Some fundamental considerations for the control of residual
atness in tension levelling
J.W. Morris a,* , S.J. Hardy b , J.T. Thomas c a Department of Materials Engineering, EngD Centre, University of Wales, Swansea, Singleton Park, Swansea SA2 8PP, UK b Department of Mechanical Engineering, University of Wales, Swansea, Singleton Park, Swansea SA2 8PP, UK c Corus Group R, D & T, Welsh Technology Centre, Port Talbot, South Wales SA13 2NG, UK Received 16 February 2001; accepted 19 September 2001 Abstract Tension levelling is a process used in the steel industry in order to remove any shape defects present in temper rolled strip. It is generally the nal process before the cold rolled product is despatched to the customer, and therefore plays an important role in delivering the desired material properties and the product standards required. In this paper a designed factorial analysis has been employed in order to study the effect that tension levelling process parameters have on the shape characteristics of the levelled product. These characteristics include residual atness (longitudinal curvature or longbow), centreline elongation and residual (or internal) stress imbalances; criteria which determine whether the customer accepts the material or not. It has been identied that in a basic ve-roll leveller the nal, adjustable roll wrap angle, has the most signicant inuence on these characteristics in over 70% of cases. It has also been determined that nal atness is dictated by many second-order process interactions. # 2002 Published by Elsevier Science B.V. Keywords: Tension levelling; Statistical process control; Mathematical modelling 1. Introduction With the introduction of sophisticated roll congurations in the accumulators of steel processing lines, an ever increasing requirement of the rolling mills is to supply material which is of very slight full-centre. The introduction of this shape into the material, coupled with processing through a series of concave/convex rolls, has the advantage of allowing high tracking control. Such control is essential when strip material in coil form is processed through the furnaces of continuous annealing processing lines. In some cases, however, the material may be over-rolled at sections across the width due to problems at the rolling mills. In thinner, high strength materials, and to a certain degree in heavier gauge materials, this can lead to the introduction of excessive, manifest shape defects such as edge or centre waves. These manifest shape defects can also be introduced by the skin-pass operation at the temper mill due to poor rolling conditions. Since these defects are unacceptable to some customersfor example in ofce furniture applica- tions, where strip of near-perfect shape is requiredthe strip has to be tension levelled in order to achieve the required degree of atness. In tension levelling, the mechanism that provides this shape removal is a combination of bending and signicant longitudinal tension, Fig. 1. Bending occurs over relatively small diameter rolls, with a typical diameter of 60 mm used when processing steel strip. This requirement, however, will change depending on the characteristics of the processed material. For example, when tension levelling tinplate mate- rial, it is not uncommon for the machine to use 2025 mm diameter bend rolls. As a result of processing material with longer longitudinal bres (in the case of centre buckle, the centre longitudinal bres across width will be longer) in some sections across the width, strain equalisation occurs. Preferential elongation of the shortest longitudinal bres ensures that all bres are of equal length. The tension applied in most cases is predominantly a function of the yield stress of the incoming material, with a feasible window of between 10 and 40% of initial yield used. A typical tension levelling machine will contain at least ve main bend rolls, with other arrangements and additional rolls included depending on the manufacturer and the type of material most commonly processed through the machine. It is usual to x bend rolls one, three and ve at the passline Journal of Materials Processing Technology 120 (2002) 385396 * Corresponding author. 