A Review of Rolling System Design Optimisation PDF

International Journal of Machine Tools & Manufacture 46 (2006) 912–928
www.elsevier.com/locate/ijmactool
Short Survey
A review of rolling system design optimisation
V. Oduguwa, R. Roy*
Enterprise Integration, School of Industrial and Manufacturing Science, Cranfield University, Cranfield, Bedford MK43 0AL, UK
Received 3 February 2005; accepted 5 July 2005
Available online 22 August 2005
Abstract
Rapid product development and efficient use of existing resources are key competitive drivers in the steel industry and it is imperative that
solution strategies are capable of delivering high quality solutions at low cost. However, traditional search techniques for Rolling System
Design (RSD) are ad hoc and users of them find it very difficult in satisfying the required commercial imperatives. This paper presents a
comprehensive review of approaches for dealing with RSD problems over the years in terms of modelling and optimisation of both
quantitative and qualitative aspects of the process. It critically analyses how such strategies contribute to developing timely low cost optimal
solutions for the steel industry. The paper also explores the soft computing based technique as an emerging technology for a more structured
RSD optimisation. The study has identified challenges posed by RSD for an algorithmic optimisation approach, especially for evolutionary
computing based techniques.
q 2005 Elsevier Ltd. All rights reserved.
Keywords: Rolling system; Optimisation; Engineering design; Soft computing
1. Introduction Rolling System Design (RSD) is a complex problem

due to the non-deterministic nature of the process and the
Rolling is an important steel manufacturing process to large number of processing conditions involved. It is a
provide a wide range of products used in the automotive, core skill often conducted by experts using their
construction, and engineering industries. However, due to experience and intuition. These procedures are largely
increasing competition in the steel industry, there is an ever- based on trial and error and can be expensive and time
increasing demand on the steel manufacturers to become consuming. It can significantly influence the ability to
more flexible, more responsive and to be competitive with deliver new products to the markets. The overall aim of
energy efficiency. The competition is fierce, complex this paper is to provide a comprehensive review of
products are being required at higher quality for the same process optimisation approaches for solving the rolling
cost and margins are continuously being squeezed. Optimal system design problem. The paper provides a description
rolling system designs can play a significant role in dealing of the features of a rolling system design and critically
with these challenges. Motivated by the need to enhance analyse the approaches to a rolling system design. The
new products and process development strategies to deliver study looks at Quantitative (QT) and Qualitative (QL)
a spectrum of high quality design solutions at low cost, the information involved in the design. The review is
steel industry is increasingly adopting scientific approaches classified into the following areas: approximate QT
in favour of traditional approaches. Several algorithmic models, approximate QL models, manual search methods,
optimisation approaches are emerging as alternative QL and QT optimisation, soft computing approaches in
solutions to deal with the complex search space properties RSD, and multi-pass RSD. For each classification, the
of the real world process problems. review explores how the reported techniques search for
good RSD solutions in terms of efficiency and cost. The
* Corresponding author. Tel.: C44 1234 754073; fax: C44 1234 750852. paper then critically analyses how such strategies
E-mail addresses: voduguwa@hotmail.com (V. Oduguwa), r.roy@ contribute to developing timely low cost solutions that
cranfield.ac.uk (R. Roy). are optimal for the steel industry. The paper also explores
0890-6955/$ - see front matter q 2005 Elsevier Ltd. All rights reserved. soft computing based techniques as an emerging technol-
doi:10.1016/j.ijmachtools.2005.07.023 ogy for a more structured optimisation of an RSD.
V. Oduguwa, R. Roy / International Journal of Machine Tools & Manufacture 46 (2006) 912–928 913
This paper is organised as follows. Section 2 introduces The rolling system components are classified under
RSD by identifying and classifying the features of RSD general headings to show how the individual components
while Section 3 presents a comprehensive review of the of the complex system can be decomposed. It should be
RSD approaches over the years. Section 4 discusses the noted that the relationships amongst each individual
challenges of RSD as an optimisation problem and evaluates components are not mutually exclusive but in fact a
RSD in relation to new product development strategies. complex interaction of these components forms the
Finally, Section 5 presents a summary of the reviewed inherent source of the rolling system complexity. Fig. 1
approaches and concludes with an outline of the future shows how the highly complex nature of the rolling
research directions. system is composed of the product and process that
interact with multi-faceted disciplines. The RSD problem
is an important component of the rolling process. The
following sections introduce RSD and identifies the
2. Rolling system design challenges posed by the RSD optimisation.
Rolling system design is the preparation of a series of roll

pockets and draughting sequences for roll passes that are 2.1. Features of rolling system design problems
necessary to obtain a group of roll profiles. The purpose of
RSD is to ensure the correct production of roll profiles RSD problems like most real world process problems can
within the constraints of the mill with acceptable qualities, be characterised as having chaotic disturbances, random-
minimum cost [1] and maximum output. This problem can ness, and complex non-linear dynamics. This is due to the
also be referred to as a multi-pass search problem for elastic–plastic material behaviour and potentially large
multiple passes. Effective RSD aims to achieve the optimum deformation behaviour [3]. It is widely accepted that the
number of passes with the desired geometry avoiding roll rolling processes are usually large scale, highly dimen-
defects whilst maximising the reduction in the cross section sional, non-linear, highly uncertain and involves complex
area per pass. This is normally achieved by considering the interaction of highly experienced engineers to apply
geometrical, mechanical, thermal, thermo-mechanical and empirical design knowledge to develop roll schedules [4].
metallurgical behaviours of the entire rolling system [2] (see Some of the main considerations of RSD problems are listed
Fig. 1). below, and are summarised in Fig. 2.
The rolling system as a whole is multi-disciplinary in
nature consisting of reheating, inter-stand operations, mill 2.1.1. Search space
engineering, roll pass design, straightening etc. Fig. 1 RSD can be formulated as a process optimisation
shows a fishbone diagram that classifies the rolling system problem, where the search space represents all possible
components for this paper. The fishbone diagram was designs. Here, processing parameters can be chosen to
developed by the authors in collaboration with rolling maximise productivity, minimise product defect subject to
system engineers and by reviewing the literature [1]. It is the complex synergies between the material flow, compo-
also important to note that in rolling system analysis it is sition, thermal dependencies, and intricate interactions
also common to include the influence of historical between the draughting section rolling. Like most real-
behaviours from steelmaking and casting processes. world process optimisation problems, RSD poses significant
Fig 1. Fishbone diagram of the rolling system.

914 V. Oduguwa, R. Roy / International Journal of Machine Tools & Manufacture 46 (2006) 912–928
Fig 2. Features of rolling system design problem.
challenges for traditional optimisation algorithms. The main 2.1.3. Hierarchical

features of the problem space is summarised in this section The optimisation of a rolling system is a hierarchical
using the classification given by [5]. problem in which different competing technological and
The complexity of the search space is due to the number quality objectives have to be solved at each level of
and nature of objectives and constraints. The presence of optimisation [11]. For instance, at the top levels,
multiple hard constraints significantly affects the perform- productivity of the rolling mill is one of many issues to be
ance of any optimization algorithm including evolutionary considered, while at lower levels, for example, the
search methods [6] making it difficult to find optimal determination of the number of passes and the form of
solutions. The objective and constraint functions can be grooves must be obtained [12–14]. A parameter or an
non-smooth, severely non-linear, discontinuous or even ill- objective function can also be of singular or multiple
defined/undefined in some region of the parameter space hierarchical level in nature. For example, a geometrical
[7]. The existence of several local extrema (multi-modality) parameter might be of lower level such as CAD entities, or
is also quite common. Most real-world process problems of a higher abstract level such as feature-based, or might be
also have very large number of parameters resulting in a switch between both. It is observed that roll designs switch
exponential increase in interactions [8]. This effect was levels in this way within the current unstructured design
described by [9] as the ‘Curse of Dimensionality’. The process.
search space can also be quantitative (QT) or qualitative
(QL) in nature or both [10]. QT describes the behaviour of 2.1.4. Multi-disciplinary
the problem using suitable numerical terms. However, this RSD is characterised by complex interactions occurring
can be incomplete and even there are certain behaviours amongst various disciplines. For example, material flow/-
(e.g. manufacturability), which are not, easily described composition, thermal, pass design, mill capabilities, inter-
using analytical means. In such cases human based stand dependencies, and process controls all combine to
reasoning can be formulated into rules for QL models to determine the shape and property of the product. Combining
express such behaviours. different models across various disciplines have an
important influence on the optimisation algorithm. It is
The following paragraph explores some of the RSD
necessary to ensure models have equivalent scales so that
features that are likely to influence optimisation algorithms.
they can be integrated and the optimisation algorithm can be
This also demonstrates that the RSD is similar to other
utilised.
process optimisation problems.
2.1.5. Multi-dimensional
2.1.2. Multi-staged RSD is characterised by various process and product
By definition, multiple passes is an ordered multi-stage variables such as forming sequence, shape evolution,
process. This is similar to the sequential process described forming speed, reduction ratio, temperature distribution,
for process problems where the output stock of one mill friction conditions, thermal and mechanical conditions of
stand is fed as input stock into the subsequent mill stand. the work-piece and the die [15].
Solving the sequential design problem increases the
problem size (number of variables), introduces hierarchical 2.1.6. Qualitative knowledge in rolling system design
search space problems and relationship between passes. Rolling system design is a knowledge intensive task and
These relationships are expressed in terms of variables, can be qualitative (QL) in nature [16]. Traditional RSD has
parameters, objective and constraints. The nature of the been undertaken by individual experts using a mixture of
problem can be static or dynamic, QT, QL or semi-QT empirical knowledge, based on sound engineering and
depending on the level of knowledge available. These rolling practices acquired over many years of experience.
features present several challenges for an algorithmic This knowledge is not always explicit and is not so easy to
approach to the process optimisation. capture such knowledge for generic re-use. In many
instances, there are no clear rules to design a particular pass and it was considered to be a convenient form for
shape [17]. They tend to be too specific and it is difficult to calculation.
decouple the process, product and mill engineering knowl-
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
edge [18]. P Z wm rðh1 Kh2 Þ
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1:6m Rðh1 Kh2 Þ K1:2ðhKh2 Þ
! 1C
3. Rolling system design approaches h1 C h2
2 qffiffiffiffiffiffiffiffiffi 3
This section presents a comprehensive review of 2n3 h1K R
h2
approaches for dealing with RSD problems in terms of ! 4J C 5 (3.1)

