Beruflich Dokumente
Kultur Dokumente
*
*
ISSN 1018-5593
European Commission
STEEL RESEARCH
EUROPEAN COMMISSION
Edith CRESSON, Member of the Commission
responsible for research, innovation, education, training and youth
DG XII/C.2 RTD actions: Industrial and materials technologies
Materials and steel
Contact: Mr. J.-L. Martin
Address: European Commission, rue de la Loi 200 (MO 75 1/10),
-1049 Brussels Tel. (32-2) 29-53453; fax (32-2) 29-65987
European Commission
i e&k j&t
Contract No 7210-EA/823
1 August 1990 to 31 July 1993
Final report
Directorate-General
Science, Research and Development
1998
EUR 17881 EN
LEGAL NOTICE
Neither the European Commission nor any person acting on behalf of the Commission
is responsible for the use which might be made of the following information.
A great deal of additional information on the European Union is available on the Internet.
It can be accessed through the Europa server (http://europa.eu.int).
Cataloguing data can be found at the end of this publication.
Luxembourg: Office for Official Publications of the European Communities, 1998
ISBN 92-828-3127-2
European Communities, 1998
Reproduction is authorised provided the source is acknowledged.
Printed in Luxembourg
PRINTED ON WHITE CHLORINE-FREE PAPER
CONTENTS
PAGE
1.
INTRODUCTION
2.
OPTIMISATION SOFTWARE
2.1
2.2
2.3
2.4
3.
4.
Schedule Optimisation
Evaluation of the Objective Function
Operational Constraints
Software Testing
g
11
14
15
16
3.1
3.2
3.3
3.4
16
17
18
19
20
4.1
21
4.2
24
5.
25
6.
27
7.
CONCLUSION
28
REFERENCES
29
TABLES
31
FIGURES
41
LIST OF TABLES
1.
2.
3.
4.
5.
Mechanical Properties of Softened and Descaled 302AA Grade Stainless Steel Hot Band
6.
7.
8.
9.
LIST OF FIGURES
1.
2.
Total Strip Gauge Reduction Displayed as a Function of the Individual Pass Reductions
3.
4.
True Stress Natural Strain Type 302XD Longitudinal Results of Tensile Tests
5.
True Stress Natural Strain Type 302XD Transverse Results of Tensile Tests
6.
True Stress Natural Strain Type 302XD Transverse Results of Tensile Tests at Different
Temperatures
7.
True Stress Natural Strain Type 302XD Longitudinal Results of Tensile Tests at Different
Temperatures
8.
Proportion of Martensite in the Tensile Test Specimens Against Test Temperature - Values
Determined by Ferriscope Measurement
9.
10.
11.
12.
13.
14.
15.
16.
Undeformed Mesh
17.
18.
Vertical Strain Component (E22) as Viewed from the Front of the Mill
19.
20.
21.
22.
23.
Screwdown Pressure Signal Gauge Output for Bottom Back-Up roll Eccentric Positions
24.
25.
26.
27.
28.
29.
Work Roll Torque per Unit Width Front Tension Stress for First Passes
30.
31.
32.
33.
34.
1.
INTRODUCTION
The aim of this project has been to test the feasibility of implementing the on-line calculation of time
optimised rolling schedules for Sendzimir rolling mills. Cold rolling is a complex physical process, which is
further complicated by the fact that hot rolled coils of stainless steel can have individual characteristics.
The variability in the properties of processed strip is generated by the natural fluctuations in processing
parameters that occur during casting, hot rolling, annealing, descaling and grinding. The variability of
hot rolled coil properties necessitates that on-line adaptive schedule calculations are performed during
subsequent cold rolling. The schedule calculations are therefore repeated after each pass. The new
calculations are based upon measurements of coil properties obtained during cold rolling, such that the
material in question is always rolled in an optimum manner.
Scheduling can be implemented in a number of different ways. Firstly, one can use a model to predict the
mill loadings and adjust the schedule until maximum rolling force and the minimum number of passes
have been achieved. The mill speed is free to vary from pass to pass at the discretion of the mill operators.
Adaption of the schedule is achieved by measuring roll separating force, from which the yield stress is
inferred and subsequent more accurate roll separating force are predicted. This approach is favoured for
reversing hot rolling mills, where rolling speed is limited. It would also be suitable for application on fourhigh or Z-high cold rolling mills which have limited reduction and speed capability.
Alternatively, one can generate a model which predicts both the mill rolling force and maximum mill
speed throughout a schedule. A time optimum schedule may then be calculated which seeks to find the
balance between heavier draftings and faster rolling speeds. This approach is deemed to be applicable to
rolling on Sendzimir mills. These mills are generally equipped with powerful motors which allow rolling
speeds in the range 500-1000 m/min to be achieved. However, the rolling of quality stainless strip also
requires that certain criteria are obeyed regarding drafting, strip speed and tension. To achieve maximum
productivity of a high quality product, the schedule required must respect the rules applied, yet exploit the
available mill power. Schedule adaption can be implemented by using measured load data as a means of
refining the predictions made. However, it must be realised that adaption is also a means of error
correction. One cannot claim to have time optimised scheduling if the result on rolling requires
continuous large error correction. One must therefore aim to achieve suitable accuracy in the predictions
of the model such that adaption represents a small perturbation to the predictions.
The proposed solution to this problem is a computer software package capable of providing calculated
schedules within the time required to roll one pass. The software developed must therefore contain all the
information required to predict the rolling forces generated at the work roll for any proposed reduction
sequence. The main elements of the proposed system are shown in Fig. 1. Essentially the system consists
of an optimising procedure, a model of the mill mechanics, and a model of cold rolling describing the
behaviour of stainless steel strip during rolling reduction.
The approach sought is both theoretical, in that the aim is an improved understanding of the cold rolling
processes, and practical, in that a focus has been retained on the overall aim of the project and its
importance to stainless strip manufacturers. Hence, both theoretical and empirical models of material
behaviour under deformation have been included in the design of the software package. More details of
the software and its functionality are given in Section 2.
The optimisation procedure can be successful only if it can accurately predict the behaviour of the strip
under any given processing conditions. Therefore, the mechanisms which control the yield stress of rolled
stainless steel strip have been investigated. This work is described in Section 3 below.
In order for adaptive scheduling to be implemented, a measure of a coil's properties during rolling must be
obtained; this is best achieved by measuring rolling load. Great effort has therefore been expended in
attempting to sense the roll separating force generated by a production Sendzimir mill. This work is
detailed in Section 4 of this report.
2.
O PTIMISATI
O N SO FTWARE
A large suite of software has been developed for the calculation of optimised rolling schedules. The
program has been written in FORTRAN 77 and consists of numerous subroutine modules. Great care has
been taken to ensure that the code operates in an efficient manner, both in terms of the number of
calculation steps required and the memory allocation demanded. Figure 1 shows the connectivity of the
main program modules; the function of these routines is discussed below. The source code requires a
storage capacity of 100 kb. The program can therefore be operated on a personal computer of reasonable
capacity. The code can be thought of as comprising two parts, the optimisation algorithm, which locates
the minimum of the function, and the function algorithm, which computes the objective function to be
minimised. In Fig. 1, the routines which comprise the function evaluation are located within the large
box. The two components are discussed more fully below.
2.1
Schedule O ptimisation
The optimisation routine provides the main driving force of the programme. This module has been
significantly developed since first reported. The aim has been to develop an algorithm that calculates the
rolling schedule which minimises the total processing time, and which completes the required calculations
within a time suitable for on-line mill application. To ensure these aims are met the structure of the
problem have been exploited to the maximum. The function to be minimised is the total rolling time T,
this is given by:
t.
(!)
i = 1
where t is the time to roll the i t h of passes. does not include the inter-pass delay time. This time is not
a function of the parameters which control the schedule and thus would only be presented as a negligible
additive constant if included in the above equation.
The values t are dependent upon the reductions and tensions used for each pass. The rolling time is thus
dependent upon the variables x, Sf, S b , where Sf and Sb are the front and back tensions respectively. The
parameter x is defined as:
h. , - h.
l
x. = - ^
1
..-(2)
where ho is the starting strip thickness and h, is the thickness after rolling pass i. This variable has been
specifically chosen in preference to the fractional reduction achieved at each pass. This is because the
values of x must satisfy:
o S x S 1
. . . (3)
and hence,
*,=
h0 - h.
h h
O n
h, - h2
+ . . .
h_ h
O n
= -1
...(4)
The value of n can be chosen to be as large as is convenient as x = 0 has a valid meaning. The summation
of the Xj to unity implies t h a t for n passes there are n-1 free variables. This situation is shown
diagrammatically in Fig. 2. The complete reduction is represented by the area of the circle, the values of x,
are shown as the circle segments. The parameters to be varied are the n-1 barriers which define the n
segments. The problem can thus be stated as:
minT =
n
Y
t. (. . . . , S F , S B )
.-.(5)
i = 1
It will be noted that if Xj is changed then only ti, i > j need be recomputed. The tensions Sfi, Sbi are bounded
by S m i n . < S < S m a x ., where S m i n . and S m a x . are mill dependent parameters. If either tension is varied for
pass i then only the time t, requires evaluation.
By configuring the problem such t h a t the dependent variable ranges are constrained by physical bounds, a
direct search algorithm can be employed to find the function minimum. It will be noted that if one changes
the value of x, then Xj is given by:
x. = 1 j
x, - .
k
(6)
where the values of xt, k * i , j are stationary. The change in x will however effect all the values of strip
thickness h^ k > i. Thus a multivariate search is conducted in the space of the strip thicknesses as a result
of a univariate search in the space of the variable x. This property is the driving force behind the use of
these variables. Univariate search algorithms are less complex than their multivariate counterparts, in
that they do not suffer from the 'explosion' of iterations required during the search.
A second advantage of a direct search algorithm, is that the search operates by examining the entire
functional space defined within given bounds, thus ensuring that the global minimum of the function is
obtained. Also, any constraint which is a function of the optimising variables can be simply introduced.
The constraints define regions within the optimising space of the variables which may not be entered by
the search algorithm. This mechanism dispenses with the need for penalty or barrier functions and the
associated function evaluations required to determine the geometry of the objective function.
The location of the constrained boundaries is implicit in the formulation of the direct search problem. The
designed optimiser therefore operates in two phases. The first phase locates a point within the variable
space which is not forbidden by the constraints applied. The second phase of the algorithm uses this first
point as a means of determining the function boundaries and as a starting point for the minimisation
algorithm.
2.1.1
Phase O ne Algorithm
The phase one routine uses a binary search algorithm to locate a point in the space of the optimising
variables which is not forbidden by the required constraints. The values of S/, S b i and x are set in turn
beginning with pass one. Front and back tension values are set to their maximum allowable values S m a x - .
The reduction fraction is then set using the binary search algorithm. The value of xi must lie in the
range 0 < x i < l . In practice to improve the algorithm speed the upper limit is set at 0.4. The search
algorithm locates the upper boundary between the allowable and constrained regions of the variable space
along the axis of the chosen variable. The tolerance to which the boundary edge is detected is set at 0.05.
This tolerance is large for reasons to be described below. Each of the fractions is set in turn in this way.
The upper limit for the search for the value of x is determined by the requirement that the x should sum
to unity. Problems occur at the end of this procedure; the final pass value x n will be determined by the
values of the preceding passes. However, this value will often be small and will lie in a constrained region
of the variable space. This constraint is produced as a result of requiring there to be a lower limit on roll
separating force or rolling reduction. The use of a wide tolerance when locating the boundaries of the
operational space can now be justified. Pass 1 is re-examined; if a value x i 1 was obtained during the first
search, a new search is conducted for the boundary of the constrained space in the region 1 < < x ^ +x n .
This procedure is repeated until x n is distributed amongst the preceding n-1 passes. The number of passes
required for the schedule is thus defined. The dependence of a procedure on chosen tolerances is not a
favoured practice. However, testing of the algorithm has shown that the values set will allow the required
functionality. Should the procedure fail, the tolerance used in the first pass search is increased and the
procedure repeated, this ensures successful solution.
2.1.2
The phase two algorithm firstly determines the location of the boundaries of the operating space. A direct
search method is then used to detect the function minimum within the region defined. The procedure
utilises the variables x to maximum effect. The minimum is found by a s k i n g the question, 'would
increasing the magnitude of a pass value x, at the expense of another pass value XJ, j * i , decrease the total
rolling time?' Referring to Fig. 2, this can be seen to be equivalent to asking how the area contained in two
chosen segments may be redistributed, whilst keeping the area of all other segments the same, such that
the rolling time is reduced. Thus if x t = x + Xj, then a search for the minimum can be made by using a
univariate search in the range, 0 < x < x t with
XJ
= x t - Xi
. . . (7)
In the plane of x, Xj this will be a search along the straight line as shown in Fig. 3. The search, however,
can only be made in regions where the function is unconstrained by mill or operational limits. Figure 3
depicts these regions, which have been found to be numerous and complex. The nature of the constrained
regions is a function of all the optimising variables and hence will change as the schedule is modified. The
programme therefore determines the segments of the search line which can be examined. This is achieved
by a binary search algorithm. The points x = ;1 obtained from the phase one search are known to lie in an
unconstrained region of this space, the points x = 0 and x = x t are assumed to lie in the constrained
region. The binary search is then made for the boundary of the constrained region between x; 1 and Xi = 0
and similarly between ,1 and x = x t . When a boundary is located the search is continued to find the next
point where the function can be evaluated, this enables 'islands' of constrained regions to be located. When
all the segments of the search line have been located, the minimum value of the objective function is found
for each of the segments. The search method of Fibinacci is used for this purpose 1) . This algorithm
necessitates evaluating the objective function at a series of points on the line segment. The algorithm
locates the minimum by bracketing the minimum point with two points. These points are moved towards
one another until they are separated by a preset tolerance of dx = 0.001. This value has been chosen
because, for strip rolling in the range 5.0-0.5 mm, it represents an u n c e r t a i n t y in the reduction of
approximately 0.005 mm, which is beyond the control of Sendzimir mills.