0924-0136/02/$ see front matter # 2002 Published by Elsevier Science B.V. PII: S0 9 2 4 - 0 1 3 6 ( 0 1 ) 0 1 1 7 5 - X whilst adjusting the penetration depths of Rolls 2 and 4 (Fig. 1), such that a decreasing bending moment is progres- sively achieved towards the exit end of the machine. It is convenient to think of this bending moment in terms of a so- called wrap angle, i.e. the angle subtended by the radii of the nominal points of contact over a roll, and therefore to reduce the magnitude of this wrap progressively towards the exit end of the leveller. This adjustment is usually performed on- line in relatively small increments in order to obtain opti- mum shape removal at Rolls 2 and 3 (via complete con- formation at the rollstrip interface). Rolls 4 and 5 are then set accordingly, such that at material is achieved post- levelling due to a bend radius signicantly larger than that of the roll radius. So-called ``shape'' can generally be classied into two groupslatent and manifest. Latent shape appears visible when coil or sheet is slit or cut. This latent property can make the material appear at prior to a cutting operation, and can be attributed to residual (or internal) stress differentials through-gauge and cross-width. Similarly, manifest shape is associated with defects which appear visibly (either in coiled form or under line tension ) such as edge-waves and centre/ quarter-buckling. Surface area defects such as longbow (permanent longitudinal curvature) can be classed as latent since this defect, in general, is only visible when the coiled material is reduced into sheet form. Roller levelling or multi-roll levelling is a process used to remove shape in its various latent forms. In contrast to tension levelling, the process of roller levelling is restricted in its use by the nature of the incoming shape defect. In general, roller levellers are best suited to remove defects of latent shape such as longbow due to power restrictions, specically front-line tension capabilities. The roller leveller may contain more than twice the number of work rolls compared to that of its tension levelling counterpart, with only sufcient line tension to pull the material through the machine. The two processes can be used in tandem [1], in order to produce material which is required to be perfectly at, and to produce desirable residual stress distributions through-gauge. In this situation, the tension leveller is predominantly used upstream to remove manifest defects, whilst the roller leveller, with its lowline tension and smaller diameter work rolls, is used to obtain perfectly at strip with balanced internal stresses. 2. Tension leveller analysis methodology Historically, empirical methods have been employed to analyse the process of tension levelling [2,3]. Commission- ing of leveller machines demand that the correct settings are obtained for each material that the leveller will process. This measure usually involves two explicit stages: rstly, the correct roll penetrations or plunge depths at early rolls must be obtained, which ensures full conformation of strip to the bend roll. Secondly, the nal roll(s) are adjusted incremen- tally until at strip is observed from cut samples. In some cases, a matrix of leveller settingswhich may take many man-years to completefor the whole product range is obtained. These settings are then used as a reference for subsequent incoming material. However, this method is not without its problems. In most cases, the obtained matrix leveller settings for the subsequent material do not produce at material, and can deviate from atness by 20 mm. The problem with employing such a method is that no account is taken of the natural variation in mechanical properties in the coiled material which occur from head-to-tail end. Other factors, such as linespeed, elongation/tension and bend roll wrap angle can also vary throughout processing to a small degree, which may detrimentally inuence the nal atness. The tension levelling process has also been analysed extensively using numerical techniques [48]. Analytical solutions have been proposed [9,10], mainly as a develop- ment tool or guide, in order to determine process character- istics. Analytical solutions of residual curvature are not known to exist, and it is necessary, therefore, to perform such an analysis with the aid of a numerical solution based on constitutive plasticity laws [11,12]. 3. Statistically designed experiments It is becoming ever more important to understand and, ultimately, control process parameters which affect the residual atness of the coiled product. In order to achieve at material across the product range using tension levellers, it is often necessary to perform some type of parametric study to obtain a gauge/width matrix. Frequently this matrix will be chemistry dependent. This method, however, is extremely labour intensive and, moreover, does not offer Fig. 1. Schematic layout of general five roll tension leveller, showing contact wrap angle (y at each bend roll). 