h1 C h2
modelling and optimisation. This section presents a critical
analysis of the RSD optimisation approaches in terms of
developing approximate QT models and approximate QL where PZrolling load, wmZmean width of the material
models, manual search methods, handling QL information, in the arc of contact, h1Zheight of material before the
multi-pass optimisation problems, and Soft Computing pass, hZheight of material after the pass, RZroll radius,
approaches in RSD. nZperipheral velocity of the rolls at the bottom of the
pass, mZcoefficient of friction between the materials and
3.1. Approximate QT modelling methods roll. Ekelund derived values of mZ0.848–0.000222tF
from experiments for steel rolls, tFZtemperature of
Section 2.1 stated that a pre-requisite to any rolled material. 3Zcoefficient of viscosity derived as
optimisation problem is the availability of a process 0.01 (20,164–7.89tF), J is the static compressive strength
model. This is required to evaluate the quality of a given of hot steel derived as 1003(1.4CCCMnC0.3Cr).
solution. Since physical experimentation is slow, expens- Where C, Mn and Cr are the carbon, manganese and
ive and disrupts production runs, approximate QT models chromium content of steel, respectively. Limitations in
are developed to represent the process behaviour so that Ekelund’s formula are due to large approximations.
off-line studies can be carried out in order to gain better According to Ekelund’s formula, if the temperature is
understanding of the process. Several modelling methods increased to 1400 8C, the yield stress would be equal to
have been developed for simulating the behaviour of the zero. This implies that the rolled stock would begin to
rolling process. These approaches aim to find a suitable melt. A critical analysis of Ekelund’s formula is
representation for the underlying process behaviour. This presented by [1].
section reviews the following QT based modelling Siebel and Lueg [22] performed experimental studies
approaches, classical rolling theories, recent mathematical on annealed aluminium and measured roll pressure
methods, finite element methods and approximate finite distribution. Their results were used to compare early
element strategies. theories. Orowan [23] discarded most of the mathematical
approximations and assumptions of the rolling theories
and developed the most comprehensive model. Orowan
3.2. Classical rolling theories on load and torque recognised that the flow stress of the material varies in
rolling due to the variation of effective strain, effective
Much efforts were devoted to develop theoretical strain rate and temperature and considered these
model to predict roll pressure distribution, rolling load, variations in his model. Orowan also abandoned the
torque, yield stress of the material being rolled and the assumptions of slipping friction and homogeneous
coefficient of friction between the rolls and the material. compression. He showed that the coefficient of friction
Most rolling mathematical models of hot-strip rolling are can be sufficiently high for the tangential stress to reach
concerned with the force of equilibrium of vertical cross- the shear yield stress of the material and that the strip
sections in the roll gap. Siebel [19] and Von Karman [20] sticks to the roll. He was interested in developing an
in 1925 first introduced the concept of homogeneous accurate theory that can be used as a benchmark for
compression of vertical strip segments as they move assessing simplified theories regardless of its complexity.
through the roll gap. These early authors made the Subsequently, several authors simplified the Orowan’s
fundamental assumptions of a neutral plane cross-section model, notably [24]. Sims assumed sticking friction over
and a neutral point occurs along the contact arc between the whole arc of contact and made allowances for
the rolls and the deforming metal. Ekelund [21] later homogeneous deformation. Sims’s formula shown in Eq.
developed a formula (Eq. (2.1)) based on the study of (3.2) consistently under-predicts values predicted by the
material flow during plastic deformation. Ekelund’s Ekelund formula [21,25]
approach consisted of contact area, friction hill factor
and the yield stress. It was very popular in the early years pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
due to the embodiment of the essential rolling variables P ZK R 0 ðh1 Kh2 ÞQp (3.2)
" rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffirffiffiffiffiffiffiffiffiffiffi rffiffiffiffiffiffiffiffiffiffisffiffiffiffiffi for the roll process design problems. This section introduces
p 1Kr K1 r p 1Kr R 0 h
Qp Z tan K K ln N the FE method, and reviews FE applications to rolling
2 r 1Kr 4 r h2 h2 problems.
rffiffiffiffiffiffiffiffiffiffisffiffiffiffiffi #
The aim of FE application to metal forming problems is
1 1Kr R 0 1
C ln ð3:3Þ to predict physical quantities such as stress, strain, velocity,
2 r h2 1Kr
and temperature throughout the deformed solid. This
where R 0 is radius of curvature of the elastically implies that the physical quantities are required at infinite
deformed roll, QP is a complex function of the roll gap number of points, which defines a problem with infinitely
given by Eq. (3.3), and rZ ðh1 Kh2 Þ=h1 , h1 and h2 is the many degrees of freedom. The FE method is based on the
initial and final thickness of rolled strip, respectively. idea of discretisation where the deformation zone is divided
into a finite number of sub-zones called elements. The
elements are connected together at the corners and at
3.2.1. Classical rolling theories: observations selected points at the edges called nodes. For each element,
Although applications of classical rolling theories have the individual relationship between the applied nodal forces
been relatively successful in delivering roll design and the resulting nodal variables such as velocity and
solutions, their limitations influence the RSD in terms of temperature are calculated and the element stiffness
the quality of solutions, and poor response to a broad range obtained. The global stiffness matrix of the whole body is
of problems. They tend to be related to specific disciplines. then assembled by combining the elemental stiffness. A
Typical classical rolling theories tend to consider spread, detail description of the FE method is given in [4].
elongation, and load. This limits the information obtainable
from such methods since rolling processes are influenced by 3.3.1. Application of FE method in rolling problems
many more factors such as metallurgical and thermo- The fundamentals of the rigid-viscoplastic FE method
mechanical properties. Since limited number of rolling and the mathematical formulation are well established and
design variables are considered, this can result in unrealistic given in literature [31,32]. Several authors [33–37] have
design solutions. applied the FE method to determine more detailed
Several authors [26–28] have investigated the perform- information of material flow for shape rolling in the analysis
ance of a number of classical rolling theories and found that process. In 1984, Park and Kobayashi [38] expanded the
in most cases the rolling load and torque are severely rigid-viscoplastic FE method to three-dimensional cases.
underestimated. Alexander [29] conducted experimental Hacquin et al. [39] presented a coupled model of thermo-
studies on annealed copper strip with front tension and made elastoviscoplastic strip deformation and thermo-elastic roll
comparison on earlier theories. It was concluded that deformation. The model was used to predict profile defects,
previous rolling theories were not capable of predicting strain, and stress maps, including residual stresses in hot and
the roll torque with adequate precision. The fundamental cold rolled strips. In rod rolling 3D finite element was used
stress equilibrium approach for which the rolling theories to determine strain distribution for prediction of micro-
were based fails to take account of the non-homogeneity of structural material evolution [40–42]. Several authors [43–
the deformation. For this reason, under-estimates of the 47] have used FE to determine the thickness accuracy of the
rolling loads resulted in many cases. Most of the empirical strip products. These studies calculated the required rolling
rules tend to focus on flow mechanism related to spread and force, torque, and power in terms of other process
elongation for various roll groove geometries without parameters such as roll speed, thickness, and temperature.
considering the temperature and strain rate effects. Research Recently, Kim and Im [48] applied a three-dimensional FE
efforts have focused on applying the Finite Element program for analyzing shape rolling processes considering
Analysis (FEA) to the metal forming processes to address heat transfer based on rigid thermo-viscoplastic approach.
some of the above mentioned issues. Round–oval, oval–oval, square–oval, and square–diamond
passes (Fig. 3) were simulated at different friction
3.3. Finite element methods in RSD conditions with and without temperature effects.
In spite of the computational burden, FE models are
The concept of Finite Elements (FE) was originally becoming powerful and are becoming popular for investi-
introduced in the 1950’s as a method of structural analysis gating technological windows of different processes in the
for the aircraft industry [30]. Since then, it has evolved and steel industry. Three-dimensional (3D) FE based codes
its wide application covers various classes of metal forming enable the modelling of large plastic deformation. The FE
problems such as rolling. The FE technique can be used to modelling technique is reported to be more accurate and it
simulate the metal forming before performing actual provides detailed information of plastic deformation when
experiments, which can reduce the cost and development compared to classical rolling theories. It provides an
time for the process design. However, due to the extensive description of the metal flow in the roll gap and
computational burden of the FE method there are still few of loads and motion of the whole strip and is thus widely
challenges for using the FE method as embedded optimisers used for simulating and analysing various metal forming
Fig 3. Typical breakdown sequence.
processes [49–52]. The FE technique has a considerable along with the current against some pre-defined criteria.
advantage to the traditional mathematical approaches Although this approach intuitively suggests significant
because it can calculate most of the parameters characteris- reduction of computational cost, storing previous solutions
ing the process (forces, stresses, velocities, displacements, could significantly disrupt the pattern formation during the
temperatures etc.) in a single simulation. search.
3.3.2. Approximation strategies for FE methods 3.3.3. FE method in RSD: observations

