MSW Research Projects
Optimisation of Roll Profiles in the Hot Rolling of Steel
This research is a collaborative project between The Open University, The University of Glamorgan, Fachhochschule Bielefeld, and Metrology Systems Wales.
There is a world-wide over-capacity for wide steel strip. In such a buyers’ market, producers need to offer a high quality product at a competitive price in order to retain existing customers and win new ones. Producers are therefore under pressure to improve their productivity by automating as many task as possible and by optimising their process parameters to maximum efficiency and quality. One of the critical processes is the hot rolling of the steel strip.
Aim of Project
The aim of the project is to develop software to assist in the optimisation of the ingoing ground profiles of the work rolls of a hot rolling finishing mill, in order to improve the physical parameter ‘flatness’ of the strip (Figure 1). Initially the project concentrates on the hot rolling of steel, however it is hoped that the findings can be used in other rolling processes, for example cold rolling of steel or rolling of aluminium.
Figure 1 - Flat steel strip.
The flatness of the steel strip depends on a number of factors, but particularly on the profiles of the work rolls. Usually these profiles are initially specified by the mill stand producer, and are subsequently improved, usually empirically, by the rolling mill technical personnel. A system which could assist in optimising the roll profiles would clearly be of benefit.
The Rolling Operation
In a rolling mill the thickness of a steel slab is reduced by rolling between two work rolls in a mill stand (Figure 2).
Figure 2 - Principal Rolling Operation.
Figure 3 - Real Roll Stand.
As a consequence of the high forces employed, the work rolls bend during the rolling process. To compensate for the bending and thermal expansion, work rolls are usually ground to a convex or concave camber, which is sinusoidal in shape (Figure 4). Due to roll wear, the rolls need to be periodically reground after a specified duty cycle (which is normally in the order of four hours).
Because there is no suitable process model for a complete mill train, the problem is to find suitable work roll profiles - for each rolling programme - capable of producing strip-flatness to the specified tolerances.
Figure 4 - Initially ground profile with 70° - sine curve segment shape ( not to scale!)
Figure 5 - Pair of Work Rolls.
Figure 6 shows how the initially ground camber can ideally compensate for the effects of bending and expansion.
Fig 6(a) Unloaded Rolls
Fig 6(b) Loaded Rolls
The "Intelligent Roll Profile Optimisation System" should be able to suggest optimum profiles for all the work rolls in a hot strip finishing mill while minimising the number of different profiles required.
Traditional optimisation methods are local in scope, i.e. they tend to find only local optimas. They depend on the existence of derivatives of the error function, this existence cannot be ensured in this real-world-problem: The error function of the technical systems "mill train" is disturbed from noise and is not smooth, i.e. a derivation may not exist for all points of function. This class of problems cannot be solved by mathematical methods. Genetic Algorithms are here an alternative. Due to a lack of access to real hot rolling mill trains for experimenting, the use of a process model would seem to be unavoidable. Due to the absence of a suitable analytical process model the transfer function of a particular finishing train should be learned by a neural network (Figure 7).
Figure 7 - The suggested system for roll profile optimisation
It seems likely, that the approach presented can be used not only for the optimisation of the initial ground work rolls in a hot strip mill, but also in a wide range of real-live combinatorial optimisation problems where experiments are impractical on the real process and no analytical knowledge of the process - and hence no process model - is available.
Nolle, Armstrong, Ware, Biegler-König: "Optimisation of Roll Profiles in the Hot Rolling of Wide Steel Strip", GALESIA'97, 2-4 September 1997, Glasgow, UK, IEE Conference Publication Number 446, pp 133-138
Dr. L. Nolle
Senior Lecturer in Computing
Nottingham Trent University
Dr. D. A. Armstrong
The Open University in Wales,
Faculty of Technology
Cardiff, CF1 9SA, UK
Dr. J. A. Ware
The University of Glamorgan,
Division of Mathematics and Computing
Pontypridd, CF37 1DL, UK
Prof. Dr. F. Biegler-König
Fachbereich Mathematik und Technik
33609 Bielefeld, Germany