The position of the deepest minimum, obtained by searching all the line segments, is taken to be the new
value of x. Throughout, the value of Xj is defined by Equation (7). The algorithm proceeds by examining
all possible pairing of passes. At each step the rolling time will be either reduced or remain constant, as
the initial starting point of the search must be improved upon. The cycle of adjusting the values for each
pairing continues until no further improvement in rolling time can be found. The values obtained are
assumed to represent the optimum schedule. The points (x = 0.0, XJ = xt) and (x = xt, XJ = 0.0) are
evaluated during each cycle, it is therefore possible for the algorithm to reduce the number of passes
required for the schedule.
10
2.1.3
The front and back tension values are determined by the same procedure as is used for the variables x. The
values are constrained to lie in the limits Smin. < S < S max ., unless otherwise restrained by operational
requirements. The binary search for the constraint boundaries and a Fibinacci search for the function
minimum are performed as described above. The optimisation of front and back tensions for a pass i is
made everytime a search along the value of x is completed. This is required, because although the
variables x, S^ and SD are independent, the position of the boundary of the constrained space, at the point
of intersection with one variable axis, is a function of the other variables. For example, the upper limit on
the reduction variable may be controlled by a load constraint, varying values of Sf and S b will move the
position of this constraint along the axis. In principle this dependence should require that the searches
over the three variables should be nested. However, the partially nested approach described above has
been found to yield the required results, with a great saving in computational effort. The algorithm has
been found to function satisfactorily by setting the tensions to their maximum values in the phase one
computations. The tensions set therefore allow the largest possible reduction space to be searched during
the evaluation of the variables. Usually the tension will only be altered so as to reduce the torque on the
coiler motors and allow greater rolling speed to be achieved. For a well designed mill one would expect this
condition never to be enforced.
2.2
At each point in the space of the control variables defined by the search algorithms, the rolling time given
by Equation (1) must be evaluated. In addition, if the schedule in question violates the constraints
imposed, then a logical flag must be set and returned to the optimising routine. The optimiser therefore
requires only two pieces of information to operate, the total rolling time and an allowed/forbidden or
pass/fail flag (Fig. 1).
To calculate rolling time for a given reduction, the torque required at the work roll and hence the torque
which must be supplied by the mill motor, must be either known or calculated. The rolling time
calculation therefore depends upon two mathematical models. The first provides the roll separating force
and torque for a given reduction, the second models the behaviour of the mill mechanics and evaluates the
motor conditions required to generate the forces needed at the work roll. These routines are described in
more detail below.
2.2.1
Mill Mechanics
The code has been constructed so as to represent a 20-roll Sendzimir mill equipped with two coilers. The
modular structure of the program would allow this code to be replaced by that describing, two-high, 4-high,
Z-high or any other roll cluster design operating as a reversing mill. Application to tandem mills could
only be achieved through modifications to the overall code structure. Clearly the exact code used in this
module is unimportant as the formulae are mill dependent and would have to be constructed so as to
simulate the characteristics of the mill in question. However, the basic calculations performed will be
common to all cold rolling mills and these are described in more detail below.
The first routine called in this module provides the input of data describing the mills construction and
capabilities. The required data is stored in a file preprepared by the user. The parameters required are
listed in Table 1; for instance, a set of typical data for a commercial operational Sendzimir mill is shown.
The remaining routines in the module are called during the pass time calculation. The torque which is
required to be generated at the work roll and the applied tensions are inputs into the routine. If the torque
required is within the capacity of the mill then the maximum rolling speed possible is returned to the
calling routine. The key elements of the calculations are as follows;
II
1.
Calculate the mill motor torque required to produce the input work roll torque.
2.
Check that the input roll separating force is less than the maximum roll separating force on
the mill
3.
Check that the roll separating force is within the c o n s t r a i n t s applied for q u a l i t y or
operational reasons.
4.
Check that the input torque is less than the maximum torque rating of the mill
5.
Calculate the maximum mill motor speed possible for the required torque.
6.
7.
If the strip speed is greater than the maximum mill speed, reset the strip speed to the
maximum value.
8.
If the strip speed exceeds the limit required for quality or operational reasons reset to the
speed to this limit.
9.
Calculate the rotational speed of both the coiler motors, at maximum and minimum coil
diameter at the set strip speed.
10.
Calculate, using the required tension values, the torque acting on the coiler motors at
maximum and minimum coil diameter.
11.
12.
If the coiler power required is greater than the maximum power rating of the coiler motor,
calculate the motor speed which will allow maximum coiler power to be utilised. Recalculate
the strip speed obtainable.
13.
Calculate the rolling time using the exit strip speed calculated.
The rolling time is calculated by allowing for uniform acceleration and deceleration of the strip at the pass
start and end. To represent true mill actions, modifications would be required to this formula. At the start
of a pass, the crews often require the mill to run more slowly whilst the As-U-Roll (shape control) settings
are adjusted and good shape is achieved. One can not take account of such events, which may occur at any
time in the rolling process, and their omission is not expected to invalidate the choice of optimised
schedule.
2.2.2
The function of this module is to provide an estimate of either the rolling force or the rolling torque
generated at the work roll. A series of parameters defining the material to be rolled, its prior rolling
history and the pass for which the calculations are required, is input into the routine. These parameters
are given in Table 2. The rolling model option flag, allows the user to select one of several cold rolling
models as a means of roll separating force and torque calculation. The models of Stone, Ford and Orowan
have been encoded. It is not necessary to give details of these algorithms as adequate descriptions of the
mathematical procedures required and the assumptions used in these models may be found in References
2-6. All of the models require yield stress information for the material being considered; this is supplied
by subsidiary routines and will be discussed more fully below (Fig. 1).
A fourth rolling model option also exists, that of an empirical or stochastic model based on continuously
monitored plant data. It is this model which is most likely to be applied in practice. The reasons for this
choice are discussed more fully in Section 3. Here the software required to implement both the theoretical
and empirical methods is outlined.
12
(a)
zii-vyiii-i
ijk...
'
"
(8)
The constants aj.. of this multidimensional polynomial may be obtained by a least-squares fit
to the data contained in the database. This procedure a s s u m e s t h e torque values a r e
normally distributed at each point in the space of the control variables. On the collection of a
large statistics base, a parameterfitbased upon the principle of maximum likelihood would
be possible and preferable. The fitted equation represents the rolling torque value most
likely to be required for a pass described by the given control parameters. The rolling torque
values predicted by this equation are therefore suitable for a n initial computation of the
optimised schedule. Once a characteristic data set has been collected for each product type,
the fitting procedure to the data collected need only be repeated a t regular time intervals so
as to account for any gradual variation in prior processing conditions. The evaluation of the
polynomial is computationally highly efficient and aids the desire for speedy optimisation.
Most importantly by adjusting the number of terms in the polynomial, and examining the
dependence of the residual errors on each of the control p a r a m e t e r s , it is possible fully to
account for the dependence of the rolling torque on each control parameter.
(b)
. . . (9)
describes the torque values of the remaining unrolled passes. The residual r will be different
for each coil rolled and will be a function of the p a r a m e t e r s which describe t h a t pass.
Therefore, for adaption to be possible, t h e torque for a pass discriminated by control
parameters V, must also be a function of the residual r for the previous pass and the control
parameters u which describe that pass. This modifies Equation (8) as:
= - ^ . . . ^ . v ^ i ^ - ^ ^
i
(10)
13
(c)
Yield Stress
As described above, the cold rolling models incorporated into the programme require
material yield stress data in order to perform roll separating force and rolling torque
calculations. A database system for the storage of yield stress curves has therefore been
developed. The data contained in the database can be obtained either by laboratory testing,
or calculated from measured roll separating force load and/or torque data. The yield stress
curve is represented by a series of discrete coordinates of reduction and stress. Values on the
curve are obtained by linear interpolation between the measured data points. The yield
stress element of the programme consists of three separate routines concerned with file
handling, linear interpolation and adaption.
A yield stress file is created for each product type, in this is stored the set of reduction - stress
coordinates obtained from each coil rolled. On start-up of the programme this unit checks for
the existence of the required file, if this file cannot be found default yield stress data is used.
If the file is located, the stored data is used to calculate the average value of the yield stress at
eight equally spaced points in the range 0 to 80% reduction. The average values represent
the most likely state of the coil to be rolled given no prior knowledge of its condition. Each of
the cold rolling models encoded utilises the linear interpolation routine to predict the yield
stress at the required value of total reduction.
The yield stress values are obtained from the measurements of rolling force or torque by
using one of the encoded cold rolling models. A binary search algorithm is used to propose
yield stress values in the range 0-3000 N/mm 2 . At each point in the search the rolling model
is used to predict the roll separating force or torque. This is compared with measured value
and the search continues until the difference between predicted and measured values lies
within a preset tolerance.
The yield stress values calculated after each pass may be used to adapt the calculated rolling
schedule. At the end of rolling, the complete yield stress record for the coil is stored in the
database. Thus, the system continuously builds its knowledge of the behaviour of the
material and improves the accuracy of prediction.
(d)
2.3
Operational Constraints
Mathematical modelling can provide a description for many of the physical processes which occur during
cold rolling. However, there are many effects, associated with the product surface and shape quality,
which at present may not be represented in a simple mathematical form. All processes are to some extent
a function of the p a r a m e t e r s which describe the pass, assuming the processes to be a repeatable
phenomenon and not dependent on random 'environmental' factors. A series of simple operational
constraints can be applied to these descriptive variables, so as to constrain the mill operation within an
envelope which forms a barrier between good and poor quality production.
14
The application of the constraints is achieved through a database of operational conditions. A file is
created for each product type in which the following information is stored.
Constraint type
Constraint limit
Constraint variable
Descriptor list
Two types of constraint have been encoded at present. Application of the programme to the production
environment will inevitably produce additional requirements. The code has been structured with these
developments in mind. Both constraint types enforce upper or lower limits on the constraint variables.
The constraint variables are:
Roll Separating Force
Pass reduction
Front Tension
Back Tension
Rolling speed
The first method applies constraints which are conditional on the pass type. The pass types available are;
First, Penultimate, Last and Middle. The second constraint type is conditional on the total reduction
applied to the strip lying in a specified band. This mechanism allows the application of a constraint on
pass reduction to act as a constraint on mill load. Upper and lower constraints on roll separating force are
of importance to strip shape control. The model must be constrained to operate in a regime where the
cambers ground onto selected rolls can offset the roll stack deflection and allow shape to be controlled
through As-U-Roll movement. The roll separating force constraint can thus be applied directly if a force
prediction is available, else it may be imposed through limits on the pass reduction.
The code consists of an initialisation and file handling routine and a constraint checking routine. The
initialisation routine searches for a constraint file for the product to be rolled. If the file is not located, a
new file is opened and the user is prompted to provide the relevant information. Once the constraints have
been defined the information is stored in memory for rapid access by the checking routine.
Calls are made to the constraint checking and application routine by various elements of the optimisation
programme. The variable to be checked and its value are input to the subroutine. It is checked against the
constraints for this variable type. If the value lies outside the set limit, a logical flag is set and returned to
the calling routine. This procedure is followed for all the variables except speed. If the strip speed value is
outside the set limits, the value is reset to that limit. This procedure can be adopted as speed is not an
optimising variable, but a variable calculated by the mill mechanics module.
Limits on the reduction and tension set for a pass are checked immediately after they are provided by the
optimising routines, this prevents unnecessary calculation of the rolling time. Limits on roll separating
force and speed are investigated by the modules concerned with the descriptions of the mill mechanics,
again care is taken to prevent unnecessary calculations being performed.
2.4
Software Testing
The optimisation algorithm seeks to find the minimum of a function in a multidimensional space.
Checking the results of this procedure may be performed by evaluating the function throughout the space
and recording the minimum value encountered. A series of test problems were chosen, and in each case
the objective function was evaluated at points defined on a preset three dimensional matrix. For each pass
the reduction parameter was evaluated in the range 0.0 to 0.5, at intervals of 0.005. Front and back
tensions were evaluated in the range 0, S m a x at intervals of 10 kN. The number of passes was set at ten in
all cases, this being judged sufficient so as to explore all possible function minima. The problems chosen
were also evaluated using the optimisation routines. The optimisation programs were run on a VAX 3100
workstation and a COMPAQ 3086/20 personal computer. The function tests were evaluated on a
DEC VAX 6100 computer. In all cases the optimisation program was found to have located the function
15
minimum to within the resolution of the multipoint search. In Table 3, the computation time required by
the optimisation procedure is displayed. The examples chosen correspond to actual rollings on a
production Sendzimir mill at Avesta Sheffield Shepcote Lane Works. The optimised and actual number of
passes rolled are indicated in Table 3.
It can be seen that the procedures operate at a speed sufficient for implementation in a n on-line
environment using a computer of relatively low capacity and cost. The time required for a solution is
related to the number of passes required and hence the number of parameters to be set. However, the
application of constraints to the variables may significantly reduce or increase this time. This result
reflects the complexity of the optimising space in which several competing local minima may be created.
In a production environment it may be prudent to store all calculated schedules with key accesses by
product grade, width, starting and finishing gauge and surface finish type. The relevant file may be
searched prior to the commencement of the optimising procedure. A date marker for each file would also
be essential, this would ensure that the schedule is re-calculated regularly and g r a d u a l changes in
operational conditions are incorporated.
3.