Wen et al. [53] presented a simplified FE for modelling FE methods are capable of predicting quite accurate
complex three-dimensional deformation in section rolling. deformation behaviour such as roll contact stresses, rolling
In their work, previous analysis of various passes with the load, and rolling torque required for RSD. However, they
three-dimensional models showed CPU times between 26 can be computationally expensive and the time required for
and 52 h compared to pseudo-two-dimensional models of 2 a fully converged solution limits its wider application for
and 3 h. Kiuchi and Yanagimoto [54] developed a complex optimisation problems. CPU time of 4–5 h for 5000–6000
element method based on the functional combination of nodes and 1000 time steps is not unusual [60]. The three-
rigid-plastic FE and the slab method. Kim et al. [36] further dimensional modelling even with the super computer is still
simplified the approach and applied the two-dimensional time consuming and quite complex to investigate the
rigid-plastic FE method for the generalised plane strain influence of all the potential variables [61,62,28]. Although,
condition, and combined it with the slab method. Kim et al. the FE method provides various detailed information, it can
[55] extended the work by developing a computer program be difficult to analyse. The FE based design process often
called TASK which was used by Shivpuri and Shin [56] to lacks structure and appears to be done sporadically. Since,
investigate RSD. In truss optimisation problems, where various design parameters are considered simultaneously,
similar FE computational burden is also experienced, quite often practitioners seems to be content with any
several approaches were reported to address the problem. solution that appears feasible in light of the computational
Deb and Gulati [57] in design of truss-structures, introduced burden. In addition, since this approach is mainly based on
the concept of basic and non-basic node to emphasise QT information, it cannot address integrated QT and QL
creation of user-satisfactory trusses and reduce compu- search space problems [4].
tational time. This avoided expensive FEA for unsatisfac-
tory trusses. Quagliarella and Vicini [58] proposed a 3.4. Recent mathematical modelling methods
hierarchical approach for the fitness evaluation. This
involves using several solvers with different levels of In spite of reported model accuracy of the FE methods in
accuracy, in order to use the more computationally metal forming problems, there is still an increasing interest
expensive models only when needed. These approaches to use simplified mathematical formulae for predicting
can be regarded as ‘good house keeping measures’ that complex metal forming behaviours [63,64]. There are
improves on the computational expense of large FE runs; various reasons why the mathematical approach is still
however, they fall short of alleviating the problem in the popular in the industry. It is a simplified alternative to the
context that makes them applicable to complex real life complex FE formulations and it requires less computational
problems. Siteo [59] developed a Genetic Algorithm with efforts. They allow formulation of dependencies between
Ancestry Record (GALGAR) within a framework of hybrid rolling parameters [65]. Rudkins and Evans [25] and Lenard
techniques based on simple GA, hill-climbing and simulated et al. [28] performed off-line calculations to aid in the
annealing. The basic idea is to save computational FE finishing of mill set-ups. In their study, they use the [21] and
simulation cost by storing previously processed information [24] formulae to determine the likely rolling loads for
a given reduction pattern and compared the results obtained This also incurs expensive computational cost since the
with Finite Element calculations. They concluded that the models are generated from FE data.
FE results gives accurate prediction of mill loads while the
Sims model [24] slightly under-predicts the values 3.6. Manual design search methods in RSD
compared to that predicted using the [21] formula. Joun
and Hwang [63] developed an approximate mathematical A significant part of RSD carried out in the industry is
approach for predicting roll pressure and tangential stress at manual. This relies on adopting existing solution to produce
the roll–strip interface. The authors reported that theoretical the required design where design variables are changed one
predictions agree with FE results and experimental results. at a time. Individual experts carried out traditional RSD by
Although there was no optimisation study reported in the combining empirical knowledge developed through years of
paper, the authors suggested that the method could be used experience with rolling theory. Prior to the 1980s, RSD had
for optimising rolling parameters. Wusatowski [1] and been largely based on empirical formulae. Several other
Chitkara and Hardy [66] applied the groove rolling principle authors have developed design rules empirically from
and proposed an empirical formula for determining spread production data [72,73] and experiments [74–77]. Based
in flat rolling. Mauk [67] adopted a similar method and on such research work, design guidelines and the empirical
proposed an iterative approach. Hwang et al. [41] presented formulations were developed to speed up the development
an analytical model for predicting the mean effective strain process and to enable a more structured RSD approach.
defined as a maximum average effective (equivalent) plastic Although their applications have been relatively successful
strain at a given pass in bar (or rod) rolling process. The in delivering roll design solutions, their difference in
information obtainable by the mathematical modelling functional characters render them unsuitable for dealing
method is limited compared to the FE method since it is with complex geometries. In RSD design, where the
difficult to accommodate most of the complex factors. Like problem involves many variables the design problem
the FE method, it is based on QT information only, and as a becomes too complex due to curse of dimensionality.
result cannot simultaneously deal with integrated QT and QL Since the human can only deal with up to 5–10 variables at
search space problems. any single time [78], the search problem reduces to costly
trial and error and it becomes slow and very ineffective.
3.5. Approximate QL modelling methods The manual approach when used in the industry takes on
various forms with respect to the type of modelling
QL information is a non-numeric method used in approach adopted. The classical rolling theories as well as
reasoning about the behaviour of the physical environment. the recent mathematical models when used alone are
It is mostly used in the presence of incomplete knowledge. inherently trial and error search process from an initial
QL models include QL abstractions of the relations in a design (single point). Each process normally requires
numerical model, the casual dependencies between its significant effort and it results in long process development
variables and explicit representations of the modelling times and high cost of the products [79–81]. It is also
assumptions used to describe a physical system. This section observed that attempting to use this traditional approach
reviews QL modelling methods reported in literature to deal may result in optimising one criteria at a time [82]. Another
with RSD problems. important criterion for the RSD is to be able to search a large
Pataro and Helman [68] developed an approach to design space as quickly as possible and to deliver multiple
determine sequence of passes for the strip rolling process good solutions, which can be beneficial for locating a
using fuzzy logic rules. The authors suggested that the rules variety of good solutions. Here, the classical methods fail
could be extracted from data obtained from theoretical miserably. The manual RSD approach cannot produce
model, available databases or production data. Jung et al. multiple good solutions due to its inherent one point
[69] developed a fuzzy control system based on production solution method. In addition, since this approach is mainly
data and operational knowledge. The fuzzy controller based on QT information it cannot simultaneously deal with
calculates the change of roll force to improve the flatness QL aspects in RSD. Although the FE method is a
of the thin strip based on the developed fuzzy rules and computational technique, a design search conducted using
fuzzy inference. These approaches are only suitable for the FE method on its own is a manual process. Designers use
readily available data [70], and would incur significant their experience to search by adopting existing design
computational cost if used for large simulations. within the neighbourhood of a known solution point and use
Shivpuri and Kini [71] in roll pass design, generated a FE simply to evaluate the chosen design point [18]. The
fuzzy model from finite element data using two fuzzified computational cost of FEA and the nature of the search
design variables. In the approach adopted, design variables method inhibit elaborate exploration of the design space.
were fuzzified and used directly as inputs in the fuzzy This can result in low quality solutions, as the search is
reasoning module. Since the rolling process is highly limited. The FE method when used alone is not suitable for
dimensional, increasing the number of design variables generating designs at the initial design stage, where a large
also increase the number of rules required exponentially. number of alternative design solutions are required at low
cost and high response time. The FE model based simulations. Osakada et al. [87] later investigated an
optimisation also gets more complex and time consuming automatic generation of pass sequences using pattern
in the presence of large number of design parameters [83]. recognition. The interactive operations use rules to
This section has shown that the manual methods are determine the shapes and dimensions of the intermediate
single point search methods and are not suitable for multiple products. The authors experienced difficulties in expressing
objective problems and cannot evolve multiple good empirical knowledge due to its implicit nature and
solutions. The next section considers algorithmic concluded that simple knowledge engineering may not be
approaches. adequate for metal forming.
In a parametric optimisation based approach, Perotti and
3.7. Classical algorithmic optimisation approaches Kapaj [88] proposed a CAD based technique for automatic
shape design of square passes. This technique adopts a
Classical algorithmic optimisation approaches are point-by-point direct search method. The search method
alternatives to the manual methods. The aim is to change finds parameter values of stock shape by iterative
values to all the design variables simultaneously to achieve calculations using empirical formulae. The method is
a set of objectives while satisfying a set of constraints. slow, requires many function evaluations for convergence
Although, it is faster and more effective in exploring a and tends to get stuck to sub-optimal solutions [89]. Since
design space than the manual methods, this approach has the shape forming is a highly knowledge intensive task,
received very little research interest for RSD problems. Osakada et al. [90] extended their previous work by building
Yamada et al. [84] developed an algorithm to optimize load expert system from FE simulations results. The authors
distribution using mathematical models for set-up calcu- integrated expert system, FE method, and neural networks
lation. Lapovok and Thompson [12] formulated a math- for knowledge acquisition. In similar approaches, several
ematical geometrical problem for tool-form optimisation for authors [91–94] integrated an expert system with a CAD
roll pass design. The authors used a derivative-based software for a forming sequence. A number of expert
optimisation method to design a breakdown sequence systems have also been developed to produce more efficient
system for rolling simple square profiles. This approach solutions in other areas, e.g. steel composition design [95].
requires the problem to be twice differentiable and the While the expert system determined the feasible pre-form
knowledge of the search space. However, these require- sequence for the multi-stage forging process, the CAD
ments are not always satisfiable due to the complexity of the provides the designer with the necessary tools to perform
RSD problems. complete analyses for suitability. Predictions from the