386 J.W. Morris et al. / Journal of Materials Processing Technology 120 (2002) 385396 a robust theoretical consistency which is applicable between machines and consecutive coils. A process tool employed for such an analysis is the area of statistical process control known as robust design. Experimental or robust design is a tool used by statistical process control engineers in order to determine those pro- cess variables which affect the quality and consistency of a product. Fig. 2 gives a simplied view of robust design. The number of experiments can be reduced by using the Taguchi approach. Robust design was developed in the early 1950s by Taguchi [13]. In Taguchi methodology, the experiments are not designed. Instead, so-called orthogonal arrays are given for a particular project, such as the popular L 8 (2 7 ) array. This particular design is capable of analysing up to seven process variables, with each variable set at two levels (hence the descriptor 2 7 ), using as few as 16 individual analyses. Unfortunately, there are major shortcomings of this method. For example, the aliasing structure is not self- evident from the interaction tables given; this structure is also time-consuming to obtain. (All location effectsor effects which show the extent to which the average response is inuenced by a change in the level of that factor estimated from a fractional factorial design are so-called aliased effects, i.e. the effects of changes in the factors on the response cannot be distinguished from one another and are, therefore, said to be aliased). In addition, the orthogonal arrays can be interpreted as fractional factorial arrays in most cases, which can give misleading results to the designer. This is of particular concern when dealing with the interactions between the factors as partial aliasing of two-factor interactions with main effects can occur. On the other hand, experiments which have been statis- tically designed for a particular project offer a more suitable and effective alternative. In contrast with the Taguchi method, statistically designed experiments involve planning the whole analysis. This results in a more controlled and systematically designed set of experiments, including a full alias structure, and is the method chosen here to analyse the basic ve-roll tension levelling process. 4. Factorial analysis An engineering experiment has been designed in order to investigate the effect particular parameters of the tension levelling process have on the characteristics of the levelled product. The aim is to nd levels of the controllable factors of the process that are least inuenced by noise factors, and to reduce variability around a mean target value or response, Y. Each experiment has been carried out using a fully validated ABAQUS nite element (FE) model, the details of which can be found elsewhere [14]. Fig. 3 shows the levelling conguration and material geometry used in the FE simulations. The response, Y, takes several forms, although each test response shares the same base and test design. Each response has been identied as follows: residual atness (Dr), elongation (e), topbottom surface residual stress imbal- ances including whole surface, edge/centre imbalances and residual stress levels (normalised to yield stress) at Fig. 2. General model of a manufacturing system. Fig. 3. Levelling configuration and materials geometry used for the factorial experiments. Centreline symmetry has been assumed, giving a total strip width of 100 mm. The gauge used is 0.7 mm. J.W. Morris et al. / Journal of Materials Processing Technology 120 (2002) 385396 387 mid-thickness. Residual atness is taken from a fully levelled, 600 mm length of strip, with elongation response taken as the total centreline elongation of the material. The stress balances were calculated by taking the sum of the differences in stress magnitudes at each element integration point across-width. For the strip edge, residual stress imbal- ances were calculated from 25% of the full width, i.e. at a distance, k edge , of k edge = distance total width = 25 100 mm = 25% section Similarly, for the strip centre k centre = 25 100 mm = 25% section The sum k edge k centre is equal to 50% since centreline symmetry in the FE model has been assumed. A graphical description of each response is given in Fig. 4. The meth- odology used to establish the experiments, including a description of the factors, their associated levels and coded test conditions, is given in Appendix A. The full factorial analysis requires n k tests, where n designates the number of levels allocated to each factor and the superscript k refers to the number of factors (linear dependency is assumed). For ve factors each at two levels this yields 2 5 or 32 experiments. It has been identied that a one-half fractional factorial of design 2 k1 V can be employed in order to reduce the number of tests performed, giving 2 51 or 16 tests. (The subscript ``V'' indicates that the design is of resolution ve.) Therefore, no main or rst-order interaction location effects are aliased with any other main or rst-order interaction location effects. First-order interactions are actu- ally aliased with higher order interaction location effects, allowing all main and rst-order interactions to be identied. For robust design, a resolution of III should not be considered as an alternative. Since an uncontrollable or noise factor is present, the emphasis on the initial design must be on making the process robust to noise. The 15 aliased relation- ships for the 2 51 V design are given in Appendix A.4. These effects have been obtained from the design generator E = ABCD = ABCDE (1) according to Montgomery [15] for a 2 51 V one-half fractional factorial design. To identify the important parameters of the tension level- ling process, it is assumed that each location effect has a mean Fig. 4. Description of responses used in the factorial analysis. 