The success of a scheduling model depends upon the accuracy with which the mechanical properties of the
rolled strip can be predicted. This entails locating the dominant parameters which give rise to variations
in the material properties during rolling. To this end, the effect of strip temperature and annealing
practice on the yield stress of cold rolled strip have been investigated. To determine if the yield stress is
dependent on the type of deformation mechanism, tensile and compression testing of stainless steel strip
has been performed, in conjunction with cold rollings on a small experimental mill. The data obtained
have enabled the suitability of laboratory derived yield stress values for the theoretical calculation of
rolling forces, to be determined.
3.1
In determining the applicability of an on-line adaptive scheduling system, the metallurgical changes
which occur within stainless steels and the subsequent variation in macroscopic properties are of
importance. For example, it is known that certain grades of a austenitic stainless steel are susceptible to
the formation of martensitic phase which is produced via a diffusionless transformation from the parent
austenite. This phase change occurs at low temperatures and is initiated by deformation.
The effect of this change has been investigated for grade 302XD stainless steel, the chemical composition
for the grade is given in Table 4(a). Material samples were obtained from the same coil of softened and
descaled hot-band material both before and after each of a series of cold reductions was made on a
production Sendzimir mill. Tensile test pieces were cut from the strip both longitudinally and transverse
to the rolling direction. Tensile tests were performed on each of the rolled samples to record the ultimate
tensile strength and 0.2% proof stress. Further tensile tests were carried out on the undeformed strip to
describe the initial part of the stress-strain curve in more detail.
The results of the different tests were combined to form the complete stress-strain curves shown in Figs. 4
and 5. It can be seen that there is a discontinuity at the junction between the rolled and un-rolled sample
curves. This difference is attributed to martensite formation during the relatively low room temperature
tensile tests on the un-rolled samples, but the high first pass deformation temperatures on the Z-mill
prevent the formation of martensite in the pre-rolled samples. No appreciable difference between the yield
stress measured transverse and longitudinally to the rolling direction could be discerned, indicating that
the material is isotropic in the plane of the strip.
Martensite transformation is associated with a shear strain and may be regarded as a deformation that
competes with slip. With decreasing temperature the resistance to slip increases, whereas the resistance
to martensite formation decreases. Above a certain temperature, the Mj temperature, no amount of
deformation can bring about the transformation. The Md3o temperature is an arbitrarily chosen reference
at which a volume fraction of 50% martensite is formed for a true strain of 0.3<7\ An empirical equation
relating steel composition to the ^3 temperature has been derived<8);
Id
Md3o (C) = 497 - 462 [C + N] - 9.2 [Si] - 8.1 [Mn] -13.7 [Cr] - 20 [Ni] -18.5 [Mo]
. . . (11)
Using the values given in Table 4(a) for Grade 302XD stainless steel, the M<j30 temperature is expected to
be 25.8C. To investigate this phenomenon further, the M<j and M3o temperatures for Grade 302XD have
been determined experimentally. Eighteen NFT2 test pieces were cut from softened and descaled hot
band, both longitudinally and transverse to the rolling direction. The test pieces were subject to simple
tension, isothermally , for a range of temperatures from 8C to 100C. The amount of martensite formed
was determined by ferriscope measurement and plotted as a function of testing temperature.
To enable isothermal testing to be carried out for a given temperature, a special thermal jacket was
designed with an insulated heating element which could be regulated to 3C. For the low temperature
tests the specimens were surrounded by ice. The problem of rising temperature due to mechanical working
was overcome through the use of a low testing speed of 0.5 mm/min. The testing was carried out on an
Instron tensile testing machine with automatic recording of the load/time curve. The true stress/strain
curves for longitudinal and transverse specimens are shown in Fig. 6 and 7.
Ferriscope measurements were made for each sample using material taken near to the necking region.
Figure 8 illustrates the proportion of martensite produced at each of the test temperatures. It can be seen
that the M temperature is 70C and the Mj3o temperature is approximately 15C.
These results confirm that significantly different yield stress curves may arise owing to the temperature at
which Grade 302XD is initially deformed. This is significant to cold rolling operations, because
measurements have shown that first pass reductions of the order of 25% result in strip temperatures of
some 40-50C. Clearly variations in reductions and rolling speed will effect the percentage of martensite
produced during rolling and hence the yield strength behaviour on further rolling.
3.2
Annealing Treatment
Another contributing factor to the possible variations in material properties is the annealing procedure
used prior to rolling. A survey was made of the hardness, ultimate tensile strength and 0.2% proof stress of
several post annealed coils of Grade 302AA material produced at a commercial s t a i n l e s s wide strip
production facility. This data is displayed in Table 5. The variation in UTS and 0.2% proof stress is
indicative of a difference in the metallurgy of the coils investigated. To determine how the incoming
hardness variation effects yield stress, a series of cold rolling trials and tensile tests were performed.
Samples of 5 mm thick hot band material were taken from the front end of a coil. Full width (1500 mm)
samples were cut from the coil on receipt from the hot mill. The chemical analysis of the sample is shown
in Table 4(b). It was noted that there was some variation in the thickness of the hot band material and
samples with a minimum gauge variation were selected. The samples were cut into pieces of dimensions
5 mm thick X 31 mm wide X 305 mm long, which are slightly greater t h a n s t a n d a r d sized tensile
specimens.
In order to study the effect of annealing and subsequent cold rolling on the yield strength of the material, it
was decided to anneal the test samples, in the laboratory, to the centre, maximum and minimum of the
possible commercial hardness range. However, prior to annealing, the initial hardness was measured.
This showed that a considerable initial hardness variation (182-254 VPN) exists. To determine the heat
treatment required to produce samples of a certain final hardness, a series of tests was made. Samples
were placed in an annealing furnace at a temperature of 1100C for various times. The hardness was then
measured, enabling the approximate times required to produce material in three hardness bands to be
determined.
5.5 m i n a t l l 0 0 C
3 minatll00C
2 minatll00C
139-141 VPN
145-162 VPN
160-182 VPN
17
Figure 9 shows the furnace residence time against the resulting range of hardness. It can be seen that the
final hardness cannot be exactly controlled and the samples produced therefore form a continuum of
hardness values in the range 137-185 VPN. Representing partially to fully annealed material conditions.
All three batches of annealed specimens were then descaled by Swinden Laboratories' corrosion laboratory
in a dilute mixture of nitric and hydrofluoric acids at a temperature of 60C.
Laboratory cold rolling of the specimens was then carried out on a two-high Hille mill fitted with 100 mm
diameter, 100 mm wide rolls. The mill was fitted with accurately calibrated load cells, to measure the roll
separating force generated during rolling. The cells were found to be accurate to within 0.5% of the
measured load value. The specimens from each of the hardness ranges were rolled down to final reductions
of 10, 20, 30,40, 50, 60 and 70%. Single pass reductions of between 10% and 35% were also carried out on a
number of samples from each batch. Tables 6-8 show the roll separating force and percentage reduction
data obtained from the rolling experiments. In Fig. 10, the cumulative rolling force for the samples from
the three hardness ranges is shown. It can be seen that the load variation between the hardest and softest
material is approximately 10% after a 10% reduction is applied. However, after the initial reduction is
performed, the loads produced are effectively uniform for all samples over the remaining reduction
sequence.
After rolling, the specimens were machined into Type NFT2 tensile specimens and then tested to produce a
load extension curve. The envelope of these curves describes the true-strain curve. Figure 11, shows the
envelopes produced for the three hardness ranges. It can be seen that despite the large difference in
hardness, the yield stress measurements produced indicate little difference in the material mechanical
strength. This is confirmed by the rolling force measurements shown in Fig. 10. It appears that a small
variation in yield strength can be produced by annealing practice. However, this variation is rapidly
overcome by the reduction process, with a 10% reduction being sufficient to render the strip to a uniform
condition. It must be concluded that in the industrial production of this product, where tight control is
kept on furnace temperature and line speed, the variation of hardness is not a reflection of greatly varying
yield stress.
3.3
Cold rolling of stainless steel on Sendzimir mills is performed using higher tensions than those associated
with the two and four high type mills for which the cold rolling theories of Stone, Orowan, and Bland and
Ford were originally developed' 2 - 6 '. The behaviour of yield stress under both tensile and compressive
forces is therefore of interest. Plane strain compression tests were therefore performed to complement the
tensile work described above.
A plane strain testing rig was designed and constructed. To ensure plane strain conditions are achieved
certain criteria were used in the design of the compression tools. The platen breadth, b, was chosen to lie
in the range:
l/4<h/b<l/2
where h is the mean stock thickness throughout the compression stroke. The limits on platen breadth
ensure that the load measurements obtained are equal to the yield stress in plane strain to within a
maximum error<9> of 2%. At values of h/b less than \, frictional effects begin to introduce errors into the
measurement. At values of h/b greater than ^ the loading pressure required to achieve deformation is
greater than the plane strain yield stress and rises with stock thickness.
A platen width w, such that w > 5b is required to ensure negligible lateral spread during compression. In
order that the tests could be performed on samples of 5 mm gauge 302AA hot-band material, previously
obtained for the tensile tests described above, the following tool dimensions were chosen; b = 10 mm,
w = 100 mm. The tools were designed by Swinden Laboratories' design office.
18
The tools were manufactured so as to conform to British Standard BS4659-BD2, and were hardened and
tempered to 54/56 RG. An EXACTA die set of design type MCP was chosen to house the tools. This design
was chosen so as to ensure that there would be no deflection of the tools when under a load of 200 t. The
tools were mounted centrally onto the die set and great care was taken to ensure the tool faces remained
aligned and parallel throughout the compression stroke.
Stainless steel samples were precision cut to a width of 100 mm and an approximate length of 300 mm.
These samples were then annealed in an annealing furnace at 1100C for six minutes. The hardness of the
samples was measured and was found to span 142-155 VPN.
Compression testing was performed by an hydraulic press with a 250 t load capacity. The load was
measured by a proving ring calibrated to a NAMAS accredited standard before the compressions were
performed. Each test sample was lubricated using cold rolling oil obtained from a Sendzimir mill. The
compressions were performed as a series of increasing indentations with each test sample able to
accommodate up to ten individual compressions. Due to the confined space available and safety
considerations, the reductions were performed to a pre-determined load. The test piece was then removed
from the tools and the indentation measured at several positions across the width with a micrometer.
The results of these tests are shown in Fig. 12. It can be seen that the yield stress is significantly higher
than that derived from tensile testing of fully annealed material, as shown in Fig. 11. The ratio of
compressive to tensile yield stress may be calculated for the fully annealed samples which were tested at
room temperature. In Fig. 13 this ratio is plotted against fractional reduction. The results are surprising;
theoretically one would expect the ratio to be 1.155(9) and independent of reduction. However, it can be
seen that there is a marked dependence of this factor on reduction. In Fig. 13, the results obtained by Ford
for copper are also shown. The conclusion is that metallurgical anisotropy may exists in metals and the
assumptions used in cold rolling models may have to be significantly modified.
Roll separating force data, obtained during the cold rolling trials detailed above, has been modelled using
the rolling theories of Stone and Ford. For each sample rolling, a simple binary search algorithm was used
to determine the yield stress which must be entered into the model, so as correctly to predict the measured
force. The ratio of this inferred yield stress and the measured tensile yield stress is shown in Fig. 13 also.
It can be seen that the ratio indicates that the yield stress required by the cold rolling model is
significantly greater than the tensile yield stress value, but less than the compressive test value. This is to
be expected as cold rolling is predominantly a compressive operation in which smaller tensile forces are
also present.
One could therefore assume that the use of compressive yield stress data is valid for use in describing
rolling on two and four high mills, where little tension is applied and large work rolls are used.
3.4
The tensile tests performed on Grade 302XD lead to the conclusion that the possible variations in yield
stress produced by industrial annealing of austenitic stainless steels will be negligible when compared to
the possible variations produced by thermal effects during rolling (see Figs. 6 and 7). For metastable
grades the yield stress curve followed during rolling is therefore heavily dependent on the rolling schedule
applied. By varying the reductions and rolling speeds used in a schedule, the metallurgy and temperature
of the rolled strip will be altered. Prediction of the yield stress will only be possible if the functional
dependence on reduction, rolling speed and interpass delay time can be determined.
The rolling models of Stone, Orowan and Ford are all extensions of a simple theory of slab compression*2-6'.
In this theory the horizontal stress s, and vertical stress t, acting on a simple slab of material are related to
the yield stress in plane strain compression k, by an equation
t= s+k
19
This is this yield criterion for plane strain compression for the material. The validity of this equation is in
doubt for stainless steels as it takes no account of the anisotropy evident in the results reported here. The
suitability of compressive yield stress for the calculation of rolling force values to mills which apply high
tensile forces is therefore in great doubt.
One could expect the theories to give reasonable agreement with roll separating force measurements
where relatively large diameter rolls are used and the roll bite conditions approximate to those of plane
strain compression. This fact is demonstrated by comparing the yield stress produced in compression with
that inferred from the roll separating force measurements made during experimental cold rolling trials.
In Fig. 13, the plotted yield stress ratios indicate that for rolling without tensions the models may predict
roll separating force to within an accuracy of better than 10%. The work roll torque, however, is highly
dependent upon the relative strip-roll velocity throughout the roll gap and is thus sensitive to the yield
stress under tension. The modifications required to rolling theory to account for material anisotropy can
only be determined from trials on an experimental facility which can provide both high compression and
tension simultaneously.
The dependence of yield stress on the mode of deformation, leads to the conclusion that the yield stress
value must be deduced from measurements of roll separating force or torque. The machinery to do so has
been put in place in the model and has been used to generate Fig 13. Such an approach is attractive as it
provides a means of achieving exact answers for repeatable phenomena, and is a l r e a d y a common
component of scheduling models for hot rolling mills. However, the results of this work on materials
testing and the measurements of torque on a production Sendzimir mill (to be discussed below), have led to
the conclusion t h a t such an approach would be inappropriate for Sendzimir cold rolling operations.