expert system are further evaluated with FE simulations.
3.8. QL and QT optimisation strategies in metal forming Although the developed systems were more accurate, the
search method was still slow due to the expensive FE
Several integrated computational systems have been simulations and the inherent trial and error of QL evaluation.
reported in literature combining Computer-Aided Design Several CAD commercial applications such as LINE-
(CAD), Expert Systems (ES), and the FE method in various BOW [96] and CARD [97] were developed to help improve
configurations. The integrated approach can save expensive the design efficiency and increase productivity. These
design time by re-using knowledge. In the 1970s, integrated applications are composed of several modules of math-
computer-aided systems were developed to incorporate ematical pass design models designed to reduce the
mathematical design techniques and experience based repetitive task (such as calculations and drawings)
intuitive skills for metal forming process design. The performed by the designers. Since the designers evaluate
Expert System (ES) automated most of the routine and the overall design solution, the evaluation process of the
repetitive tasks such as die drawings and design calculations system is still slow and unstructured due to the inherent trial
performed by designers. The benefit of the approach is that and error nature of the designer interaction. Recent
it reduced the drafting time as well as some of the trial and integrated strategies attempt to focus on automatic
error in the process design. Subramanian and Altan [85] generation of process design solutions. These strategies
integrated a CAD and ES system and named it ‘DIE aim to provide a more structured design approach free of the
FORGE’ for die design. The designer interacted with the ‘on-line’ designer’s QL evaluation. Im and Kim [98]
system through QL evaluation of designs obtained from the developed a Knowledge Based Expert System (KBES)
ES. Although the system reduced repetitive tasks, it is still using Object Oriented Programming (OOP) for the design
slow and less accurate. In an attempt to improve the of roll pass, profile sequence and shape rolling of round and
accuracy of such systems, Akerman et al. [86] integrated FE square bars. The authors built KBES using empirical design
method with a more interactive CAD system. The authors rules adopted from basic design theory and practice.
developed two computer programs for modelling shape Farrugia and Jennings [18] developed the current state of
rolling for airfoil designs. The first program predicts lateral the art integrated KBES/CAD system. The authors
spread and the second program simulates rolling of simple combined KBES, CAD, and FE analysis for automatic
parts. The system relied on expensive trial and error FE generation of roll design solutions. The KBES records,
maintains, and automates designs, the CAD generates This aims to achieve the desired product quality without
drawings and the interface software and the FE simulator violating the operating constraints of the overall process.
provide a broad range of advanced analysis capabilities. Approaches for addressing this problem includes the FE
Although the system can deal with a broad range of method, backward tracing [49], forward tracing [101], and a
problems, the search for good solutions is still based on a derivative based approach [102,103].
point-by-point method. This can present difficulties for Despite the high computational expense, the FE method
problems with large number of variables and evaluation can is still very widely used for solving metal forming multi-
be time consuming since the solution is analysed by the FE pass optimisation problems [33–36]. There are various FE
solver. It was also argued that KBES is not suitable for RSD based approaches reported in the literature for dealing with
design optimisation because an infinite number of rules are process optimisation for metal forming. Two of such
required [13]. approaches are backward tracing and the derivative based
This section has shown that there are attempts to combine approach. The backward tracing starts with the final phase
QT and QL information in metal forming process design of the deformation stage and traces the loading path
problem. These approaches offer interesting combination of (deformation process in non-steady forming) backward to
knowledge based systems, CAD, and FE analysis. The predict optimal pre-form design associated with the finished
KBES approach is capable of producing simulation results part. Park et al. [49] and Kobayashi [104] applied a
quickly and is suitable for metal forming applications where backward tracing technique to design a pre-form in a shell
the knowledge is well known. The CAD/KBS reduces housing. The thermo-viscoplastic deformation was mod-
repetitive detailing work performed by the designers and elled as a boundary value problem after the FE discretisation
saves valuable design time that can used to generate more and the solution satisfies the system of non-linear coupled
creative design output. It can also incorporate the QL algebraic equations. The backward tracing technique has
information from the engineers to guide the search for good also been applied to plane-strain rolling problems [105] and
realistic solutions. disk forging where a uniformly deformed disk is sought
under the influence of friction at the interface of the die and
3.8.1. QT and QL of integrated strategies in metal forming: work-piece [106]. Although, the technique has been shown
observations to discover the desired final shape in various forming
The following issues pose several difficulties for the problems, the backward tracing technique, when used alone,
hybrid approaches mentioned above: cannot uniquely determine the optimal solutions due to the
presence of diverse and multiple loading solution paths
† Development of KBES is a complex task because the [107]. The back tracing technique tends to be more efficient
metal forming knowledge is often ill-structured, implicit, when the loading path is known, however, this can lead to
difficult to systematise and involves a large number of difficulties in real world problems where the search space is
rules [90,99]. complex and unknown, especially when multiple diverse
† The KBES are not suitable in areas where the knowledge loading paths are present.
is not well known for example in hot strip rolling areas Several authors have used the derivative based approach
such as roll damage, roll wear and sheet shape profile to solve multi-pass optimisation problems. For the
[100]. derivative approach, the optimal process design problem
† The approach is not capable of dealing with multi- is initially formulated mathematically based on a penalty
objective optimisation problems and is not suitable for rigid-viscoplastic FE method. The solution approach
simultaneously dealing with integrated QT and QL search obtains derivatives of the design objective and design
space problems. The search method produces single constraints with respect to the design variables. Current
point solutions, which can be slow and tedious due to its design variables for a given design iteration in the process
inherent trial and error approach. optimisation are then evaluated, using the values of the
† These methods do not conduct an algorithmic search and derivatives obtained. Joun and Hwang [102,103] developed
as a result often offer satisfying solutions and not the a FE based process optimisation technique and applied it to
optimal. a die profile design extrusion problem. A non-linear
iterative optimisation algorithm was adopted to minimise
the forming energy. In the proposed algorithm, the authors
3.9. Multi-pass RSD optimisation approaches integrated design sensitivity analysis with finite element
analysis for the deformation analysis. The authors extended
The multi-pass RSD problem is synonymous to the the work for pass schedule optimal design in multi-pass
sequential process optimisation problem outlined in Section extrusion and drawing process [108]. Barinarayanan and
2. The multi-pass problem aims to provide optimal design Zabaras [109] and Forment and Chenot [110,111] applied
solutions for individual passes of the rolling process the method for optimisation of die and preform shapes in
including both design information such as the geometrical non-steady state forming. In their study, the initial shape of
size of a roll and the operating conditions for the mills [1]. the part as well as the shape of the preform tool was
optimised for a two-step forging operation where the shape offers the opportunity to deal with ambiguity and
of the second operation is known. Shapes were described uncertainty in real world problem and can result in more
using spline functions and optimal parameter values of the realistic solutions. The application of intelligent SC
splines were searched in order to produce a part with a techniques is increasing with successful applications in
prescribed geometric accuracy, optimal metallurgical many areas including but not limited to: engineering design
properties and minimal production cost at the end of the optimisation, manufacturing system, process control, medi-
forging sequence. The work was later extended to solve cal diagnosis, simulation and communication systems [115].
non-isothermal, non steady forming problem [112]. Some of the principal combinations of SC components are:
Shin et al. [37] conducted a simulation study on I-section fuzzy logicCgenetic algorithm, neural networkCgenetic
beam with four passes. The authors carried out a simplified algorithm, neural networkCchaos theory, neuro compu-
three-dimensional FE method and a physical experiment tingCfuzzy logic, fuzzy logicCprobabilistic reasoning, etc.
with plasticine. Since the technique is inherently a trial and [115]. This paper focuses on the combination of Fuzzy
error approach, the development of analysis tool for RSD Logic (FL)Cgenetic algorithm as a SC technology for the
permitting prediction of the effect of process variables with RSD problem.
speed, economy and a level of parameter control cannot be Although SC techniques are emerging as an alternative
attained by conventional experiments [14]. for solving real world engineering problems [114], however,
Process sequence design for new materials and complex its application to metal forming design is still not common
geometries is not easy to carry out due to the complexity of [116]. Oduguwa et al. [117,118] presented survey of EC
the process [98]. There are major difficulties involved with techniques to metal forming. There are two main research
systematic process design through the analytical approach trends emerging in literature on the application of SC
due to the non-deterministic nature of the process. For techniques for metal forming problems. The first aims to
multi-pass designs it is not economically trivial to determine automate the search using EC based techniques by adopting
the optimum process sequence without making any a process model, and the second attempts to deal with the
deviations from these various paths. complexity of the metal forming domain problem using
fuzzy based modelling methodology. The overall aim of
3.9.1. Multi-pass RSD optimisation approaches: these two broad approaches is to develop low-cost compu-
observations tational techniques to guide an algorithmic search for good
Following observations can be made from the study solutions in real-world process optimisation problems. Most
above: of the applications reported in literature use the FE solver as
embedded design evaluation tool for the RSD.
† The tracing techniques require knowledge of the search Roy et al. [119] implemented an adaptive Micro Genetic
space to perform efficiently, however, this is not always Algorithm for shape optimisation of process variables in
possible since search space can be very complex and multi-pass wire drawing processes. A modular based FE
unknown. package was used to evaluate the objective functions. In a
† These techniques are derivative based and require an similar approach, Hwang and Chung [120] proposed a
initial guess, which can influence the search. This modified micro genetic algorithm for the optimisation of
derivative based approach also tends to get stuck in design variables for die shape in extrusion. The authors
sub-optimal solutions and an algorithm that is efficient in reported that convergence was fast for the first 200 iterations
solving one optimisation problem may not be efficient in and was slow afterwards. In a recent application of the