388 J.W. Morris et al. / Journal of Materials Processing Technology 120 (2002) 385396 of zero, and that each effect has the same standard deviation. This is achieved by ranking each estimated location effect from most negative to most positive. From this, it follows that each effect has exactly the same normal distribution, and thus, each effect represents a point under this distribution. A cumulative probability (CP) is then assigned for each loca- tion effect. The CP associated with each location effect is given by the following expression: CP i = 100(i 0:5) M 1 (2) where M is the number of test conditions, i is a rank index equal to 1 for the most negative location effect, 2 for the next most negative and so on up to M for the most positive location effect. Important location effects in subsequent probability plots are seen as those factors which deviate from a linear regression line drawn through location effects which have a mean or net effect of zero. 5. Results The probability plots of estimated location effects for the seven identied responses are shown in Figs. 510. A summary of these results can also be found in Table 1. 5.1. Flatness Five levelling factors, as a combination of single factors and interactions, inuence the nal atness of the material (Fig. 5): v The single most influential factor is Roll 4 wrap angle (factor B). This design factor only has a positive influence on flatness, i.e. regardless of the setting in these experi- ments, the material will always maintain a positive (bow- ing up) longitudinal curvature. In over 70% of responses, Roll 4 wrap angle also appears as the most influential. The yield stress of the material, factor E, is the next most influential factor. v Four interactions of factors, Roll 4 wrap angle/yield stress, Roll 4 wrap angle/elongation, Roll 2 wrap angle/Roll 4 wrap angle and linespeed/yield stress (i.e. BE, BD, AB and CE, respectively) in order of rank. The BD interaction, like factor B alone, has only a positive influence whilst all other important interactions have a negative influence on flatness, i.e. these interactions Fig. 5. Probability plot of estimated location effectsflatness response. Fig. 6. Probability plot of estimated effectselongation response. Fig. 7. Probability plot of estimated location effectsresidual stress imbalance at the strip edge. J.W. Morris et al. / Journal of Materials Processing Technology 120 (2002) 385396 389 produce material that has negative longitudinal curvature (bowing down). v It is worth noting that no single parameter, with the exception of Roll 4 wrap angle, appears as influential in dictating final flatness. 5.2. Elongation Four factors affect the total centreline elongation as shown in Fig. 6. v The most influential factor affecting elongation is the line tension, factor D. This influence is always positive. Similarly, Roll 2 wrap angle, factor A, has a positive effect and is the next most influential factor. v Those factors which attempt to inhibit the elongation are the yield stress (factor E), and the interaction between Roll 4 wrap angle and the yield stress, BE. Fig. 8. Probability plot of estimated location effectsresidual stress imbalance at the strip centre. Fig. 9. Probability plot of estimated location effectsresidual stress imbalance between top and bottom surfaces. Fig. 10. Probability plot of estimated location effectsnormalised residual stress at the strip centreline (through gauge) as a function of yield stress. Table 1 Summary of all seven factorial experiments a Response Individual factor rank Interaction rank A B C D E AB AD BD BE CE Flatness 1 4 3 2 5 Elongation 2 1 3 4 Percent difference in yield 1 2 Centreline residual stress 4 3 1 2 Centre balance 1 3 2 Edge balance 1 2 Topbottom balance 1 3 2 a Numbers corresponding to each factor refer to its rank of importance or most influential factor within that experiment (number 1 indicates highest rank). 390 J.W. Morris et al. / Journal of Materials Processing Technology 120 (2002) 385396 5.3. Residual stress As with atness, the most inuential factor affecting the residual stress balances at the edge, centre and surface differentials (Figs. 7, 8 and 9, respectively), is the wrap angle on Roll 4. The normalised residual stress levels at the centre of the material (taken as the residual stress:yield stress ratio) are inuenced by the line tension or elongation applied to the centreline, Fig. 10. 