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