solving a different optimisation problem [113]. approach for process optimisation in forging [107], the
† The techniques cannot also identify multiple optimal authors reported that for 1000 FE simulations the search
solutions in a single run for the multi-objective multi- required 200 h for the convergence.
pass rolling problem. The techniques do not also In an attempt to deal with such large computational cost,
consider the relationship between passes. several authors have adopted neural network based methods
to develop process models for fast and inexpensive
predictions. Myllykoski et al. [70] developed process
3.10. Soft computing techniques in metal forming models for optimisation of rolling processes. Nolle et al.
[121] proposed an approach using neural network and GA
Soft computing (SC) is a collection of methodologies for the optimisation of roll profiles in strip rolling. Such
including, as its main constituents, Evolutionary Compu- hybrid approaches require a large number of sample data to
tation (EC), Fuzzy Logic (FL), Neuro-computing (NC), and develop the neural network model. The approach is not
Probabilistic Computing (PC) [114]. SC aims to exploit suitable for RSD where the large amount of data required to
tolerance for imprecision, uncertainty and partial truth to build the model can be expensive to obtain or are not readily
achieve tractability, robustness and low cost solutions. It available.
differs from conventional techniques, in that it incorporates Several authors are now adopting fuzzy logic in RSD for
human knowledge into the solution methodology. This modelling QL information. Ref. [68] developed an approach
to determine sequence of passes for the strip rolling process † Fuzzy Logic based design evaluation tools need to model
using fuzzy logic rules. The authors developed fuzzy logic a smaller set of critical process parameters, otherwise the
rules from production data. This proposed method, although number of rules required for the system can increase
suitable for generating fuzzy based models with real exponentially with the increase in the number of
processing conditions, is not capable of identifying optimal parameters.
sequences, as it has no optimisation capability. Jung and Im
[122] developed a fuzzy algorithm to calculate the roll speed
variations in order to improve the thickness uniformity of
hot-rolled strips considering the complex relationships of 4. Challenges: in practice and in research
process parameters such as roll speed, reduction ratio, strip
entry thickness, front and back strip tensions and the The RSD plays a significant part in delivering timely
deformation of the machinery in the mill system. This work optimal products to the markets. This paper presents a
is an interesting example of using SC techniques to deal critical analysis of the RSD optimisation approaches. The
with complex behaviour, especially if no mathematical review explored how the reported techniques search for
model exists to express such behaviour. Since the proposed good RSD solutions in terms of efficiency and cost of the
approach lacks optimisation capability, it is also not capable search approaches. The review also considered the capa-
of identifying optimal roll speed variations. bility of handling multi-objective optimisation problems
Shivpuri and Kini [71] developed an optimisation and integrated QL and QT search spaces. This section now
technique using an hybrid of empirical knowledge, FE explores the challenges poised by RSD as an optimisation
simulations and fuzzy analysis to improve product quality of problem and discusses how the reviewed approaches
RSD. In their approach, simulation points based on influence new product development in the steel industry.
experimental design were selected in the neighbourhood This discussion section concludes with an outline of
of the optimal design region and the optimal solution is then interesting research areas identified as a result of this study.
identified from the resulting fuzzy based evaluations.
Although this approach offers a semi-automatic method of 4.1. Challenges posed by optimisation of RSD
generating optimal solutions, it has the following draw-
backs. In general, the exploratory potential of the hybrid There are several challenges posed by optimisation
approach is weak due to the absence of an algorithmic search procedures for RSD problems. These challenges
search technique. The experimental design used for often inhibit the wider application of these procedures for
selecting simulation points is a random sampling plan, and rolling system design. This section identifies the challenges
may miss the optimal design points. Recently, Oduguwa and from both an industrial application viewpoint and an
Roy [123] presented an integrated design optimisation algorithmic perspective.
approach for QT and QL search space and applied it to the
RSD. The proposed solution approach is based on design of 4.2. Challenges of algorithm based optimisation in industry
experiment methods and fuzzy logic principles for building
the required QL models, and evolutionary multi-objective In spite of the potential benefits achievable from using
optimisation technique for solving the design problem. The algorithm based optimisation approaches for generating high
proposed technique was applied to a two objectives rod quality design solutions, it is very surprising that the appli-
rolling problem [124,125] and four stage multi-pass RSD cation of such approaches is still not grasping the headlines
problem [4]. The authors demonstrated that promising for industrial applications. This section presents industrial
results can be obtained from the proposed approach for perspective based on the authors’ observations from a survey
dealing with RSD taking into account the related QL conducted in steel industry on the modelling and design
evaluation of the design problem. activities [4]. These observations have been obtained through
a cross-disciplinary data collection approach adopted in this
3.10.1. Soft computing techniques in metal forming: study. This has enriched the knowledge used for generating
observations the observations outlined below.
† Soft Computing techniques can help in addressing a † Designers lack a full working knowledge of algorithmic
number of limitations of classical approaches, but still optimisation approaches. They often perceive it as a
have some limitations. highly theoretical approach, which bears little or no
† EC based techniques using the FE solver as embedded relevance to the day-to-day design activity. This issue is
design evaluation tool incurs severe computational cost. reflected by the comment made by a roll designer during a
This problem intensifies since EC techniques require to semi-structured interview including the author, a roll
evaluate large number of solutions before convergence. modeller, and the roll designer. These engineers were
This inhibits the use of the EC based optimisation asked about their viewpoint regarding the potential
approach to wider range of metal forming problems. benefits that optimisation approach could add to roll
design. One of the engineers commented that optimis- in a discrete event simulation workshop. When asked
ation is a different kind of logic which is only good for why optimisation is not part of the design activity for
academics’ purposes but not suitable for roll design rolling system design, the engineer replied ‘what is there
problems. to optimise?’, implying process models are not available
† Formulating the optimisation problem is difficult for the due to the complexity involved in building them.
average engineer who is not familiar with the optimis- † Realistic process models infer models not only based on
ation techniques. The features of the optimisation QT formulations but also models based on perception
problems (multiple objectives, constraints etc.) and the based reasoning (QL) of the engineers [126]. This shifts
availability of countless number of possible classical the complexity of the search space to a new paradigm
optimisation algorithms contribute to this difficulty. and an optimisation algorithm needs to be developed to
Selecting suitable objectives and formulating the con- deal with such complex models.
straints to reflect the real scenario is non-trivial. Classical † An important consideration for obtaining a realistic
optimisation algorithms tend to require prior knowledge solution is to ensure the solution strategy closely mimics
of the search space and this is not always possible for real the nature of the problem. This can ensure that the search
life problems due to its inherent complexities. preserves the structure of the problem and provides
† There is a cultural misconception inherent in most results that can be traced back to its solution mechanism.
engineers regarding the role of optimisation in product This also provides more transparency for the engineers
and process design. It is commonly perceived that since since they are able to understand the steps taken to arrive
optimisation can automatically generate design solutions, at the result. Developing an algorithm with such features
the approach can eventually replace their jobs. This fear can increase the complexity of the algorithm and
of losing their jobs places severe restrictions in the buy-in compromise its performance.
required from these user groups. This point was reflected
in an engineer’s comment during an informal interview-
ing session ‘I would rather not use this technique, because 4.4. Evaluation of RSD optimisation approaches for new
you know where this is going, I am going to lose my job’. product development
Also when asked about sharing the optimisation knowl-
edge with other areas of the business, the engineer This section presents an evaluation of the RSD
commented ‘we would rather keep the optimisation techniques reviewed in the previous sections in terms of
knowledge within the department as this ensures people their suitability for various stages of the New Product
always come to us when they have problems.’ Development (NPD) process. This was considered necess-
ary in order to explore the relationship between the
techniques and the different stages of the development
4.3. Challenges posed by RSD on optimisation algorithms process. This can be useful in identifying the most suitable
technique for a given stage of the product development
This section identifies some challenges of RSD optim- process.
isation poses on the optimisation algorithms. The review in the previous sections shows that the
incremental development of approaches since 1950 has
† The features of the real-life process optimisation addressed various aspects of the search problem with
problem, such as the presence of multiple objectives, varying successes. Fig. 4 shows a summary of the reviewed
constraints, interaction among decision variables, QT approaches and subjective ratings in terms of robustness and
and QL nature of the information and sequential nature of exploratory potential for a given technique with respect to
the design problem create challenges for optimisation the NPD process. A typical product development process
algorithms that are currently available for the industry. consists of the following stages: feasibility studies—
This discourages the industry to adopt these algorithms. concept design—design development and detail design
† A process model that explains the behaviour of the [4]. For such paradigm, the level of design exploration
design problem is a pre-requisite of any optimisation required at each stage of the process increases from right to
problem. It is needed by the optimisation algorithm to left, likewise the level of detail and accuracy required
evaluate the goodness of the solution. High compu- increases from left to right. For example, at the feasibility
tational expense (for example, FEA) inhibits its stage, a significant amount of exploration with relatively
application, in addition poor quality models result in small accuracy is required to identify any solution
engineers having little faith in them. This also makes alternatives. While for the detail design stage, very little
them sceptical about the results obtained from exploration is required since the design is already chosen.
the optimisation algorithms. Therefore, developing Here, the emphasis is on high accuracy and detail to ensure
RSD process models of reasonable computational cost the required objective is specified.
and acceptable quality is an important but quiet difficult Table 1 shows the technical capabilities of the
task. This point was well captured by a rolling engineer optimisation approaches in terms of multi-objective
Fig 4. Optimisation approaches and NPD.

