1428 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 52, NO. 5, OCTOBER 2005
Design of a Self-Adaptive Fuzzy Tension Controller
for Tandem Rolling
Farrokh Janabi-Sharifi, Senior Member, IEEE, and Jingrong Liu
Abstract—A fuzzy logic controller (FLC) is designed to main-
tain constant tension for tandem rolling mills. Self-adaptive tech-
niques were introduced to optimize the proposed FLC’s parame-
ters (i.e., to make it flexible and enable it to generalize). With the in-
clusion of supervision and concern for generic control criteria, the
optimal parameters of the fuzzy inference system were either tuned
by a backward propagation algorithm or determined by means
of a genetic algorithm. In simulations, the proposed neuro-fuzzy
controller exhibited the real-time applicability, while the proposed
genetic fuzzy controller revealed outstanding global optimization
ability.
Index Terms—Backpropagation algorithm (BPA), fuzzy logic,
genetic algorithm (GA), neural networks, rolling mill, tension
control.
I. INTRODUCTION
I
N TANDEM rolling mills, long metal products with different
cross sections (such as bars) are produced through multi-
stage shaping as they proceed sequentially through mill stands
[1]. Automatic gauge controllers (AGCs) and automatic speed
regulators (ASRs) are employed to meet dimensional require-
ments and to regulate mass flow, respectively in each rolling
mill stand (Fig. 1). Ideally, desired inelastic deformations are
confined to individual stands. In principle, a rolled billet’s speed
when leaving a stand must be equal to its speed when entering
an adjacent downstream stand. Otherwise, a longitudinal force,
known as interstand tension, will result inside a billet as it travels
between two stands. Such force will cause a push or pull action
and introduce variations in the thickness and/or width of the de-
livered product, thus deteriorating product quality. In an extreme
case, excessive tension might break or deform a product or even
cause equipment damage.
It is vital, therefore, to control the interstand tension to
achieve a safe, stable, and high-quality rolling process [2].
Meanwhile, to optimize AGC and ASR performance, it is
desirable to keep tension low and constant via addtitional
control action. However, interaction effects (e.g., AGC and
ASR activities) will produce tension variations; in turn, ten-
sion maintenance activity (e.g., adjusting stand speed) might
Manuscript received MArch 19, 2003; revised April 19, 2004. Abstract pub-
lished on the Internet July 15, 2005. This work was supported by the Natural
Sciences and Engineering Research Council of Canada (NSERC) through Col-
laborative Research and Development Grant CRDPJ 234028-99 and by Mate-
rials and Manufacturing Ontario Collaborative Research Grant IC403.
F. Janabi-Sharifi is with the Department of Mechanical and Industrial
Engineering, Ryerson University, Toronto, ON M5B 2K3, Canada (e-mail:
fsharifi@ryerson.ca).
J. Liu is with the Department of Electrical and Computer Engineering,
University of Waterloo, Waterloo, ON N2L 3G1, Canada (e-mail: j5liu@king-
cong.uwaterloo.ca).
Digital Object Identifier 10.1109/TIE.2005.855653
Fig. 1. (a) Schematic view of multistand tandem rolling mill. (b)
Two-dimensional shear stress profiles under two consecutive stands.
worsen the gauge and speed control and thereby complicate the
situation. Some control schemes have been proposed [1], [3] to
keep a constant tension in tandem rolling mills; for example,
conventional proportional–derivative (PD) and proportional–in-
tegral–derivative (PID) control techniques are common control
methods [4]. Conventional controllers, however, are nonin-
teractive and cannot compensate for parameter variations
and interactions within a rolling mill system. Other control
techniques, such as optimal multivariable control [5], inverse
quadratic control [6], and control [7], have also been
proposed. In general, most of these control methods require
exact mathematical models and complete knowledge of rolling
processes. The complicated characteristics of rolling mills, lack
of delicate instruments, and noisy environments make it diffi-
cult to identify rolling process from the measurement of tension
data. Typical sources of noise include mill chatter, roll thermal
expansion, and skid marks [3]. In addition, multivariable con-
trol methods based on advanced control theories, despite taking
interaction effects into consideration, are difficult to implement
and configure. Therefore, human intervention and supervision
are necessary and common in rolling mills.
In practice, human experts (operators) are assigned to ma-
nipulate rolling stands and manage interstand tension through
manual intervention. However, high skill requirement and
inconsistencies associated with operators motivate their re-
placement with automatic control systems. A fuzzy logic
controller (FLC), though, can emulate the way the human
brain applies intuition and experience to process ambiguous
0278-0046/$20.00 © 2005 IEEE