An inferred yield stress is a function of the processing and deformation conditions applied. One may
expect the inferred value to be a function of all the variables listed in Table 2. To predict yield stress
correctly, one would be required to build an empirical model describing its functional dependence on these
variables. The resulting algorithm would be identical in form and function to Equation (8) for the
empirical rolling torque model described in Section 2.2.2(a). The inferred yield stress would ensure valid
results only at the points where it has been calculated. Prediction of roll separating force and torque at
other points would require that the equations of the cold rolling model are valid and that there is no
residual dependence of the inferred yield stress on reduction and tension values. The results do not give
such confidence in the underlying equations of these theories. Therefore, the result of employing a system
based on an inferred yield stress, is an empirical algorithm which carries the computational overheads of
evaluating a cold rolling theory, which in turn may be erroneous in its predictions. The computational
overheads are of some concern, as the yield stress must be evaluated at several points in the roll bite for
each reduction proposed by the optimising routines. In addition computational time is required after each
pass for the calculation of the inferred yield stress value.
The attraction of a cold rolling theory is that it is able to provide predictions of both roll separating force
and torque. However, one must doubt the predictions of roll separating force based upon yield stress
implied by torque measurements, and similarly torque values predicted using yield stress inferred from
roll separating force measurements. A yield stress inferred from roll separating force will essentially
reflect the compressibility of the material under the prevailing conditions. A yield stress calculated from
torque will depend heavily not only on the conditions of the material but upon the friction coefficient in the
roll bite. The inferred yield stress is a single value and must be used at all points in the roll bite, hence the
rolling theories cannot be relied upon to predict accurately the position of the neutral point and the value
of torque.
It must therefore be concluded that for accurate and speedy calculations of roll separating force and torque,
empirical based models describing the behaviour of both of these variables are required.
4.
An optimisation programme can be successful only if it is given a full knowledge of the functional
dependence of rolling forces on reduction and tension. This dependence can only be e x a m i n e d if
measurements of rolling forces are obtained from a production Sendzimir mill. Throughout the course of
20
this project attempts have been made to measure the roll separating force and torque generated during
rolling. Roll separating force is by far the more accurate of the two parameters to measure; it is largely a
function of the material forces and has less dependence on frictional forces and applied tensions than
torque. Roll force measurement techniques have therefore been pursued with great effort, as detailed
below. However, this was unsuccessful and torque measurement techniques have subsequently been
applied with some success.
4.1
The Sendzimir cluster mill design does not lend itself to the direct measurement of roll separating force, as
the roll stack distributes the rolling force directly to the mill housing. The nature of the force distribution
is governed by the cluster geometry and the positions of the back-up roll eccentrics. The only method of
measurement available, which does not require significant modifications to the mill h o u s i n g or its
components, is to sense the displacement produced in the housing when roll separating force is applied.
A system has been sought which can detect this displacement and also satisfy other constraints imposed by
the mill operating conditions and maintenance requirements. These are:
(a)
(b)
(c)
(d)
The housing of a Sendzimir mill has a construction which gives it considerable rigidity, indeed the mill is
designed as a 'zero strain' housing. Figure 14 shows a sketch of the housing frame. The mill in question
has a frame of 1600 mm width. Each of the four housing legs has an approximate cross- sectional area of
385 000 mm 2 . If a roll separating force of 4 MN is generated at the roll gap, a simple calculation predicts
that a stress of 2.6 N/mm 2 will be produced in each leg. This stress will yield a strain of approximately
12 X 10"6. This small value is at the limit of the measuring capabilities of conventional foil strain gauges.
It was decided to investigate further the strain distribution produced within the complex mill geometry,
the aim being to locate an area of high strain concentration which would permit a more accurate roll force
measurement. A finite element model has therefore been devised using the commercial software package
ABAQUS0).
A three-dimensional model has been constructed which calculates the linear-elastic deformation produced
under static point loading conditions. The model considers only the mill frame and ignores the ancillary
equipment associated with the mill, including the roll stack. The model was constructed from the
drawings of Sendzimir Incorporated. The housing geometry was rationalised to allow easier coding and to
reduce the computation time required. The housing shape produced by these simplifications is shown in
Fig. 15. A symmetry in the mill design and a symmetry in the force distribution under load are assumed.
Therefore, the model considers only one quarter of the mill frame.
Roll separating force is transmitted to the mill frame via the roll stack. The model does not calculate the
roll separating force or the distribution of forces in the roll stack. These forces are calculated by a separate
program and then transferred to the mill model as input data. The loads are applied at eight concentrated
points across each back up-roll. These points correspond to the the locations where the As-U-Roll cams
connect with the saddle and transmit the force to the mill housing. In addition, point loads are applied
where the screw-down rams are bolted to the mill frame. The model is assumed to operate in the small
displacement regime. That is, under load the displacement is considered to be so small that the associated
change in the direction of the applied forces may be neglected.
In addition to the symmetry mentioned earlier, the Z-mill is also constrained by foundations which locate
the lower p a r t of the frame flanges. The locating m a t e r i a l is a s s u m e d to h a v e a stiffness of
10 6 N/mm 2 /mm.
21
The force distribution arising at the back-up roll saddle supports where they meet the housing was
calculated for a 4 MN roll separating force. The back-up eccentrics were assumed to be set in the fully
closed position. The effects of force distributions arising from misalignment of the strip in the roll gap and
strip crown effects are ignored and the rolling load is applied evenly across the As-U-Roll cams. Figure 16
shows the undeformed mesh of the housing. The rolled strip direction is along the coordinate axis 1.
Figure 17 shows, from the same viewpoint, the deformed mesh of the housing. To aid visual clarity the
displacement of the mesh has been greatly magnified. It can be seen that the mill pillars bend in a concave
manner. Further visual investigation, from different angles, shows the pillars bend in both the 1-3 and
2-3 planes. Torsion is present in the 1-3 plane. In Figs. 18 and 19 the vertical component e22 of strain is
shown. In Fig. 18, the view is from the front of the mill looking towards the roll stack and door jamb. In
Fig. 19 the view is from the mill centre looking outwards in the direction of the strip passage.
These results were used to locate the area of maximum strain. This occurs along the 2-axis (vertically)
near the inside corner of each vertical column and at pass line height and is of measurable magnitude for
foil and semi-conductor gauges. The model predicts that loading variations due to strip shape effects,
(non-uniform but symmetrical distribution of rolling contact pressure across the width of the strip) do not
have a significant effect on the stress-strain developed in the mill housing. It was not intended that the
model would provide exact numerical results, but was written primarily to provide a picture of the overall
stress-strain distribution in the housing and to indicate the region(s) of highest strain.
A survey of the mill revealed practical restrictions to mounting a strain sensor at the centre of the highest
strain area located by the ABAQUS model. The housing frame has attached to it many pieces of
equipment, including guide assemblies and supports for the oil spray system, which prevent access to the
mill frame at this point. In addition the need to locate the device in a safe place clear of any sources of
potential damage and without interference to the normal mill operation restricted the choice of location
further. The most suitable compromise position is at the external side of the mill next to the door jamb on
the hinged side and at pass line height. The door hinge design allows a clearance of 40 mm in this area,
with the door and door jamb providing a certain degree of physical protection.
4.1.1
It was uncertain whether a method could be developed which would measure strains in the order of several
microstrain in magnitude. A laboratory test rig was therefore constructed for the evaluation of potential
measurement techniques. A steel bar of length 300 mm, width 30 mm and thickness 10 mm was drilled at
its top and bottom ends with 10 mm diameter holes. The bar was then attached to a fixed point at its upper
end. A stack of calibrated weights was then attached to the lower end so as to place the bar under a known
and variable tensile stress. The induced strain could thus be altered in the range of approximately
0-10 microstrain, assuming the steel to have a Youngs modulus of 210 000 N/mm 2 .
Three potential strain measurement devices were evaluated: conventional foil strain gauges applied to
the bar so as to form a Wheatstone bridge circuit, a strain link device utilising semi-conductor strain
gauges, and a Wheatstone bridge circuit of semi-conductor strain gauges.
Each of the devices was attached in turn to a test bar. The circuits were powered by a ten volt power source
and the output sensed using an ultra-violet chart recorder equipped with variable amplification circuits.
Conventional foil backed strain gauges have a gauge amplification factor of approximately 2 and as such
may not be sensitive to such low strains when applied directly to the test bar. Testing confirmed that the
signal output at 10 microstrain was less than 0.1 mV and amplification of the signal demonstrated that
thermal and electronic noise dominated the gauge output.
A strain-link device was also constructed and tested, this is shown in Fig. 20. The device consists of two
halves which form a simple vice. These elements are attached to the surface under strain by means of
bolts. Dowel pins are introduced to prevent twisting about the fixings and to aid relocation should the
device have to be removed and replaced at any time.
22
The two portions of the vice concentrate the displacement produced between their location points into the
region between the vice jaws. This is only possible if the material placed in the vice carries little or no
stress. A block of perspex with a Youngs modulus of 3 GN/mm2 was used for this purpose.
A bridge circuit of semiconductor strain gauges, each with a gauge amplification factor of 140, was
mounted on the block so as to sense the strain concentrated by the link mechanism.
Testing of the device revealed that it was indeed sensitive to strains as low as 1 microstrain. However, the
results of the test also revealed that a bending moment was set up about the fixing bolts used to attach the
device. It became evident that the flexibility of the bolts was masking the true strain produced in the test
bar and that any design based upon this principle would also be effected by this problem, which is inherent
in the measurement of very low strain levels.
The direct application of the semi-conductor gauges to the test bar yielded better results. Two
semi-conductor gauges, each with a gauge amplification factor of 140, and two precision resistors were
applied to the test bar so as to form a Wheatstone bridge circuit. The response of the circuit when applying
strain in the range 1-10 microstrain, is shown in Fig. 21. Slight deviations from linearity were noted;
these arise as a result of small variations in the loading path, which in turn induce bending stresses in the
test bar. It was therefore decided to pursue direct application of semi-conductor strain gauges to the mill
as the preferred means of roll force indication.
4.1.2
The laboratory tests allowed the selection of semi- conductor gauges as an appropriate method of strain
sensing on the mill. The small size of the gauges (5 mm X 5 mm) provided some difficulty when mounting
on the mill frame and during the connection of the associated bridge circuitry. The gauges were placed in
the region located by the finite element model as discussed above. Four semi-conductor gauges, each with
a gauge amplification factor of 140, were applied such that two gauges were aligned vertically and two
horizontally.
The mill frame was firstly abraded so as to provide a smooth clean surface on which to affix the gauges.
Several layers of coating material were applied so as to provide protection from moisture, mechanical
damage and fluctuations in atmospheric temperature.
The gauge circuit was powered by a ten volt power supply and the output was recorded by an ultra-violet
chart recorder equipped with variable amplifier circuits. Recordings were made during the mill operation.
In Fig. 22 a typical response of the bridge circuit during rolling is sketched. At point one the mill is closed
and the housing is placed under strain by screwdown. The gauge output rises abruptly at this point,
indicating vertical stretching of the housing. At point two equal front and back tension is applied to the
strip and the signal falls. This result indicates that the strain induced by screwdown is relieved by the
tension forces. At point three rolling commences with the application of full tension and a further fall in
the signal is observed. Adjustments of As-U-Roll and intermediate roll settings can be seen to produce
small fluctuations in the signal output. As rolling continues the signal rises owing to the thermal
expansion of the mill frame and gauge material.
Studies of the output from the gauges for full rolling schedules indicated that the signal output was not a
simple function of the strip reduction. It became evident from observations that the magnitude of the
signal was heavily dependent on the stack geometry also.
To aid the interpretation of these results, static screwdown trials were conducted. For each trial a
different setting was chosen for the bottom back-up roll eccentric position. The position of the eccentric is
indicated on the mill by a dial gauge with a 0 (fully open) to 10 (fully closed) scale. The positions of the side
eccentrics were adjusted so as to allow the roll gap to be closed on a 1.5 mm thick strip. Screwdown
pressure was then applied by the top stack and the bridge output noted. The results of these trials are
shown in Fig. 23.
23
These results indicate t h a t the strain produced in the housing is highly sensitive to the distribution of
vertical and horizontal forces applied to the mill frame. The relation between roll separating force and
gauge output is therefore a complex function of the roll diameters and the back-up roll eccentric positions.
This complication, coupled with the considerable effects of strip tension and thermal drift on the housing
s t r a i n , led eventually to the decision to abandon any further a t t e m p t s a t roll s e p a r a t i n g force
measurement using strain gauges. However, if free access to a Sendzimir mill could be obtained, a system
of force indication based upon strain gauges could be applied to resolve these problems.
4.2
To achieve a measurement of coil properties for use in the optimisation model, it was decided that, as roll
force could not be easily sensed, a measurement of roll torque would be attempted. Drive shaft torque may
be measured by use of suitable strain gauges arranged in a standard Wheatstone bridge configuration.
Problems arise when attempting to record the bridge imbalance when the shaft is rotating. Slip rings may
be used for this purpose but have been found to be affected by electrical noise generated at the ring
interfaces. Radio telemetry was therefore chosen to transmit the bridge signal from the rotating shaft to a
recording unit. The units purchased consist of a radio frequency transmitter and receiver. The aerial is
wrapped around the drive shaft so as to ensure the signal is radiated uniformly from the shaft. A signal
processing unit is also placed on the shaft; this receives the gauge imbalance voltage and provides the
appropriate radio frequency signal to the aerial. The bridge signal level is modulated by the processing
unit; this enables the radio frequency signal (4 MHz) to be transmitted at constant amplitude so to ensure
that the device can operate throughout a full range of torque values. The pick-up unit is placed as close as
possible to the the aerial so as to limit the reception of stray electromagnetic fields. The pick-up signal is
carried by coaxial transmission cable to a demodulation unit. This unit provides a signal output in a
calibrated range suitable for recording by chart recorders or as an input to an analogue to digital
processor. Calibration of the signal is achieved by use of a calibration resistor which simulates the bridge
imbalance at a known level of torque. There is therefore no direct calibration of the torque signal and all
measurements will therefore be subject to an unknown systematic error.