T L T L
optimisation problems, Q and Q , Q , Q and exploratory † Integrated QT and QL information: studies are required
potential. The exploratory potential signifies the capability to develop optimisation algorithms that can deal with
to search a design space with reasonable effort. The rating various combinations of QT and QL information, such as
given here is subjective from the authors’ viewpoint and is complementary and non-cooperative forms, in a single
meant to illustrate the relative strength of exploration for the framework. Scalability of integrated QT and QL design
approaches surveyed. Approaches with one star ‘*’ confirms optimisation strategies to higher dimensional problems
that they are based on manual search. It is clear from this is an important success criterion for wider applications.
evaluation that SC techniques have the highest exploratory This is influenced by the feature of the problem (large
potential and it is capable of handling multi-objective number of parameters) and the nature of the resulting
problems. However, most of the reported approaches are search space (discontinuous).
computationally expensive as shown by the location of item † Techniques are required for representing the native
7 in Fig. 4. It is also observed that most of the FE based parameter space of the QL information within the
approaches (4–7) shown in Fig. 4 are ranked with high optimisation framework. This could provide capabilities
quality. FE based approaches are computationally expens- for tuning the correlation between the granularities of the
ive but can produce good quality results. The figure also QL models with the measurement scale of the QT
shows the need to develop suitable approaches for the models. Search algorithms that consider such features of
feasibility stage of the RSD. The following sections outline the problem should give better performance.
future research directions to deal with these challenges. † Within the multi-pass optimisation contribution from the
several stages may even out fluctuation caused by many
4.5. Future research directions of the underlying factors influencing the search
directions, this may not always be the case and could
Future research directions can be summarised as follows: result in deceptions in the search problem. Further study
could consider investigating the behaviour between
† Approximate strategies for QT multi-objective optimis- different rolling stages during the evolutionary search.
ation problems: optimisation approaches that require a Such information could be valuable in understanding the
smaller population for convergence whilst minimising convergence behaviour of the search space, which could
the error between the simulation and the approximate result in improved algorithms.
models should be further investigated. Emerging EC † An inherent feature of the process optimisation problem
based sampling techniques such Kriging, MARS can be is the hierarchical nature. There is a need to develop a
explored. hierarchical optimisation approaches that can recognise
differences at the subsystem level (each mill stand) but
Table 1 performs a system level optimisation (e.g. for the rolling
Technical capabilities of the RSD optimisation approaches operation as a whole).
Current Multi-objective QT and QL QT QL Exploratory
approaches potential
Manual ! ! # ! *
Rolling ! ! # ! * 5. Conclusions
theories
FE ! ! # ! * The key conclusions from the review can be summarised
KBS/ES ! # # # * as follows:
Classical ! ! # ! ***
optimisation
† Classical rolling theories have limited success in
SC # ! # # ****
delivering roll design solutions. Their limitations can
influence optimal RSD in terms of the quality of [5] R. Roy, A. Tiwari, O. Munaux, G. Jared, Real-life Engineering
solutions, and poor response to a broad range of Design Optimisation: Features and Techniques CDROM Proceed-
ings of the Fifth Online World Conference on Soft Computing in
problems. Industrial Applications (WSC-5), Finland, 2000.
† The FE method provides a variety of detailed and high [6] Z. Michalewicz, A Survey of Constraint Handling Techniques in
quality information, however, the FE based design Evolutionary Computation Methods Proceedings of the 4th Annual
process often lacks structure and appears to be done Conference on EP, MIT Press, Cambridge, MA, 1995. pp. 135–155.
sporadically. It is computationally expensive and often [7] M.S. Elred, Optimization strategies for complex engineering
applications, Sandia 1998.
tends to inhibit its application to a broader range of RSD [8] H.W. Ray, J. Szekely, Process Optimization with Applications in
problems. Metallurgy and Chemical Engineering, Wiley, New York, 1973.
† When generating a fuzzy model from FE data, increasing [9] R. Bellman, Dynamic Programming, Princeton University Press,
the number of design variables used in developing the Englewood Cliffs, NJ, 1957.
fuzzy model also increases the number of rules [10] G. Munda, P. Nijkamp, P. Rietveld, Qualitative multicriteria
methods for fuzzy evaluation problems: an illustration of econ-
exponentially. This incurs expensive computational
omic-ecological evaluation, European Journal of Operational
cost. Very little research has been reported to deal with Research 82 (1995) 79–97.
the QT and QL RSD problems. Most of the soft [11] V. Oduguwa, R. Roy, Bilevel Optimisation Using Genetic Algorithm
computing based efforts in RSD either focus on 2002 IEEE International Conference on Artificial Intelligence
evolutionary computing or fuzzy logic based techniques. Systems, (ICAIS 2002), Divnomorskoe, Russia, 2002 pp. 322–327.
[12] R.Y. Lapovok, P.F. Thompson, The Mathematical Basis of Optimal
† There has been little focus in developing optimisation
Roll Pass Design, Engineering Mathematics: The Role of Math-
approaches with high exploratory potential that generate ematics in Modern Engineering Biennial Conference, Melbourne,
a large number of alternative optimal RSD solutions or 1994 pp. 435–444.
have the capability to deliver the required quality of [13] W. Shin, Development of techniques for pass design and
solutions for the concept and feasibility design stage of optimisation in the rolling of shapes, PhD Dissertation, The Ohio
State University, 1995.
the new product development.
[14] R.Y. Lapovok, P.F. Thompson, An approach to the optimal design of
rolling passes, International Journal of Machine and Tool Manu-
This paper has presented a comprehensive review of facture 37 (8) (1997) 1143–1154.
approaches for rolling system design optimisation and [15] W.L. Roberts, Hot Rolling of Steel, Marcek Dekker, New York,
discussed their advantages and disadvantages. The review 1983.
has critically analysed how each approach contributes to [16] V. Oduguwa, R. Roy, Qualitative and Quantitative Knowledge in
Engineering System Design 17th National Conference on Manu-
developing timely low cost optimal solutions and how they
facturing Research, Professional Engineering Publishing, Cardiff
influence new product development strategies for the steel University, UK, 2001. pp. 81–86.
industry. The challenges, posed by the RSD problem and the [17] D. Jennings, Advances in Roll Pass Design, vol. 12, Saskatchewan
future research directions are also identified. Music Educators Association, 2000.
[18] D.C.J. Farrugia, D. Jennings, Advanced Analysis System for Roll
Pass Design 11th Abaqus User’s Group, Warrington, UK, 2000.
[19] E. Siebel, Kraft und materialflub bei der bildsamen formanderung,
Acknowledgements Stahl Eisen 45 (37) (1925) 1563.
[20] T. Von Karman, Bietrag zur theorie des walzvorganges, Z. angew
Math. Mech 5 (1925) 1563.
The authors are grateful for support from EPSRC and [21] S. Ekelund, The analysis of factors influencing rolling pressure and
Corus for the initial research. The authors also recognise power consumption in the hot rolling of steels, Steel 93 (8) (1933)
contributions from members of the Decision Engineering 27.
research team within the Enterprise Integration department [22] E. Siebel, W. Lueg, Mitteilungen aus dem Kaiser Wilhelm, Institut
Fur Eisenforschung, Dusseldorf, 1933.
at Cranfield University.
[23] E. Orowan, The calculation of roll pressure in hot and cold flat
rolling, Proceedings of the Institute of Mechanical Engineers 150
(1943) 140.
[24] R.B. Sims, The calculation of roll force and torque in hot rolling
References mills, Proceedings Institute of Mechanical Engineers 168 (1954)
191.
[1] Z. Wusatowski, Fundamentals of Rolling, Pergamon, NY, USA, [25] N. Rudkins, P. Evans, Mathematical modelling of set-up in hot strip
1969. rolling of high strength steels, Journal of Material Processing
[2] C.G. Sun, C.S. Yun, J.S. Chung, S.M. Hwang, Investigation of Technology 80–81 (1998) 320–324.
thermomechanical behavior of work roll and a roll life in hot strip [26] R. Shida, H. Awazuhara, Rolling load and torque in cold rolling,
rolling, Metallurgical and Materials Transactions A 29A (1998) Journal of Japan Society Technological Plasicity 14 (147) (1973)
2407.52–2424.52. 267.
[3] M. Kleiber, Z. Kulpa, Computer-assisted hybrid reasoning in [27] J.G. Lenard, Study of the Predictive Capabilities of Mathematical
