
Fig. 16. The register error of the 3rd and 4th units.

Table 7 the principle of mass conservation and linear approximation

Comparison of absolute maximum errors of the 3rd to 7th units with MFPD and method. The dynamics of register fluctuation caused by register
traditional PD (in mm).
control performed upstream is described in detail. Then the con-
Control method Unit no. trol law is derived from this dynamics. A MFPD control method is
proposed to suppress the downstream propagation of register
3 4 5 6 7 fluctuations. The effectiveness of the proposed MFPD controller is
verified by simulation and experiments through a 7-unit roto-
MFPD 0.0842 0.0842 0.0737 0.1053 0.1263
Traditional PD 0.85 0.35 0.23 0.17 0.13 gravure printing press. The results show that MFPD controller can
reduce initial register error rapidly and keep register fluctuations
within 70.1 mm, which cannot be realized by the traditional PD
Table 8
control, MPC and nonlinear control method. Furthermore, the
Comparison of the settling time of the 3rd to 7th units with MFPD and traditional proposed MFPD control method does not require the realtime
PD (in period). tension data that is hard to measure in each section, which makes
it easy to be taken into practice in industrial printing presses.
Control method Unit no.

3 4
MFPD 1 1
Traditional PD 69 71
plant. Some detailed register errors of the 2nd to 7th units are
shown in Fig. 15 and are listed in Table 6.
It can be observed in Table 6 and Fig. 15 that when the printing
press is under the proposed MFPD controller, the register error of
the 2nd unit remains in the range of 70.1 mm after 19 periods,
and the register errors of the 5th and 7th units also have the ac-
ceptable precision. The practical running data proves that the
proposed control approach can rapidly reduce the register error
and effectively reduced the register fluctuation.
Next, the comparison of MFPD controller and traditional PD
controller is shown in Fig. 16 and the detail compared data is listed
in Tables 7 and 8. The register errors of the 3rd and 4th units with
the proposed controller reach and remain in the range of
70.1 mm more rapidly and the precision are higher compared
with the traditional PD controller. Moreover, the absolute max-
imum errors and settling time of the 3rd to 7th units with MFPD
controller are much smaller than those with traditional PD. It must
be noted that the results are not limited to the technical para-
meters in Table 2. The proposed method can be applied with dif-
ferent operating conditions.
5. Conclusions
A R2R web printing press is studied in the paper. The me-
chanism model of R2R printing system is established according to
5 6 7
Acknowledgment
1 19 43
87 106 122
The authors thank the financial support for this work from
Chinese National Natural Science Foundation (61374139, 61370105,
U1501248).
References
Abjadi, N. R., Soltani, J., Askari, J., & Markadeh, G. R. A. (2009). Nonlinear sliding-
mode control of a multi-motor web-winding system without tension sensor.
IET Control Theory & Applications, 3, 419–427.
Benlatreche, A., Knittel, D., & Ostertag, E. (2008). Robust decentralised control
strategies for large-scale web handling systems. Control Engineering Practice, 16,
736–750.
Branca, C., Pagilla, P. R., & Reid, K. N. (2012). Governing equations for web tension
and web velocity in the presence of nonideal rollers. Journal of Dynamic Systems
Measurement & Control, 135, 359–370.
Choi, K.-H., Thanh, T.-T., & Kim, D.-S. (2009). A precise control algorithm for single-
span roll-to-roll web system using the back-stepping controller. In IEEE inter-
national symposium on industrial electronics, 2009. ISIE 2009 (pp. 1709–1714).
http:dx.doi.org/10.1109/ISIE.2009.5222079.
Choi, K. H., Tran, T. T., Ganeshthangaraj, P., Lee, K. H., Nguyen, M. N., Jo, J. D., et al.
(2010). Web register control algorithm for roll-to-roll system based printed
electronics. In 2010 IEEE conference on automation science and engineering
(CASE) (pp. 867–872).
Dong, E. C., Lévine, J., Jo, J., & Choi, K. H. (2013). Control of roll-to-roll web systems
via differential flatness and dynamic feedback linearization. IEEE Transactions
on Control Systems Technology, 21, 1309–1317.
Fletcher, J. (2010). Motion planning and observer synthesis for a two-span web roller