Torque telemetry equipment has been installed on a commercial Sendzimir mill. Torque strain gauges
were applied to the main drive shaft of the mill. The gauges were located on the drive shaft between the
motors and the pinion gearbox unit. The relevant telemetry equipment was then connected. A ten volt
battery pack was used to power the bridge circuit and telemetry unit. This proved capable of sustaining a
signal for a period of up to fourteen days. A length of coaxial cable was required to carry the pick-up signal
to a convenient place where monitoring equipment could be housed. The outer sheath of the cable was
earthed so as prevent the generation of electromagnetic noise signals along its length.
The amplified signal output was fed to an ultra-violet chart recorder in the first instance. After an initial
evaluation period, this was replaced by a DASH-16 analogue to digital converter driven by a COMPAQ
386/20 personnel computer. In addition to the torque signal, the main drive rotational speed was also
recorded. An isolating transformer was placed in parallel with the output of the main drive shaft
tachometer. The transformer output was suitably amplified to allow monitoring to be performed by the
recording devices.
Access to the telemetry equipment has been limited as a result of production requirements and safety
considerations. Considerable problems and delay were encountered owing to the reliability of the
transmitter circuitry placed on the shafts. The problems were associated with mechanical damage to the
circuits created by the high rotational forces produced during mill operation. The mill in question
regularly rolls at speeds close to the maximum motor speed of 600 rpm. The problem was finally resolved
by setting the circuitry in a silicon rubber gel. This material holds the circuits in place without
interference to the circuit operation and provides protection from the ingress of oil and water.
Monitoring of the torque and speed signals was performed continuously throughout the lifetime of the
battery power. Fourteen days of continuous monitoring was achieved using the COMPAQ computer. The
data logging package ASYST was used to drive the analogue to digital converter card. The digital
measurements for torque and mill speed were processed to provide an output in units of full load torque
(i.e. motor torque at base speed) and revolutions per minute respectively. The data were stored in a
24
compact binary image on a 30 Mb hard disk. A logging frequency of 1 Hz was used allowing continuous
operation for up to 24 h. In practice the accumulated data was transferred to magnetic disk at 12 h periods.
The parameters describing the reduction sequence and the product rolled are automatically recorded by
the supervisory computer installed on the mill. Thus, the average through pass strip gauge, front tension,
back tension, strip speed and differential screwdown pressure were available from the mill computer
archives. The average of each parameter is based on readings made at regular rolled strip lengths rather
than at a set time interval. The time at pass start and end was also noted, so as to allow the alignment of
the monitored signals with the mill computer records.
5.
The data logged on the mill consists of large files containing two words of data for every one second of
recording. A method has been devised for separating these data into the portions representing individual
rolled passes. A histograming technique has been employed for this purpose, which allows for a
considerable compression of the recorded data. The main drive torque data in each file was recorded as a
histogram using 400 sampling bins in the range -1 to + 1 times full load torque. One h i s t o g r a m is
generated for a recording period of 360 s. A pass appears as a distinct peak in the histogram at the
appropriate torque value. Data obtained from a pass which spans two or more histograming periods can be
simply combined so as to give the full histogram distribution. The method has the advantage that the
distribution in torque values during a particular pass is immediately available. Similarly, torque
variations associated with sundry mill operations, such as strip threading and coil unloading, can be easily
detected and ignored. The torque values are signed according to the direction of rolling. This fact ensures
that consecutive passes cannot be confused. Similarly, the 360 s histogramming period ensures t h a t
passes rolled in the same direction will not appear in the same histogram. During the summation of the
histogram the bin entries are weighted by the mill speed signal. The mill speed signal is representative of
the strip speed, and thus provides a weight to the torque value which reflects the length of strip rolled
during the one second logging period. This method prevents the histogram being biased by torque values
obtained when the mill is rolling slowly for long periods. The histograms generated are stored in a file
such that only the non-zero bin entries are recorded. This method allows considerable compression of the
data and could be used to form the basis of an improved ASYST data logging routine.
In Fig. 24, the main drive torque and speed signal output are shown for a single pass. In Fig. 25, six
histograms are shown corresponding to each pass of a six pass schedule. It can be seen that for each pass
there is a considerable distribution in torque values. The distributions a r e symmetric a n d have a
characteristic width of 0.02 X full load torque. The time dependence of the distribution of torque values
within a pass, can be examined by a Fourier transform technique. In Fig. 26, the power spectrum for the
torque data shown in Fig. 24 is shown. The Fourier technique may give information a b o u t signal
frequencies only below the 1 Hz logging rate. The full frequency spectrum covers the range 0 to 1 Hz, the
spectrum is flat and of low amplitude in the range 0.01 to 1 Hz. There are two distinct peaks at 0 Hz and
0.01 Hz. The 0 Hz component represents random noise interference. The source of t h e peak at
approximately 0.01 Hz represents slow changes in the mean torque signal as displayed in Fig. 24.
The noise signal most probably arises from the detection of stray electromagnetic waves by the radio
telemetry receiver. The mill in question is driven by three motors via a common shaft. The magnetic field
of the motors may extend to the location of the telemetry circuitry generating stray electrical interference.
If this is the source of the noise signal, the construction of a simply Faraday cage around the transmitter
and receiver units may help to exclude the interference. Alternatively, the variation in the torque could
arise from changes in roll bite friction, tension and gauge fluctuations or slippage between rolls in the
stack. This uncertainty may be resolved by running the mill with the rolls removed from the housing. If
the variation is due to noise, one would expect to see a similar distribution of torque values centred at zero.
The data recorded was found to cover a wide variety of products with varying
diversity has limited the statistics available for any one product type. Work has
the analyses of 11 coils of Grade 302XD material, all of which were rolled to
Other coils of 302XD material rolled to a standard finish were also investigated
and confirmation of the results obtained.
25
In Fig. 27, the torque per unit width at the work roll is plotted as a function of pass input gauge. For
clarity only five coils of data have been displayed. The peak values of the histograms have been indicated,
the error bars represent the standard deviation of the weighted torque distribution. The schedules used for
these coils are summarised in Table 9.
It can be seen that there is considerable variation in the torque values, especially for the first pass, above
that which is associated with the measurement uncertainty. The torque varies erratically; a coil which
has the lowest torque value for one pass may well have the highest torque value for the next. However,
each of the coils in the sample was processed with slightly different conditions of reduction, tension, rolling
speed. The dependence of the torque on each of these variables must therefore be taken into consideration.
Any remaining torque variation, may then be ascribed to unknown conditions dependent upon the coil
properties and rolling conditions, such as yield stress and friction.
Although the collected data are sparse we have described the torque values using a function of the form
given by Equation (8). The procedure applied was as follows. The torque was plotted against the following
variables:
Total Fractional reduction =
Pass reduction
=
Front tension stress
=
Back tension stress
=
Starting gauge
Rolling speed
Screwdown cylinder differential
In Fig. 28, all the torque data are displayed against total fractional reduction. The torque is clearly a
polynomial function of total fractional reduction, with the shape of the polynomial being dependent upon
the starting gauge of the coil. A simple polynomial of the form given by Equation(8) was fitted to the data
and the residual error given by Equation (9) was calculated. The residual error was plotted against each of
the dependent variables in turn, so as to determine if any remaining dependence on these variables could
be distinguished. A quadratic dependence on fractional reduction was noted and a second fit was made.
This was found to successfully remove all dependence of the torque on these variables, except for first pass
data. As one may expect the first pass torque falls with the applied front tension stress in a linear manner,
as shown in Fig. 29. A linear relation to front tension stress was therefore included in the fitted equation.
The result of the fitting procedure is shown in Fig. 30, where plotted the 'fitted' torque has been plotted
against the measured values. The fit was found to be in error for some of the first pass results. This may
be expected, the fitting procedure attempted to describe all passes in terms of the front tension applied, a
variable which only discriminates the first pass torques. By fitting equations separately, to the first
passes and remaining passes, the results may be improved.
In Fig. 31 shows a histogram of the ratio of the residual error r for each pass to the average measurement
uncertainty o m for the sample. If the fitted equation represents the true relationship between torque and
the dependent variables, it would be expected that the residual errors would be drawn from the same error
distribution as the measurement errors. Thus, the ratio of the two errors at each point will form a normal
distribution. The number of entries in each of the histogram bins will be Poisson distributed and the error
bars indicated are V(number of bins entries). The statistics are too sparse to apply any conclusive tests
to this data. However, one may draw confidence in the fact that the distribution is not significantly
skewed, which is indicative of a poor fit. There is an indication that the equations, and hence the variables
fitted, account for all variation in the torque for passes two to six. The first pass torque may have some
small dependence on other parameters.
This residual dependence may arise from variations in the friction conditions in the roll bite. For the first
pass, heavy front tension is used which moves the neutral point towards the entry plane of the roll bite.
The measured torque at Pass 1 may thus be more dependent upon the relative strip-roll velocity and the
lubrication conditions in the bite than at subsequent passes where equal tensions are used. The coils in
question are ground prior to rolling and one could therefore postulate that there may be a variation in strip
2b
surface roughness not only from coil to coil but through the coil also and these effects would be most severe
at Pass 1. The torque variation may also be an effect of yield stress variations arising from fluctuations in
annealing conditions or the production of martensite as discussed in Section 3.1.
The data may therefore be adequately described in terms of the processing parameters. One could not
apply an adaptive procedure to these coils as it is evident that they are statistically equivalent. It must be
added that the sample size is small and further monitoring may result in a much wider distribution to
which the adaptive procedure may be relevant. Thus it is possible to consider for adaption those points
which satisfy the condition:
IL > 3
...(i2)
Using the routines described in Section 2.2.2, the yield stress has been calculated for each pass by
inverting Fords method. In Fig. 32 the calculated yield stress for the five coils shown in Fig. 27 is plotted
against reduction. The erratic behaviour of the predicted yield stress confirms the postulate of Section 3.4,
that the yield stress calculated in this manner is not a true measure of the material properties. The value
of the friction coefficient is fixed for all calculations, thus the yield stress is a function of the true value of
the friction coefficient. Hence, the sudden rise in the last pass yield stress shown in Fig. 32, may be
attributed to the change from standard to polished work rolls.
The yield stress values obtained suggest that little work hardening of the strip has occurred. This effect
can also be noted in Fig. 27 where, despite the higher initial pass reduction applied to coils with a 1.75 mm
starting gauge, the torque generated in the latter passes is no greater than that for coils with a 1.45 mm
starting gauge. This effect is most probably due to a corresponding rise in strip temperature with
reduction, which produces material softening and/or reduces the propensity for martensite formation.
To test the validity of the results, the operation of the data collection and torque fitting procedures in an
on-line environment has been simulated. The full data sample consisted of 65 passes representing 11 coils
rolled from starting gauges of 1.45-1.50 mm and 1.75 mm. The first 40 passes were used to fit the torque
equation discussed above, there being 36 free parameters to be determined. The fitted data consisted of
only one 1.75 mm coil and thus the fitted coefficients were biased towards the 1.45-1.50 mm coil data. The
fitted equation was used to predict the torque for the remaining 15 passes in the sample; the residual
errors are plotted against torque in Fig. 33(a). It can be seen that a good prediction is made for the
1.45 mm coils, whereas the 1.75 mm predictions are significantly poorer. The process was repeated using
50 and 60 points in the sample to obtain the fitted coefficients respectively. The residual errors for the
remaining passes are shown in Fig. 33(b) and 33(c) respectively. The data in Fig. 33(c) corresponds to a
1.75 mm coil and represents an accurate fit to the data. Thus, the system is capable of describing any
possible reduction sequence once a statistically significant sample has been devised. It is
important to note that the statistical fitting procedures aim to minimise the total residual error and hence
will be biased towards satisfying the schedule for which most data have been recorded. With a greater
data sample it would be prudent to generate separate equations for groups of product types.
For example, Fig. 34 illustrates a plot of the torque against input gauge for thicker 302XD material and
the thinner bright annealed coils discussed above. It can be seen that the thicker material torque
reduction curve is slightly different from the thinner coils analysed above. This may be expected, as more
energy of deformation must be provided for thicker material to raise the material temperature and hence
softening due to thermal effects may occur later in the schedule.
6.
The data collected demonstrates that one can predict rolling torque to within the accuracy associated with
the through pass signal variation. Dynamic monitoring of input and output gauge, and front and back
strip tension will allow more insight to be gained into the through coil variation of torque. Reduction of
the torque measurement error will allow the dependence of torque on coil properties and frictional effects
to be more easily resolved.
27
The extent to which the spread of torque for a single pass affects the scheduling model must also be
considered. The predicted torque values will, by the nature of the statistical methods used, represent the
mean of the mean through pass torque. For scheduling, one is interested not only in the range of torque
obtained, but the length of coil for which the torque values are maintained. The standard deviation of the
weighted histogram distributions obtained for each pass may be taken as a measure of the through pass
torque variation. It may be expected that the width of the torque distributions may vary as a function of
the parameters which control the pass. Therefore, an equation of the form used for torque predictions
(Equation (8)), may be used to predict the variation in torque at any point in the space of the dependent
variables. The predicted torque at any point is then simply the sum of the predicted mean torque value
plus twice the standard deviation. Provided the spread in torque values follows a normal distribution, 98%
of a coil chosen at random, will have torque values below the predicted figure. The predicted torque will
thus give a reasonable measure with which to compare the merits of different schedules.