simulation and analysis of physical systems, Computer Assisted Models of Flat Rolling Fourth International Steel Rolling Con-
Mechanics and Engineering Sciences 2 (1995) 165–186. ference, Deauville, France, 1987.
[4] V. Oduguwa, Rolling System Design Optimisation Using Soft [28] J.G. Lenard, A. Said, A.R. Ragab, M. Abo Elkhier, The temperature,
Computing Techniques, EngD Thesis, Cranfield University, Cran- roll force and roll torque during hot bar rolling, Journal of Material
field, UK, 2003. Processing Technology (1997) 147–153.
[29] J.M. Alexander, On the Theory of Rolling Proceedings Rolling [51] B.-S. Kang, Process sequence design in a heading process, Journal of
Society, London, 1972 pp. 535–555. Material Processing Technology 27 (1991) 213–226.
[30] M.J. Turner, R.W. Clough, H.C. Martin, L.J. Topp, Stiffness and [52] T. Altan, M. Knoerr, Application of the 2D finite element method to
deflection analysis of complex structures, Journal of Aeronautical simulation of cold-forging process, Journal of Material Processing
Science 23 (1956) 805–823. Technology 35 (1992) 275–302.
[31] O.C. Zienkiewicz, The Finite Element Method, McGraw-Hill, New [53] S.W. Wen, P. Hartley, I. Pillinger, C.E.N. Sturgess, Roll pass
York, 1977. evaluation for three-dimensional section rolling using a simplified
[32] S. Kobayashi, S.I. Oh, T. Altan, Metal Forming and the Finite finite element method, Proceedings of the Institute of Mechanical
Element Method, Oxford University Press, NY, USA, 1989. Engineers (1997) 143.
[33] K. Mori, General purpose fem simulator for 3-d rolling, Advanced [54] M. Kiuchi, J. Yanagimoto, Computer aided simulation of universal
Technology of Plasticity 4 (1990) 1773–1778. rolling process, ISIJ International 30 (2) (1990) 142–149.
[34] J.J. Park, S.I. Oh, Application of three dimensional finite element [55] N. Kim, S.M. Lee, W. Shin, R. Shivpuri, Simulation of square-to-
analysis to shape rolling processes, Transaction ASME Journal of oval single pass rolling using a computational effective finite and
Engineering Industry 112 (1990) 36–46. slab element method, Journal of Engineering for Industry 114 (1992)
[35] J. Yanagimoto, M. Kiuchi, Advanced computer aided simulation 329–344.
technique for three dimensional rolling processes, Advanced [56] R. Shivpuri, H.W. Shin, A methodology for roll pass optimisation for
Technology of Plasticity 2 (1990) 639–644. multi-pass shape rolling, International Journal of Machine and Tool
[36] N.S. Kim, S. Kobayashi, T. Altan, Three-dimensional analysis and Manufacture 32 (2) (1992) 671–683.
computer simulation of shape rolling by the finite and slab element [57] K. Deb, S. Gulati, Design of truss-structures for minimum weight
method, International Journal of Machine and Tool Manufacture 31 using genetic algorithms, Finite Elements in Analysis and Design 37
(1991) 553–563. (2001) 447–465.
[37] H.W. Shin, D.W. Kim, N.S. Kim, A study on the rolling of I-section [58] D. Quagliarella, A. Vicini, Viscous single and multicomponent
beams, International Journal of Machine and Tool Manufacture 34 airfoil design with genetic algorithms, Finite Elements in Analysis
(1994) 147–160. and Design 37 (2001) 365–380.
[38] J.J. Park, S. Kobayashi, Three-dimensional finite element analysis of [59] R.V. Siteo, The Use of Genetic Algorithms for Improving the
block compression, International Journal of Mechanical Sciences 26 Dynamic Behaviour of Moving Gantry-type Wood Routers, PhD,
(1984) 165–176. Cranfield University, Cranfield, 2000.
[39] A. Hacquin, P. Montmitonnet, J.-P. Guillerault, A steady state [60] P. Gratacos, P. Montmitonnet, C. Fromholz, J.L. Chernot, A Plane-
thermo-elastoviscoplastic finite element model of rolling with strain elastoplastic model for cold rolling of thin strip, International
coupled thermo-elastic roll deformation, Journal of Material Journal of Mechanical Sciences 34 (3) (1992) 195.
Processing Technology 60 (1996) 109–116. [61] B.-S. Kang, J.-H. Lee, J.-H. Lee, Process design in multi-stage cold
[40] J.A. Nemes, B. Chin, S. Yue, Influence of strain distribution on forging by the finite element method, Journal of Material Processing
microstructure evolution during rod-rolling, International Journal of Technology 58 (1996) 174–183.
Mechanical Sciences 41 (1999) 1111–1131. [62] S. Yoshida, S. Kihara, Some experiences of shape-forming:
[41] S.M. Hwang, H.J. Kim, Y. Lee, Analytic model for the prediction of experiments and numerical simulations, Engineering Computations
mean effective strain in rod rolling process, Journal of Material 13 (2/3/4) (1996) 98–110.
Processing Technology 114 (2001) 129–138. [63] M.S. Joun, S.M. Hwang, An approximate analysis of hot-strip
[42] S. Serajzadeh, K.A. Taheri, M. Nejati, J. Izadi, M. Fattahi, rolling: a new approach, International Journal of Mechanical
An investigation on strain homogeneity in hot strip rolling Sciences 34 (12) (1992) 985–998.
process, Journal of Material Processing Technology 128 [64] I.J. Freshwater, Simplified theories of flat rolling-I, the calculation of
(2002) 88–99. roll pressure, roll force, and roll torque, International Journal of
[43] G.J. Li, S. Kobayashi, Rigid-plastic finite element analysis of Mechanical Sciences 38 (6) (1996) 633–648.
plain strain rolling, Journal of Engineering for Industry 104 [65] J. Mischke, Equations of strip equilibrium during asymmetrical flat
(1982) 55. rolling, Journal of Material Processing Technology 61 (1996) 382–
[44] K. Mori, K. Osakada, T. Oda, Simulation of plane-strain rolling by 394.
the rigid-plastic finite element method, International Journal of [66] F.R. Chitkara, G.M. Hardy, International Journal of Mechanical
Mechanical Sciences 24 (1982) 519. Sciences 19 (1977) 575.
[45] C. Liu, P. Hartley, C.E.N. Sturgess, G.W. Rowe, Simulation of [67] P.J. Mauk, Computer-aided Roll Pass Design, PhD Thesis, Technical
the cold rolling of strip using a elastic-plastic finite element University of Aachen, Aachen, Germany 1983.
technique, International Journal of Mechanical Sciences 27 [68] C.D.M. Pataro, H. Helman, Direct Determination of Sequences of
(1985) 829. Passes for the Strip Rolling Process by Means of Fuzzy Logic Rules
[46] N. Kim, S. Kobayashi, Three-dimensional simulation of gap Proceedings of the Second International Conference on Intelligent
controlled plate rolling by the finite element method, International Processing and Manufacturing of Materials. IPMM ’99, 1999 pp.
Journal of Machine and Tool Manufacturing 30 (1990) 269. 549–554.
[47] S.M. Hwang, M.S. Joun, Analysis of hot-strip rolling by a penalty [69] J.Y. Jung, Y.T. Im, H. Lee-Kwang, Fuzzy-control simulation of
rigid-viscoplastic finite element method, International Journal of cross-sectional shape in six-high cold-rolling mills, Journal of
Mechanical Sciences 34 (1992) 971. Material Processing Technology 62 (1996) 61–69.
[48] S.-Y. Kim, Y.-T. Im, Three-dimensional finite element analysis of [70] P. Myllykoski, J. Larkiola, A.S. Korhonen, L. Cser, The role of
non-isothermal shape rolling, Journal of Material Processing neural networks in the optimisation of rolling processes, Journal of
Technology 127 (2002) 57–63. Material Processing Technology 80–81 (1998) 16–23.
[49] J.J. Park, N. Rebelo, S. Kobayashi, A new approach to preform [71] R. Shivpuri, S. Kini, Application of Fuzzy Reasoning Techniques for
design in metal forming with the finite element method, International Roll Pass Design Optimisation Proceedings of the 39th Mechanical
Journal of Machine and Tool Design 23 (1) (1983) 71–79. Working and Steel Processing Conference and Symposium on New
[50] B.S. Yang, K.H. Kim, Rigid-plastic finite element analysis of plane Metal Forming Processes, 1998 pp. 755–763.
strain ring rolling, International Journal of Mechanical Sciences 30 [72] K.J. Soo, J.S. Hoggart, P.W. Whitton, International Journal of
(8) (1988) 571–580. Mechanical Sciences 17 (1975) 435.
[73] F. Bursal, K. Sevenler, P.S. Raghupathi, Computer-aided analysis of [95] H. Yasuda, Y. Nakatsuka, A. Yamamoto, I. Takeuchi, T. Hashimoto,
metal flow in the rolling of rods and structural profiles, International An expert system for the material design of large-diameter steel pipe,
Journal of Machine and Tool Manufacture 28 (1988) 475–482. The Sumitomo Search 30 (1992) 3–10.
[74] Y. Saito, Y. Takahashi, K. Kato, Calculation of spread, elongation, [96] A. Izzo, New software for desk top computer roll pass design Steel
effective roll radius, roll force and torque when rolling in the square- Times 1993.
diamond, square-oval and round-oval passes, Journal of Iron Steel [97] P.J. Mauk, R. Kopp, Computer aided roll pass design, Der Kalibreur,
Institute of Japan 2 (1978) 66–75. Heft 37 (1982) 89–93.
[75] T. Shinokura, K. Takai, A new method for calculating spread in rod [98] Y.-T. Im, S.H. Kim, A knowledge-based expert system for
rolling, Journal of Application in Metalworking 2 (1982) 147–160. roll pass and profile design for shape rolling of round and square
[76] Y. Saito, Y. Takahashi, M. Moriga, K. Kato, A new method for bars, Journal of Material Processing Technology 89–90 (1999)