6. Control of final flatness From the probability plots in Figs. 510, it can be seen that the most inuential factor in most responses is the wrap angle on Bend Roll 4. The next most inuential factors appear as interactions in the probability plot of estimated location effects. The process is, therefore, dictated by many of these second-order effects. This shows that the effect of one factor alone may not be important until its parent, interacting factor, is set at a particular level. Furthermore, the adjustment of a single factor by the operator will have wider implications on the atness response which may not necessarily be obvious. This phenomena can be used to minimise the variation in nal atness, commonly found on plant in the levelled product. The most inuential interaction is that of the wrap angle on Roll 4 (factor B) and the yield stress of the material (factor E). Since the yield stress of the material is a noise factor, i.e. difcult to control at the point of manufacture or the point of processing, the effect that this noise has on the levelled product must be minimised. This can be achieved by rst looking at the interaction between the interacting factors in a two-way diagram, Fig. 11. This gure shows a graphical interpretation of the interaction between Roll 4 wrap angle and yield stress. The two lines are not parallel, indicating that a strong interaction between the factors is present. When yield stress is at its lower level (1) the likely variation in atness to be expected is 30 mm, depending on the level at which the Roll 4 wrap angle is set. Similarly, if the yield stress is at its higher level (1), the variation is signicantly smaller, this being 10 mm. Avariation in yield stress from head-to-tail in the coil (190230 MPa), therefore, provides a rationale for observed variations in atness in levelled material through-coil. If the elongation and Roll 4 wrap angle interaction (BD) is considered, as shown in Fig. 12, it can be seen that when elongation is maintained at its lower level (0.3% in this case) the maximum variation in atness to be expected is 10 mm. If the elongation applied is at its higher level (0.6%) then the expected variation increases signicantly to 30 mm. Variation in atness may also be controlled by setting levels of Rolls 2 and 4, accordingly, as shown in the two-way diagram in Fig. 13. The interaction between these two factors shows that by setting Roll 2 wrap angle (or penetration) to its higher level, and adjusting Roll 4 accord- ingly between the two levels used on the experiment, the variation is minimised. Indeed, it is common practice to use such a method to obtain minimum nal atness. To minimise the effect that yield stress noise has on the nal atness, it is necessary to consider its interaction with other design factors. It has been identied that the wrap angle on Roll 4 interacts signicantly with the yield stress. A second factor which interacts with the yield stress is the process linespeed (factor C). This interaction is shown in Fig. 14. Although the inuence is small relative to that Fig. 11. Two-way diagram showing how the flatness characteristic of strip is influenced by the interaction between process factors of Roll 4 wrap angle and material yield stress. J.W. Morris et al. / Journal of Materials Processing Technology 120 (2002) 385396 391 shown in Fig. 11, for example, by ensuring that the linespeed is maintained at its lower level (in this case 60 m/min) the maximum expected variation in nal atness using other factors set at correct levels is 3 mm, whereas setting the linespeed to its higher level of 100 m/min generates a relatively larger variation in atness. 7. Simple model development Having found the most inuential factors and the applic- able interactions between factors, it is now possible to develop a simple linear regression model in order to predict the response under several leveller settings. This simple model is of the form: Y j = model prediction error correction where Y j is the respective response of flatness or residual stress imbalance, etc., the model prediction describing the effect of the design factors on the applicable response (the emphasis here will be on the flatness response) and the error correction or noise term accounting for any departures in the model. For linear regression the model appears in the form: Y = b 0 b 1 x 1 b 2 x 2 b n x n f e (3) Fig. 12. Two-way diagram showing how the flatness characteristic of strip is influenced by the interaction between process factors of Roll 4 wrap angle and total centreline elongation. Fig. 13. Two-way diagram showing how the flatness characteristic of strip is influenced by the interaction between process factors of Roll 4 wrap angle and Roll 2 wrap angle. 