The fitted torque curve is only valid for the range of data from which it was constructed, one cannot expect
to extrapolate too far beyond the bounds of the data range before serious discrepancies occur. This effect
was demonstrated above, where data for 1.45 mm starting gauge coils was used to predict values for
1.75 mm starting gauge coils. Care must also be taken during the fitting procedure to ensure that the
collected data is evenly distributed over the fitting range and that the number of degrees of freedom is not
minimal.
It can be seen that optimised schedules can
must be constrained close to the ranges for
naturally generate schedules which push at
expansion of the valid data range will occur,
the optimising procedure.
The gradual movement towards optimised schedules also allows the assessment of the impact of the
schedule on surface finish and strip shape to be determined. New operational limits on mill camber will
have to determined as reductions are increased. Lubrication conditions and roll preparation may also
have to be revised as draft and speed are maximised.
7.
CONCLUSION
A software algorithm has been successfully constructed which will provide speedy calculations of rolling
schedules. To operate, the optimisation routines require a knowledge of the torque which must be
generated at the roll gap and the operational limits which must be observed for quality strip manufacture.
Rigorous testing of the software has demonstrated that the algorithm developed may be operated in an
on-line environment. The software has been constructed such that the calculated schedules may be
adapted in response to measurements of rolling torque. However, this procedure is possible only if the
through pass variation in torque is significantly smaller than the torque variation noted between
individual coils.
It has been the aim of the project to demonstrate that cold rolling theories coupled to a yield strength
database could provide accurate predictions for roll separating force and torque. However, the work has
shown that this theoretical approach is inappropriate. A significant difference between the yield stress of
stainless steel in compression and tension has been revealed. This fact invalidates the basic equations on
which cold rolling theories have been developed. In particular one could not use yield stress data inferred
from m e a s u r e m e n t s of roll s e p a r a t i n g force to predict work roll torque. Therefore, under these
circumstances, both roll separating force and torque predictions must be made separately using empirical
techniques.
The experimental work performed h a s also shown t h a t , for certain m e t a s t a b l e grades, the rolling
temperature can have a significant effect upon the yield stress of austenitic stainless steel. This variation
has been found to be considerably greater than that which may be generated during the industrial
annealing of this product. The yield stress at the time of rolling is therefore a function of the deformation
history, which may be described in terms of reduction, tension, rolling speed and interpass delay time. To
28
account for this complex behaviour, the empirical models of torque and roll separating force have been
developed to account for dependency on these and any other parameters which are found to discriminate
the measured values.
During the development of the optimisation model it has become apparent that a knowledge of rolling
torque is fundamental to the operation of the scheduling system. Roll separating force only enters into the
model when one considers the constraints required for adequate shape control. These constraints are in
practice determined experimentally, and may be expressed in terms of the reductions and tensions applied.
Hence, a knowledge of roll separating force is not essential to the operation of the scheduling system. Roll
separating force is, however, less dependent on frictional forces and applied tensions than the torque and is
thus preferable for use as a measurement with which to adapt calculated schedules. Attempts have been
made to measure the roll separating force produced by Sendzimir mills through the strain induced in the
mill housing. The finite element technique has been employed to investigate the deflection of the
Sendzimir mill under load and strain measurement techniques based upon high sensitivity semi-conductor
strain gauges have been developed and tested. The complexity of the cluster mill design has been found to
result in housing strains which are dependent on the roll separating force, the roll stack geometry and
applied strip tension. A solution to this problem is feasible, but as discussed, this is not essential for the
operation of the scheduling system. Torque telemetry techniques have therefore been employed and
measurements have been made on a production Sendzimir mill. The data collected has allowed the
construction of an empirical model describing the dependence of torque on the pass descriptor variables.
The ability of the proposed system to provide accurate torque predictions, given that a statistically
significant sample base has been collected, has been demonstrated. The ability to adapt rolling schedules
could not be demonstrated, as the sample base collected was small and the through pass torque variation
was found to be significant. Further work will be required to investigate the through pass torque variation
and to eliminate external signal noise.
The dependence of finished strip surface quality on the rolling schedule necessitates that optimal
schedules may only be implemented as part of a carefully planned exploration of this phenomenon. The
'expert system' approach devised allows all such practicalities to be built into the scheduling system as
necessary.
The production time which will be saved by the optimised scheduling system cannot be postulated.
However, the methods employed have been shown to ensure that the true time optimum schedule will be
generated and that product quality will be assured.
To conclude, although the aims of the project have been met, it is clear that understanding of the
fundamental behaviour of stainless steel during rolling on a Sendzimir mill is limited. Further research
into cold rolling theory, surface quality and roll lubrication is required, and it is hoped that this work will
provide a basis for further studies in this field.
REFERENCES
1.
M.J. Box, D. Davies and W.H. Swann: 'Non-Linear Optimisation Techniques' I.C.I
Monograph Series No. 5., Oliver and Boyd, 1969.
2.
M.D Stone: 'Rolling of Thin Strip' Iron and Steel Engineer, February 1953, pp61-74 and Iron
and Steel Engineer, December 1956, pp55-76.
3.
E. Orowan: 'The Calculation of Roll Pressure in Hot and Cold Flat Rolling', Proc. Inst. Mech.
Engrs, 150, pl40, (1943).
4.
D.R. Bland and H. Ford: 'The Calculation of Roll Force and Torque in Cold Strip Rolling with
Tensions', Proc. Inst. Mech Engrs., 159, pl44, (1948).
5.
H. Ford, F. Ellis and D.R. Bland: 'Cold Rolling with Strip Tension', Part 1, J.I.S.I., May 1951,
Parts 2 and 3, July 1952.
29
6.
7.
T. Angel: 'Formation of Martensite in Austenitic Stainless Steel', Journal of The Iron and
Steel Institute, pl65-174, May 1954.
8.
I. Williams, R.G Williams and R.C. Capellaro: 'Stability of Austenitic Steels between 44-373
K', Sixth International Cryogenic Engineering Conference, p337-341.
9.
Advanced Mechanics Of Materials Hugh Ford, Longmans, Green and Co. Ltd., 1963
10.
30
TABLE 1
DATA USED IN OPTIMISATION PROGRAM OPTI
Mill Data
Coefficient of Friction (MU)
Roll Flattening Constant
Diameter of Drive Roll
Diameter of Work roll
Mill Motor Power
Mill Driving Efficiency
Mill Motor Base Speed
Maximum Motor Speed
Mill Motor Gearbox Reduction
Coiler Motor Power
Coiler Drive Efficiency
Coiler Motor Base Speed
Coiler Motor Gearbox Reduction
Maximum Rolling Load
Exit Strip Speed on First Pass
Maximum Exit Strip Speed
Maximum Back Tension on First Pass
Maximum Back Tension of Penult Pass
Maximum Front Tension Penult Pass
Maximum Back Tension of Last Pass
Maximum Front Tension Last Pass
Maximum Rated Tension
Rate of Acceleration of Strip
31
=:
=
=
=
=
=
=
=
=
=
=
=
=
=
=
=
=;
=
=
=
=
=
=
0.060
22.00 mm 2 /MN
173.00 mm
52.00 mm
1062.0 kW
84.0%
400.0 rev/min
800.0 rev/min
2.380
930.0 kW
95.0%
380.0 rev/min
9.942
5.24 MN/m width
46.0 m/min
137.0 m/min
44.0 kN
312.2 kN
312.2 kN
312.2 kN
306.0 kN
378.0 kN
340.6 m/min/min
TABLE 2
COLD ROLLING MODEL INPUT PARAMETERS
Coil Data
Grade
Processing Route Code
Surface Finish Type
Starting Gauge
Pass Parameters
Input Strip Gauge
Output Strip Gauge
Front Tension
Back Tension
Rolling Speed
Roll Radius
Friction Coefficient
Roll Flattening Constant
Deformation History (all Rolled Passes)
Input Strip Gauge
Output Strip Gauge
Rolling Speed
Roll Radius
Front Tension
Back Tension
Inter-Pass Delay Time
Rolling Model Selection Flag
32
TABLE 3
OPTIMISED SCHEDULE RESULTS AND COMPUTATION TIMES
Gauge, mm
.j
'-^1
Computation Time
No. of Passes
Reduction
%
Width
mm
Coil
Weight
kg
Optimised
Present
VAX CPUS
P.C. (s)
Input
Output
5.02
1.98
60.6
1275
26 600
14.2
85
4.85
1.98
59.1
1275
9 360
14.8
68
4.82
1.98
58.9
1306
26 260
6.3
72
4.66
1.98
57.5
1299
26 360
25
51
4.00
1.48
63.0
1542
26 400
28.7
85
4.00
1.48
63.0
1543
26 300
32.9
124
3.93
1.58
59.8
1056
24170
6.5
21
3.90
1.58
59.5
1029
24 280
6.3
25
2.98
1.17
60.7
1272
26180
12.7
32
2.95
1.19
59.6
1027
6 960
6.6
24
TABLE 4a
PERCENTAGE CHEMICAL ANALYSIS OF
TYPE 302XD HOT BAND MATERIAL
C
Si
Mn
Cr
Mo
Ni
Cu
Nb
0.05
0.40
1.04
0.002
18.1
0.30
8.20
0.28
0.04
TABLE 4b
PERCENTAGE CHEMICAL ANALYSIS OF
TYPE 302AA HOT BAND MATERIAL
C
Si
Mn
Cr
Mo
Ni
Cu
Nb
Ti
0.047
0.40
1.31
0.024
< 0.002
18.2
0.14
8.58
0.35
<0.02
<0.02
0.05
34
TABLE 5
MECHANICAL PROPERTIES OF SOFTENED AND DESCALED
302AA GRADE STAINLESS STEEL HOT BAND
Coil
No.