calculating deformation and force parameters in steel rod rolling and 145–151.
ita application to roll pass design, Journal of Japan Society [99] P.A. Manohar, S.S. Shivathaya, M. Ferry, Design of an expert system
Technological Plasicity 24 (1983) 1070–1077. fot the optimization of steel compositions and process route, Expert
[77] T. Shinokura, K. Takai, Mathematical models of roll force and Sytems with Applications 17 (1999) 129–134.
torque in steel bar rolling, Journal of Iron Steel Institute of Japan 14 [100] T. Watanabe, H. Narazaki, A. Kitamura, A New Mill Set-Up System
(1986) 58–64. for Hot Strip Rolling Mill that Integrates a Process Model and
[78] H. Simon, Theories of Bounded Rationality Decision and Expertise, Process Technology Research Laboratory, Kobe Steel
Organization, R. Radner, North Holland, 1972. Ltd, 1997.
[79] E.C. Larke, The Rolling of Strip, Sheet and Plate, Chapman and Hall [101] C.S. Han, R.V. Grandhi, R. Srinivasan, Optimum design of forging
Ltd, London, 1967. die shapes using nonlinear finite element analysis, AIAA Journal 31
[80] British Steel Corporation, Roll Pass Design, Chorley and Pickersgill (1993) 774–781.
Ltd, Sheffield, 1979. [102] M.S. Joun, S.M. Hwang, Optimal process design in steady-state
[81] T.H. Kim, H.J. Kim, S.M. Hwang, Finite Element Modeling of metal forming by finite element method-I, theoretical considerations,
Shape Rolling of Complex Shaped Parts—A Steady State Approach International Journal of Machine and Tool Manufacture 33 (1)
Third International Conference Modelling of Metal Rolling (1993) 51–61.
Processes, London, UK, 1999 pp. 98–103. [103] M.S. Joun, S.M. Hwang, Optimal process design in steady-state
[82] S.-E. Lundberg, Roll pass design: the key function in control of metal forming by finite element method-II, application to die profile
design in extrusion, International Journal of Machine and Tool
shape, dimension and mechanical properties of hot rolled products,
Manufacture 33 (1) (1993) 63–70.
Scandinavian Journal of Metallurgy 26 (3) (1997) 102–114.
[104] S. Kobayashi, Process design in metal formng by the finite
[83] H.-C. Kwon, Y.T. Im, Interactive computer-aided-design system for
element method, Advanced Technology of Plasticity 11 (1987)
roll pass and profile design in bar rolling, Journal of Material
1213–1219.
Processing Technology 123 (2002) 399–405.
[105] S.M. Hwang, S. Kobayashi, Preform design in plane-strain rolling by
[84] F. Yamada, K. Sekiguchi, M. Tsugeno, Y. Anbe, Y. Andoh, C. Forse,
the finite element method, International Journal of Machine Tool
M. Guernier, T. Coleman, Hot strip mill mathematical models and
Design and Research 24 (1984) 253–266.
set-up calculation, IEEE Transactions on Industry Applications 27
[106] S.M. Hwang, S. Kobayashi, Preform design in disk forging,
(1) (1991).
International Journal of Machine Tool Design and Research 26
[85] T.L. Subramanian, T. Altan, Application of computer-aided
(231) (1986).
techniques to precision closed-DIE forging, Annals of the CIRP 27
[107] J.S. Chung, S.M. Hwang, Process optimal design in forging by
(1) (1978) 123–127.
genetic algorithm, Journal of Manufacturing Science and Engineer-
[86] N. Akerman, G.D. Lahoti, T. Altan, Computer-aided roll pass design
ing 124 (2002) 397–408.
in rolling of airfoil shapes, Journal of Application in Metalworking 1 [108] M.S. Joun, S.M. Hwang, Pass schedule optimal design in multi-
(3) (1980) 30–40. pass extrusion and drawing by finite element method, Inter-
[87] K. Osakada, T. Kado, G.B. Yang, Application of AI-technique to national Journal of Machine and Tool Manufacture 33 (5) (1993)
process planning of cold forming, Annals of the CIRP 37 (1) (1988) 713–724.
239–242. [109] S. Barinarayanan, N. Zabaras, Preform design in metal forming
[88] G. Perotti, N. Kapaj, Roll pass design for round bars, Annals of the Proceedings of the Fifth International Conference on Numerical
CIRP 34 (1990) 283–286. Methods in Ind. Forming Processes, Ithaca, New York 1995 pp. 533–
[89] K. Deb, Multi-Objective Optimization using Evolutionary Algor- 538.
ithms, Wiley, London, 2001. [110] L. Forment, J.L. Chenot, Optimal design for non-steady-state
[90] K. Osakada, G.B. Yang, T. Nakamura, K. Mori, Expert system for metal forming processes. I. Shape optimization method, Inter-
cold-forging process based on FEM simulation, Annals of the CIRP national Journal for Numerical Methods in Engineering 39 (1996)
39 (1) (1990) 249–252. 33–50.
[91] K. Sevenler, P.S. Raghupathi, T. Altan, Sequence Design for [111] L. Forment, J.L. Chenot, Optimal design for non-steady-state metal
Multistage Cold Forging, Journal of Mechanical Working Technol- forming Processes. II. Application of shape optimization in forging,
ogy 14 (1987) 121–135. International Journal for Numerical Methods in Engineering 39
[92] P. Bariani, W.A. Knight, Computer-aided cold forging process (1996) 51–60.
design: a knowledge-based system approach to forming sequence [112] J.S. Chung, S.M. Hwang, Application of genetic algorithm to process
generation, Annals of the CIRP 37 (1) (1988) 243–246. optimal design in non-isothermal metal forming, Journal of
[93] M. Brucker, D. Keller, J. Reissner, Computer-aided drawing of Materials Processing Technology 80–81 (1998) 136–143.
profiles from round and square bar, Annals of the CIRP 37 (1) (1988) [113] K. Deb, Optimization for Engineering Design: Algorithms and
247–250. Examples, Prentice-Hall, New Delhi, India, 1995.
[94] D.Y. Yang, Y.T. Im, Y.C. Yoo, J.J. Park, H.J. Kim, M.S. Chun, [114] R.A. Aliev, R.R. Aliev, Soft Computing and its Applications, World
C.H. Lee, Y.K. Lee, C.H. Park, J.H. Song, D.Y. Kim, K.K. Hong, Scientific Publishing Co Ltd, Singapore, 2001.
M.C. Lee, S.I. Kim, Development of integrated and intelligent [115] A. Tettamanzi, M. Tomassini, Soft Computing: Integrating
design and analysis system for forging processes, Annals of the CIRP Evolutionary, Neural, and Fuzzy System, Springer, New York,
(2000) 177–180. 2001. p. 328.
[116] C.A. Conceicao Antonio, N. Magalhaes Dourado, Metal-forming [122] J.Y. Jung, Y.T. Im, Fuzzy algorithm for calculating roll speed
process optimisation by inverse evolutionary search, Journal of variation based on roll separating force in hot rolling, International
Material Processing Technology 121 (2002) 403–413. Journal of Mechanical Sciences 42 (2) (2000) 249–272.
[117] V. Oduguwa, R. Roy, A. Tiwari, Evolutionary computing in [123] V. Oduguwa, R. Roy, An Integrated Design Optimisation Approach
manufacturing industry: an overview of recent applications, Journal for Quantitative and Qualitative Search Space Proceedings of
of Applied Soft Computing, Accepted for publication. ASME: 2003 ASME Design Engineering Technical Conference,
[118] V. Oduguwa, A. Tiwari, R. Roy, Genetic Algorithm in Process Chicago, IIIinois, 2003.
Optimisation CDROM Proceedings of the Eighth Online World [124] V. Oduguwa, R. Roy, Multi-Objective Optimisation of Rolling Rod
Conference on Soft Computing in Industrial Applications (WSC-8), Product Design using Meta-Modelling Approach Proceedings of the
Finland, 2003. Genetic and Evolutionary Computation Conference (GECCO-2002),
[119] S. Roy, S. Ghosh, R. Shivpuri, Optimal design of process variables in New York, Morgan Kaufmann Publishers, San Francisco, USA,
multi-pass wire drawing by genetic algorithms, Journal of 2002. pp. 1164–1171.
Manufacturing Science and Engineering 118 (1996).
[125] V. Oduguwa, R. Roy, D. Farrugia, Fuzzy Multi-Objective
[120] S.M. Hwang, J.S. Chung, Application of genetic algorithm to
Optimisation Approach for Rod Shape Design in Long Product
optimal design of the die shape in extrusion, Journal of Materials
Rolling Fuzzy Sets and Systems-IFSA 2003, 10th International
Processing Technology 72 (1997) 69–77.
Fuzzy Systems Association World Congress, Istanbul, Turkey,
[121] L. Nolle, D.A. Armstrong, J.A. Ware, F. Biegler-Konig, Optimiz-
Springer, Berlin, 2003. pp. 636–643.
ation of roll profiles in the hot rolling of wide steel strip Genetic
[126] C. Robson, Real World Research, Blackwell Publishing, Oxford,
Algorithm in Engineering Systems: Innovations and Applications
1997 pp. 133–138. 2002.

A Review of Rolling System Design Optimisation PDF

Uploaded by

Copyright:

Available Formats

A Review of Rolling System Design Optimisation PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

A Review of Rolling System Design Optimisation PDF

Uploaded by

Copyright:

Available Formats

International Journal of Machine Tools & Manufacture 46 (2006) 912–928

Keywords: Rolling system; Optimisation; Engineering design; Soft computing

1. Introduction Rolling System Design (RSD) is a complex problem

Rolling system design is the preparation of a series of roll

Fig 1. Fishbone diagram of the rolling system.

Fig 2. Features of rolling system design problem.

challenges for traditional optimisation algorithms. The main 2.1.3. Hierarchical

approaches for dealing with RSD problems in terms of ! 4J C 5 (3.1)

Fig 3. Typical breakdown sequence.

3.3.2. Approximation strategies for FE methods 3.3.3. FE method in RSD: observations

Fig 4. Optimisation approaches and NPD.

You might also like