machine (Master thesis). Ontario, Canada: University of Waterloo.
Hur, S. H., Katebi, R., & Taylor, A. (2011). Modeling and control of a plastic film
manufacturing web process. IEEE Transactions on Industrial Informatics, 7,
171–178.
Jeon, S. H., Kim, J. M., Jung, K. C., Sul, S. K., & Choi, J. Y. (1997). Decoupling control of
68 Z. Chen et al. / Control Engineering Practice 51 (2016) 58–68
bridle rolls for steel mill drive system. In Industry applications conference, 1997.
Thirty-second IAS annual meeting, IAS'97, conference (pp. 119–125).
Kang, H., & Baumann, R. R. (2014). Mathematical modeling and simulations for
machine directional register in hybrid roll-to-roll printing systems. Interna-
tional Journal of Precision Engineering & Manufacturing, 15, 2109–2116.
Kang, H., & Lee, C. (2015). Effect of tension on conductivity of gravure printed ag
layer in roll-to-roll process. International Journal of Precision Engineering and
Manufacturing, 16, 99–104.
Kang, H., Lee, C., & Shin, K. (2013). Modeling and compensation of the machine
directional register in roll-to-roll printing. Control Engineering Practice, 21,
645–654.
Kang, H. K., Lee, C. W., & Shin, K. H. (2010). A novel cross directional register
modeling and feedforward control in multi-layer roll-to-roll printing. Journal of
Process Control, 20, 643–652.
Kang, H. K., Lee, C. W., Shin, K. H., & Kim, S. C. (2011). Modeling and matching design
of a tension controller using pendulum dancer in roll-to-roll systems. IEEE
Transactions on Industry Applications, 47, 1558–1566.
Koc, H., Knittel, D., De Mathelin, M., & Abba, G. (2002). Modeling and robust control
of winding systems for elastic webs. IEEE Transactions on Control Systems
Technology, 10, 197–208.
Lee, C., Kang, H., & Shin, K. (2012). Advanced taper tension method for the per-
formance improvement of a roll-to-roll printing production line with a winding
process. International Journal of Mechanical Sciences, 59, 61–72.
Lee, J., Shin, K., & Lee, C. (2015). Analysis of dynamic thermal characteristic of
register of roll-to-roll multi-layer printing systems. Robotics and Computer-In-
tegrated Manufacturing, 35, 77–83.
Lynch, A. F., Bortoff, S. A., & Röbenack, K. (2004). Nonlinear tension observers for
web machines. Automatica, 40, 1517–1524.
View publication stats
Pagilla, P. R., & Knittel, D. (2005). Recent advances in web longitudinal control. In
Proceedings of 8th international conference on web handling.
Pagilla, P. R., Siraskar, N. B., & Dwivedula, R. V. (2007). Decentralized control of web
processing lines. IEEE Transactions on Control Systems Technology, 15, 106–117.
Seshadri, A., & Pagilla, P. R. (2013). Interaction analysis of control systems employed
in roll-to-roll printing. In American control conference (ACC), 2013 (pp. 4251–
4256).
Seshadri, A., Pagilla, P. R., & Lynch, J. E. (2013). Modeling print registration in roll-to-
roll printing presses. Journal of Dynamic Systems Measurement & Control, 135,
815–826.
Shi, X. P., & Zheng-Bing, L. V. (2011). The application of dynamic web tension
control in winders. China Pulp and Paper Industry, 18, 82–84.
Whitworth, D. P. D., & Harrison, M. C. (1979). Tension variations in pliable material
in production machinery. Applied Mathematical Modelling, 7(189–196).
Xin, L. L., & Hoang, N. (2011). Lateral control of roll-to-roll system using fuzzy
control logic and vision sensor. Advanced Materials Research, 317, 1541–1544.
Yoshida, T., Takagi, S., Muto, Y., & Shen, T. (2008a). Modeling and resister control of
sectional drive gravure printing press. Transactions of the Japan Society of Me-
chanical Engineers, 74.
Yoshida, T., Takagi, S., Muto, Y., & Shen, T. (2008b). Register control of sectional drive
rotogravure printing press. In Manufacturing systems and technologies for the
new frontier (pp. 417–420). Springer, London.
Yoshida, T., Takagi, S., Muto, Y., & Shen, T. (2011). Register control of rotogravure
printing press. Application of nonlinear control theory to sectional drive
printing press. Electronics and Communications in Japan, 94, 17–24.
Zhang, S., Dubay, R., & Charest, M. (2015). A principal component analysis model-
based predictive controller for controlling part warpage in plastic injection
molding. Expert Systems with Applications, 42, 2919–2927.