0.2%
Proof Stress
N/mm2
UTS
N/mm2
Hardness
HV
Gauge
mm
Test
Location
292
302
612
611
155
155
4.19
4.35
Front
Back
301
289
654
648
174
170
4.76
4.76
Front
Back
297
313
608
625
166
183
5.00
5.00
Front
Back
286
292
640
637
168
166
4.76
4.76
Front
Back
290
286
636
633
172
167
3.00
3.00
Front
Back
281
279
579
584
153
152
3.21
3.17
Front
Back
266
255
594
584
158
152
4.00
4.00
Front
Back
301
304
659
661
169
169
5.00
5.00
Front
Back
289
313
637
644
170
178
Front
Back
10
299
285
631
621
173
164
6.00
6.00
Front
Back
11
297
288
589
593
173
165
6.00
6.00
Front
Back
12
282
298
625
633
167
172
6.00
6.00
Front
Back
13
293
283
639
633
177
177
6.00
6.00
Front
Back
14
308
297
620
627
176
176
6.00
6.00
Front
Back
15
293
304
624
613
176
176
6.00
6.00
Front
Back
291.1
12.3
313
255
58
623.1
22.2
661
579
82
168.3
8.3
183
152
31
Mean
Standard Deviation
Maximum
Minimum
Range
35
TABLE
DETAILS OF COLD ROLLING EXPERIMENTS, STARTING HARDNESS 139141 VPN
Specimen
Number
Initial
Thickness
mm
Thick
nes
mm
%
Reduc
tion
Load
t
MW/12/1
5.055
4.570
959
10.8
MW/12/2
5.171
4.624
10.58
122
MW/12/3
5.164
4.625
10.44
12.2
MW/12/4
5.048
4.578
931
10 5
MW/12/5
4.950
3.373
31.86
30.1
MW/12/6
5.158
4.612
10.59
MW/12/7
5.070
4.582
MW/12/8
5.089
4.592
MW/12/9
5 149
MW/12/10
Thick
eu
mm
Overall
%
Reduc
tion
Load
t
Thick
ness
mm
Overall
%
Keduc
tion
Load
t
Thick
ness
mm
Overall
*
Reduc
tion
Load
t
4085
1901
17.3
11.7
4.115
20.22
18.0
9.63
10.9
4.105
19.03
17.6
9.77
11.4
4.108
19.28
18.3
3.531
30.62
4.610
10.47
12.0
4.123
19.93
18.6
3.541
31.23
26.0
5.151
4.608
1054
11.8
4.112
20.17
183
3.537
31.33
25.7
MW/12/11
5.111
4.596
10.08
11.6
4.119
19.41
18.4
3.547
30.60
25.9
3.015
41.01
30.0
MW/12/12
5 169
4.617
10.68
12.2
4.131
20.08
18.7
3.550
31 32
263
3.U18
41.61
30.0
4.130
20.44
18.8
3546
31.09
26.5
3.010
42.02
29.8
Thick
ness
mm
Overall
*
Reduc
tion
Pass 7
Pass 6
Pass 5
Pass 4
Pass 3
Pass 2
Puss 1
Load
t
Thick
ness
mm
Overall
%
Reduc
tion
Load
t
MW/12/13
5.191
4.632
10.77
12.5
MW/12/14
5.160
4.620
10.47
12.3
4.125
20.06
18.8
3.554
31.12
20.1
3.007
41.72
29.6
2.656
50.47
31.7
MW/12/15
5.169
4.618
10.66
12.3
4.137
19.97
18.7
3 555
31 22
26.1
3.041
41.17
29.5
2.551
50.65
31.5
MW/12/16
5.172
4.620
10.67
12.4
4.126
20.22
18.7
3540
31.43
26.3
3.009
41.82
29.8
2.548
50.73
31.5
MW/12/17
5.147
4.608
10.47
11.9
4.122
19.91
18.5
3.538
31.20
26.3
3.009
41.54
29.5
2.547
50.51
31.6
2/43
60.31
9.40
10.5
4 098
18.71
17.4
3.521
30 15
25.1
2.990
40 69
28.9
2.530
4981
30.6
2.034
59.47
37.3
51.11
30.4
2.051
60.28
35.5
MW/12/18
5.041
4.567
Thick
ness
mm
Overall
%
Reduc
tion
37.9
MW/12/19
5.163
4.610
10.71
11.9
4.137
19.87
18.1
3.540
31.43
25.3
2.994
42.01
28.5
2.524
MW/12/20
5.045
4.568
9.45
10.6
4.090
18.93
17.6
3.515
30.33
25.1
2.987
40.79
29.0
2.538
49.69
30.9
2.064
59.09
36.6
1.565
68.98
MW/12/21
5.089
4.591
9.79
115
4 108
19.28
182
3.533
3058
25.7
3.005
40.95
29.6
2.542
50.05
31.5
2,058
59.56
37.4
1.593
68.70
4.117
19.90
18.7
3.542
31.09
26.1
3.017
41.30
30.0
2.557
50.25
32.0
2.070
59.73
38.4
1.555
69.75
MW/12/22
5.140
4.615
10.21
12.1
M W/12/23
4 916
3.275
33.38
29.9
M W/12/24
5.173
3.322
35.78
32.9
MW/12/25
5.050
4.087
19.07
18.8
MW/12/26
5.162
4.103
20.52
21 0
MW/2
5.078
3.022
40.50
MW/3
5.107
3.032
28.88
MW/4
5.122
3.637
28.99
29.4
MW/6
5.012
3.531
29.55
28.1
29.1
Load
t
TABLE 7
DETAILS OF COLD ROLLING EXPERIMENTS, STARTING HARDNESS 145-162 VPN
Specimen
Number
Initial
Thickness
mm
Thickness
mm
Reduction
Load
t
Thick
ness
mm
Overall
Reduction
Load
t
Thickness
mm
Overall
Reduction
Load
t
Thick
ness
mm
Overall
Reduction
Load
t
Thickness
mm
Overall
Reduction
Load
t
5.132
4.574
10.87
5.153
4.613
10.48
13.9
MW/13/3
4.874
3.353
31.21
33.4
MW/13/4
5.136
4.579
10.85
13.0
4.534
9.41
11.2
4.094
18.20
17.1
3.531
29.45
25.2
3.022
39.62
28.2
2.504
49.97
32.7
10.47
12.7
4.110
19.74
18.2
3.549
30.70
25.9
3.025
40.93
28.4
2.514
50.91
33.2
5.005
MW/13/6
5.121
4.585
MW/13/7
5.098
4.568
10.40
12.7
4.102
19.54
18.3
MW/13/8
5.143
4.591
10.73
13.3
4.119
19.91
18.7
MW/13/9
5.115
4.587
10.32
13.1
4.123
19.39
18.7
MW/13/10
5.155
4.590
10.96
13.4
4.122
20.04
18.7
3.554
31.08
26.4
10.18
13.5
4.132
19.25
19.0
3.565
30.33
27.1
31.07
26.5
30.33
26.0
3.033
40.24
28.6
2.519
50.36
33.7
MW/13/11
5.117
4.596
Thickness
mm
Overall
Reduction
Load
t
Thickness
mm
Overall
Reduction
Load
t
12.6
MW/13/1
MW/13/2
MW/13/5
Pass 7
Pass6
Pass 5
Pass 4
Pass 3
Pass 2
Passi
MW/13/12
5.166
4.599
10.98
13.4
4.124
20.17
18.8
3.561
MW/13/13
5.075
4.564
10.07
12.5
4.104
19.13
18.2
3.536
MW/13/14
5.126
4.583
10.59
13.0
4.110
19.82
18.5
3.544
30.86
26.2
3.047
40,56
29.0
2.521
50.82
33.0
2.017
60.65
38.0
MW/13/15
5.014
4.534
9.57
11.1
4.071
18.81
17.0
3.508
30.04
24.8
3.004
40.09
27.7
2.489
50.36
32.0
1.992
60.27
36.5
13.4
4.124
20.17
18.7
3.553
31.22
26.4
3.032
41.31
29.0
MW/13/16
5.166
4.599
10.98
MW/13/17
4.982
3.596
27.82
27.9
MW/13/18
4.977
3.604
27.59
28.0
MW/13/19
5.149
4.591
10.84
13.4
4.124
19.91
18.9
3.557
30.92
25.7
3.043
40.90
29.3
2.522
51.02
33.2
2.017
60.83
37.6
4.536
9.35
11.1
4.076
18.54
17.2
3.524
29.52
25.0
3.021
38.51
28.1
2.497
50.10
32.5
1.996
60.11
37.0
1.517
69.68
30.81
26.3
3.044
41.65
29.4
2.524
50.76
33.4
2.017
60.67
37.9
1.543
69.91
46.3
2.513
50.85
33.7
2.009
60.71
37.8
1.534
70.00
45.9
MW/13/20
MW/13/21
5.004
5.128
4.582
10.65
13.0
4.111
19.83
18.6
3.548
MW/13/22
5.106
4.590
10.11
13.1
4.131
19.10
19.0
3.562
30.24
26.9
3.055
40.48
29.9
M W/13/23
5.113
4.589
10.25
13.4
4.122
19.31
19.0
3.553
30.51
26.9
3.050
40.35
29.3
MW/13/24
5.071
4.560
10.08
12.0
4.091
19.33
18.0
3.533
30.33
25.6
3.015
40.54
28.5
4.093
18.93
17.5
3.515
30.38
25.1
3.013
40.32
28.0
MW/13/25
5.049
4.553
9.82
11.6
M W/13/26
5.083
4.103
19.28
20.8
MW/8
4.991
MVV/10
5.098
3.386
33.58
35.2
TABLE 8
DETAILS OF COLD ROLLING EXPERIMENTS, STARTING HARDNESS 160-182 VPN
Specimen
Number
Initial
Thickness
mm
Thick
ness
mm
Reduc
tion
MW/14/1
4 960
3.296
33 55
MW/14/2
5.187
4.634
10.66
Load
t
Thick
ness
mm
Overall
Reduction
Load
t
Thick
0058
Overall
mm
Reduc
tion
Load
t
Thick
ness
mm
Overall
<*
Reduction
Load
t
Thick
ness
mm
Reduc
tion
28.44
25.0
2.923
43.64
36.7
2.553
50.78
29.7
2.155
58.45
33.1
1.611
68.95
49.4
3.672
28.28
25.6
2.956
42.25
34 5
2.557
50.05
30.5
2.127
58.45
34.7
1.590
68.94
48.9
2.953
41.69
32.2
2.551
50.63
303
2.082
59.71
35.5
1.548
70.04
47.1
2564
50.47
30.8
2.091
59.70
35.8
35.6
35.9
4.591
11.59
14.7
5.186
4.595
11.39
14.7
MW/14/6
5 104
4 565
1056
134
133
4.105
19.29
18.5
12.5
4.093
18.68
17.9
MW/14/7
5086
4.559
10.36
MW'14/8
5.033
4.534
9.91
MW'14/9
5.157
4.579
11.21
14.1
4.118
20.15
190
MW/14/10
5.040
4.540
9.92
12.6
4.090
18.85
18.2
3.583
28.91
25.7
MW/14/11
5.155
4.590
10.96
14.6
4.134
19.81
196
3.551
31.11
27.9
MW/14/12
5064
4 504
9.87
134
4 118
1808
18 n
3.549
29 92
27 0
11.23
143
4.129
20.09
19.1
3.551
31.28
27.7
2.984
41.68
30.1
MW/14/14
5.117
4.568
10.73
13.5
4.108
1972
18.5
3.535
30.92
26.8
MW/14/15
5.177
4588
11.38
14 4
4 127
20.28
19.2
3554
31.35
275
2,980
42.32
30.4
MW/14/16
5.151
4.580
11.08
14.3
4.127
10.88
19.3
3.549
31.10
27.7
2.995
41.86
30.8
MW/14/17
4.916
3.617
26.42
27 9
MW/14/18
5.075
4.549
10.36
13.1
4.099
19.23
18.2
3.525
30.54
26.6
2.974
41.40
29.8
5.190
4.594
11.48
14.7
4.129
20.44
19.4
3.547
31.66
27.6
2.982
42.54
30.4
2.558
50.71
31.0
MW/14/20
5050
4.533
10.24
124
4.086
19.09
186
3.514
30.42
26 1
2.958
41.43
29.4
2.543
49.64
30.1
MW/14/21
4.982
3.001
27.22
28.2
MW/14/22
5.189
4.593
11.48
14.5
4.137
20.27
19.2
3.544
31.70
27.5
2.U85
42.47
30.7
2.560
50.55
30.S
12.4
4.095
18.94
17.8
3.517
30.38
26.4
2.906
41.29
29.7
2.541
49.70
30.2
2.081
58.81
10.95
13.8
4.128
19.58
18.8
3.541
31.02
27.2
2.984
41.87
30.5
2.563
50.06
30.9
2.094
59.21
18.78
21.4
4.138
20.22
22.3
5.127
3.366
34.34
35.0
5.080
3.386
33.35
34.5
MW/14/24
5.133
4.571
MW/14/25
5.061
4.119
MW/14/26
5.187
MW/9
MW/ll
Load
t
3.712
5.193
10.06
Reduction
19.6
MW/14/5
4.544
Overall
20.2
MW/14/4
5.052
Thick
ncss
mm
18.75
9.45
MW/14/23
Load
t
19.99
4.635
MW/14/19
Overall
4.159
5.119
4.587
Thick
ness
mm
4.150
MW/14/3
5.167
Reduc
tion
Load
t
31.5
12.5
MW/14/13
Overall
Pass 7
Pass 6
Pass 5
Pass 4
Pass 3
Pass 2
Passi
TABLE 9
302XD BRIGHT ANNEALED ROLLING SCHEDULES
Pass
%
Reduction
Front Tension
(kN)
21-24
350 - 400
100
13-16
400 - 450
400 - 450
13-14
350-400
350 - 400
12-14
300-330
300-330
10-14
270 - 300
270 - 300
10
250 - 280
250-280
39
Back Tension
(kN)
OPTIMISATION ROUTINE
Rolling Time
Pass/Fail Flag
Reduction and
Tensions
Torque/Load/Reduction/Speed
Pass/Fail Flag
Operational Constraints
Database
Cold Rolling
Theory
Yield
Stress
Database
Empirical
Model
Torque/
Load
Database
41
0/100%
75%
25%
50%
FIG. 2
TOTAL STRIP GAUGE REDUCTION DISPLAYED AS A
FUNCTION OF THE INDIVIDUAL PASS REDUCTIONS
42
(R4/1144)
Constrained Region
FIG. 3
43
(R4/1145)
1800
1600
1400
1200
>v
tn
1000
SOO
600
400
Pre-rolled Samples
eoo
0.0
0.2
0.4
0.6
O.B
1.0
1.2
1.6
NRTURAL STRAIN
44
1.8
1800 +
1600.
HOO.
1200
, >ooo
(
lu
oc
to
800!
600
Pre-rolled Samples
400
200
0.0
0.2
0.4
0.6
0.8
1.0
1.2
1.4
1.6
NATURAL STRAIN
45
1.8
24.5C
32C
1.3 .
1.2
1.0
0.9
0.8
0.7
0.6
0.3
0.4
0.2
0.1
0.0
0.00
0.03
0.10
0.15
0.20
0.23
0.30
0.33
0.40
0.45
0.50
0.55
Natural Strain
46
0.60
True S t r e s s (kN/mm2)
1.4
31C
1.3
!7.5C
1-2
1.1
i-o
0.9
0.8
0.7
0.6
0.5
0.3
o.i
0.0
0.00
0.05
0.10
0.15
0.20
0.25
0.30
0.35
0.40
0.45
0.30
0.35
0.60
Natural Strain
47
% Martensite (max)
60,
Transverse
+ Longitudinal
No Transformation
48
H a r d n e s s , VPN
260
250
240
230
220
2)0
200-
190
180
170
160
X
X
150
X
X
X
140 -
130,
2
FIG. 9
-L
4
Time i n f u r n a c e
-L
5
at
J-
6
1 100C ( m i n )
J-
(R4/2)
Cumulative roll
separating force, t
220 i -
200
145-162 V?
160
139-141 VPN
100
30
40
Total % reduction
FIG. 10
50
in h e i g h t
50
80
(R4/6)
N/mm2
1600
1400 -
Starting Hardness
160-182 VPN
139-141 VPN
145-162 VPN
1200
1000
800
600
400
200
20
30
% Reduction
40
50
60
70
(R4/111)
1900
1800
1700
1600
1500
1400
1300
1200
100
1000
900
800
700
600
-L
J_
J_
70
60
50
40
Reduction %
YIELD STRESS OF GRADE 302AA STAINLESS STEEL
OBTAINED BY COMPRESSION TESTING
_L
20
500
FIG. 12
-L
30
52
J
80
(R4/1146)
Compressive
Ratio =
Tensile
Yield Stresses
1.6
Measured Data from Compression
and Tensile Tests
Fords Results for Copper
Calculated Ratio (Fords Rolling
Theory)
1.5
1.4
1.3
1 .2
1. 1
1.0
10
FIG. 13
20
-L
JL
30
40
% Reduction
50
60
53
_J
70
(R4/1147)
'.
i-
(R3/297)
'Vi
FIG. 15
(R3/9298)
'-/
FIG. 16
UNDEFORMED MKSH
MAG.
FACTOR
=+3.9E+03
FIG. 17
E22
VALUE
CO
FIG. 18
VERTICAL STRAIN COMPONENT (E22) AS VIEWED FROM THE FRONT OF THE MILL
i
E22
VALUE
3 . S 9 E - 0 S
2 . 14E-05
6 . 92E-06
+7.58E-06
+2 . 20E-0S
+3. 65E-05
+ 5 . 1 IE-OS
+6 . 56E-0S
+ 8 . OIE-OS
+ 9 . 46E-0S
+ 1 . 09E-04
(V,
VO
-+1.23E-04
FIG. 19
Threaded
shaft for
compressing block
Vice element
Gauges mounted
on block
Vice element
FIG. 20
(R3/9299)
61
(R4/109)
5r
10
FIG. 22
12
14
16 18
Time (s)
20
(R4/1148)
Gauge Output, mV
9
X
O
Eccentric Position
Eccentric Position
Eccentric Position
Eccentric Position
Q |_
X
X
X
X
1
2
3
4
5
Transducer Signal (Screwdown Pressure)
(Arbitrary Units)
FIG. 23
63
(R4/1149)
Speed (rpm/100)
Torque (Base xlO)
4 i-
Speed
Torque
300
400
700
Time (s)
FIG. 24
64
(R4/1150)
Number of Events
180 160 140 120
100
80
60
40
20
-0.25 -0.2
X
X
-0.15 -0.1 -0.05
0
0.05
0.1
0.15
Fraction of Full Load Torque
FIG. 25
65
0.2
J
0.25
(R4/1151)
Signal Amplitude
( Full Load Torque)
0. 11 r
0. 1
FIG. 26
0.2
0.3
Frequency (Hz)
66
0.4
0.5
(R4/1152)
.6
0.8
1.2
1.4
FIG. 27
WORK ROLL TORQUE PER UNIT
WIDTH PASS INPUT GAUGE
67
(R4/1153)
I
I
I
'i
fl'
it
I
J
0.25
I
0.3
1
L
X
0.35
0.4
0.45
0.5
0.55
Total Fractional Reduction
FIG. 28
WORK ROLL TORQUE PER UNIT
WIDTH TOTAL FRACTIONAL REDUCTION
68
0.6
0.65
(R4/1154)
5 r
4 -
I
I
150
200
I
X
I
I
I
I
250
300
350
FIG. 29
WORK ROLL TORQUE PER UNIT
WIDTH FRONT TENSION STRESS FOR 1ST PASSES
69
(R4/1155)
I I
J
L
J
L
1
2
3
4
5
6
'Fitted' Work Roll Torque per Unit Width (kN m/mm)
FIG. 30
70
(R4/1156)
Number of
Passes
20
15 -
X
X
X
X
X
- 2 - 1
0
1
2
Residual Error/Mean Measurement Error
FIG. 31
71
(R4/1157)
Stress N/mm2
1600
1400
200
1000
800
600 -
400 -
200
0.2
0.3
_l_
0.4
0.5
0.6
0.7
Reduction
FIG. 32
72
(R4/1158)
I
I
I I
I I1 I
I
I
1.75 mm
Coil Data
. .I
1.75 mm
Coil Data
I
6
X
-L
- 4 - 2
0
2
4
Residual Error/Mean Measurement Error
-6
6 -
1.75 mm C o i l
Data
II
I
I
X
-10
Measured Work Roll
Torque per Unit Width
(kN m/mm)
JL
I
X
I
II
I I
-8
-6
- 4 - 2
0
2
Residual Error/Mean Measurement Error
(c) Fit to 60 Passes Residual Error for
Remaining 5 Passes in Sample
4.5 r
4
3.5
3
I .75 mm
Coil Data
2.5
-0.6
X
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
Residual Error/Mean Measurement Error
FIG. 33a-c
73
.2
1.4
12
10
0.5
1.5
FIG. 34
X
X
2
2.5
Gauge (mm)
74
3.5
(R4/I160)
CORDIS
The Community Research and Development Information Service
The databases - nine in total - are accessible on-line free of charge. As a user-friendly aid for on-line
searching, Watch-CORDIS, a Windows-based interface, is available on request. The databases are also
available on a CD-ROM. The current databases are:
News (English, German and French version) - Results Partners - Projects - Programmes - Publications Acronyms - Comdocuments - Contacts
Telephone:
Fax:
E-mail:
WWW:
+352-401162-240
+352-401162-248
helpdesk@cordis.lu
http://www.cordis.lu/
European Commission
EUR 17881 Mechanical working (rolling)
On-line calculation of time optimal pass sequences for cold rolling on
Sendzimir mills
O. Davies, P. Yates, J. Woodisse
Luxembourg: Office for Official Publications of the European Communities
1998 74 pp. 21 29.7 cm
Technical steel research series
ISBN 92-828-3127-2
Price (excluding VAT) in Luxembourg: ECU 13.50
This report describes the development of an on-line adaptive time optimal scheduling system
for Sendzimir mills. The software required to calculate the time optimal schedules is detailed
and the data required for optimisation are specified. The system developed ensures that the
calculated schedules respect the constraints required for quality strip production whilst
maximising the production capacity of the mill.
An investigation of austenitic stainless steel yield stress has been performed using both
compressive and tensile testing techniques. The results of these tests have indicated a high
dependence of the yield stress on the mode of deformation and the deformation
temperature. Standard cold rolling models have been evaluated and found to be inadequate
for the prediction of rolling forces on Sendzimir mills. An empirical model relating rolling
torque to pass descriptor variables has been developed. This model has been applied to
data obtained from a production Sendzimir mill using torque telemetry techniques.
A method of sensing the strain induced in the Sendzimir mill housing has been investigated
and a complex relation between strain, roll separating force and roll stack geometry has been
observed.
The ability to provide adaptive calculations of predicted torque values has not been
demonstrated owing to limited size of the sample base and a significant through pass
variation in the measured torque values. A procedure for the gradual installation of optimised
schedules has been proposed which respects strip surface and shape quality requirements.
Venta Salg Verkauf Sales Vente Vendita Verkoop Venda Myynti Frsljning
BELGIQUE/BELGIE
NEDERLAND
CYPRUS
INDIA
EBIC India
3rd Floor, Y. B. Chavan Centre
Gen. J. Bhosale Marg.
400 021 Mumbai
Tel. (91-22)282 60 64
Fax (91-22) 285 45 64
E-mail: ebic@giasbm01 .vsnl.net.in
STERREICH
MAGYARORSZAG
PORTUGAL
MALTA
Jean De Lannoy
Avenue du Roi 202/Koningslaan 202
B-1060 Bruxelles/Brussel
ISRAEL
DEUTSCHLAND
Bundesanzeiger Verlag
Breite Strae 78-80
Postlach 10 05 34
D-50667 Kln
Tel. (49-221)20 29-0
Fax (49-221) 202 92 78
E-mail: vertrieb@bundesanzeiger.de
URL: http7/www.bundesanzeiger.de
/GREECE
G. C. Elettheroudakis SA
International Bookstore
Panepistimiou 17
GR-10564Athina
Tel. (30-1)331 41 80/1/2/3
Fax (30-1) 323 98 21
E-mail: elebooks@netor.gr
ESPANA
Mundi Prensa Libros, SA
Castell. 37
E-28001 Madrid
Tel. (34-1)431 33 99
Fax (34-1) 575 39 98
E-mail: libreria@mundiprensa.es
URL: http://www.mundiprensa.es
Boletn Oficial del Estado
Tralalgar, 27
E-28010 Madrid
Tel. (34-1)538 21 11 (Libros)/
384 17 15 (Suscripciones)
Fax (34-1) 538 21 21 (Libros)/
384 17 14 (Suscripciones)
E-mail: webmaster@boe.es
URL: ht1p://www.boe.es
Akateeminen Kirjakauppa/Akademiska
Bokhandeln
Pohjoisesplanadi 39/
Norra esplanaden 39
PL/PB 128
FIN-00101 Helsinki/Helsingfors
./tin (358-9) 121 41
F./fax (358-9) 121 44 35
E-mail: akatilausstockmann.mailnet.fi
URL: http://booknet.cultnet.fi/aka/index.htm
BTJ AB
Traktorvgen 11
S-221 82 Lund
Tin (46-46) 18 00 00
Fax (46-46) 30 79 47
E-post: btjeu-pub@btj.se
URL: http://www.btj.se/media/eu
UNITED KINGDOM
SLOVENIA
Gospodarskl Vestnlk
Zalozniska skupina d.d.
Dunajska cesta 5
SLO-1000 Ljubljana
Tel. (386) 611 33 03 54
Fax (386) 611 33 91 28
E-mail: belicd@gvestnik.si
URL: http://www.gvestnik.si
ISLAND
IRELAND
NORGE
NIC Info A/S
Ostenjoveien 18
Boks 6512 Etterstad
N-0606 Oslo
Tel. (47-22) 97 45 00
Fax (47-22) 97 45 45
TRKIYE
Dnya Infotel AS
Istikll Cad. No: 469
TR-80050 Tnel-Istanbul
Tel. (90-212)251 91 96
Fax (90-212) 251 91 97
AUSTRALIA
Hunter Publications
PO Box 404
3167 Abbotsford, Victoria
Tel. (61-3)94 17 53 61
Fax (61-3) 94 19 71 54
SCHWEIZ/SUISSE/SVIZZERA
CANADA
OSEC
Stampfenbachstrae 85
CH-8035 Zurich
Tel. (41-1)365 53 15
Fax (41-1) 365 54 11
E-mail: uleimbacher@osec.ch
URL: http:/Avww.osec.ch
LUXEMBOURG
BLGARIJA
Europress-Euromedia Ltd
59. Bid Vitosha
BG-1000 Sotia
Tel. (359-2) 980 37 66
Fax (359-2) 980 42 30
Abonnements:
Messageries Paul Kraus
11. rue Christophe Plantin
L-2339 Luxembourg
Tel (352) 49 98 88-8
Fax (352) 49 98 88-444
E-mail: mpkOpt.lu
URL: httpyAvww.mpk.lu
CESK REPUBLIKA
ITALIA
Euromedia
Str. G-ral Berthelot Nr 41
RO-70749 Bucuresti
Tl. (40-1)21044 01.61406 64
Fax (40-1) 210 44 01, 312 96 46
SLOVAKIA
Journal officiel
Service des publications des CE
26, rue Desaix
F-75727 Paris Cedex 15
Tl. (33) 140 58 77 01/31
Fax (33) 140 58 77 00
Licosa SpA
Via Duca di Calabria. 1/1
Casella postale 552
1-50125 Firenze
Tel. (39-55) 64 54 15
Fax (39-55) 64 12 57
E-mail: licosa Itbccit
URL: http://www.rtbcc.it/licosa
ROMANIA
SVERIGE
FRANCE
POLSKA
Ars Polona
Krakowskie Przedmiescie 7
Skr. pocztowa 1001
PL-00-950 Warszawa
Tel. (48-22)826 12 01
Fax (48-22) 826 62 40, 826 53 34, 826 86 73
E-mail: ars_pol@bevy.hsn.com.pt
ROY International
17, Shimon Hatarssi Street
PO Box 13056
61130 Tel Aviv
Tel. (972-3)546 14 23
Fax (972-3) 546 14 42
E-mail: royil@netvision.net.il
NIS CR prodeina
Konviktsk 5
CZ-113 57Praha 1
Tel. (420-2) 24 22 94 33, 24 23 09 07
Fax (420-2) 24 22 94 33
E-mail: nkposp@dec.nis.cz
URL: http://www.nis.cz
CCEC
60-letiya Oktyabrya Av. 9
117312 Moscow
Tel. (70-95) 135 52 27
Fax (70-95) 135 52 27
SOUTH AFRICA
Satto
5th Floor Export House,
CNR Maude & West Streets
PO Box 782 706
2146Sandton
Tel. (27-11)883 37 37
Fax (27-11)883 65 69
SOUTH KOREA
Kyowa Book Company
1 F1. Phyung Hwa Bldg
411-2 Hap Jeong Dong. Mapo Ku
121-220 Seoul
Tel. (82-2) 322 67 80/1
Fax (82-2) 322 67 82
E-mail: kyowa2@ktnet.co.kr.
THAILANDE
EBIC Thailand
Vanissa Building 8th Floor
29 Soi Chidlom
Ploenchit
10330 Bangkok
Tel. (66-2) 655 06 27
Fax (66-2) 655 06 28
E-mail: ebicbkk@ksc15.th.com
UNITED STATES OF AMERICA
EGYPT
The Middle East Observer
41, Sheril Street
Cairo
Tel. (20-2) 393 97 32
Fax (20-2) 393 97 32
HRVTSKA
Mediatrade Ltd
Pavia Hatza 1
HR-10000 Zagreb
Tel. (385-1)43 03 92
Fax (385-1) 43 03 92
Bernan Associates
4611 -F Assembly Drive
MD20706 Lanham
Tel. (800) 274 44 47 (loll free telephone)
Fax (800) 865 34 50 (toll free lax)
E-mail: query@beman.com
URL: http:/Avww.beman.com
ANDERE LANDER/OTHER COUNTRIES/
AUTRES PAYS
Bitte wenden Sie sich an ein Bro Ihrer
Wahl / Piense contact the sales office of
your choice / Veuillez vous adresser au
bureau de vente de votre choix
cc
co
CO
Euroabstracts
The European Commission's periodical on research publications, issued every two months.
For more information, contact:
RTD help desk, European Commission, DG XIII, L-2920 Luxembourg
Fax (352) 43 01-32084; e-mail: rtd-helpdesk@lux.dg13.cec.be
EUW
op *
*
--317-
F
OF ICE F OR OF F ICIAL PUBLICATIONS
OF THE EUROPEAN COMMUNITIES
789282"831274
L-2985 Luxembourg