392 J.W. Morris et al. / Journal of Materials Processing Technology 120 (2002) 385396 where Y represents the response for n variables of x, f e the error term and b are coefficients. Eq. (3) may be modified to contain only those factors and interactions which are the most influential in affecting flatness. The coefficients used are those effects in the factorial analysis which most influ- ence the response, and the variables are the relevant factors. This newly formed equation can then be used to predict the flatness under many different levelling conditions. The factors which most inuence the atness deviation in order of rank (rank 1 indicates most inuential) are 1. The wrap angle on Roll 4, factor B; 2. The wrap angle on Roll 4 and yield stress interaction (BE); 3. The wrap angle on Roll 4 and elongation interaction (BD); Fig. 14. Two-way diagrams showing how the flatness characteristic of strip is influenced by the interaction between process factors of material yield stress and line speed. Fig. 15. Validation of simple model. J.W. Morris et al. / Journal of Materials Processing Technology 120 (2002) 385396 393 4. The wrap angle on Roll 2 and wrap angle on Roll 4 interaction (AB); 5. The linespeed and yield stress interaction (CE). The simple linear atness model now takes the form Y j = 1:06 17:81B 11:32BE 9:98BD 4:65AB 3:55CE (4) The model is based on the five important effects that have been identified from the analysis. These are the independent variables, with the coefficients taken from the main b effect (the average of all responses at each test condition). The first term in Eq. (4) is the mean of all test responses (b 0 ). The error term (f e ) is the sum of the insignificant effects, which also gives the so-called prediction limits for the model. A series of conrmation tests have been performed in order to test the validity of the simple model expressed by Eq. (4). From the interaction plots of Figs. 1114, it has been decided that the coded test conditions which give the most desirable atness results are v Factor A high (248 wrap angle on Roll 2); v Factor B high (12.58 wrap angle on Roll 4); v Factor C low (60 m/min linespeed); v Factor D low (0.3% total elongation); v Factor E high (230 MPa yield stress); which yields a flatness deviation of 5:65 11:04 mm. Other tests show that lower levels of residual atness can be achieved. However, the factor levels required to achieve this magnitude would produce a considerable variation in atness due to the changes in mechanical properties throughout the coil, as described earlier. Fig. 15 shows the full set of conrmation tests using Eq. (4) measured against ``actual'' values of atness for each coded test condition in the experiment. The ``actual'' values in Fig. 15 have been obtained from tests performed with the validated FE model. It is important to note that all predic- tions given by the linear model fall well within the prediction limits given by the error term, f e . 8. Conclusions A series of statistically designed experiments have been used to analyse the inuence of the tension levelling process parameters on cold rolled strip characteristics in a basic ve- roll leveller. The following conclusions can be drawn from the analysis: 1. The process parameter which has the most significant influence on final flatness is Roll 4 wrap angle. The process is influenced by many second-order interactions in order of rank between (i) Roll 4 wrap angle and yield stress, (ii) Roll 4 wrap angle and elongation, (iii) Roll 2 wrap angle and Roll 4 wrap angle and (iv) the linespeed and yield stress interactions. 2. Various process parameters only become influential when parent (interacting) factors are set or changed to particular levels. No single parameter, with the excep- tion of Roll 4 wrap angle, appears as influential. The net effect on flatness when adjusting a single parameter is not just a function of that parameter, since many factors appear only as part of an interaction. 3. Finite changes in yield stress through-coil (1520%) cause significant variation in the final flatness. 4. Variability in flatness in the levelled product through- coil can be eliminated or minimised by setting those process factors which transmit noise to appropriate levels accordingly. 5. A single prediction equation can be used with ease and reasonable accuracy to estimate final flatness, given values of the input factors. Acknowledgements The authors would like to thank British Steel Strip Products (now part of Corus Group plc) for permission to publish this paper and gratefully acknowledge the sup- port of the Engineering and Physical Sciences Research Council. Appendix A. Derivation of base design, coded test conditions and test matrix A.1. Factor definitions A.2. Base design Once the factors and the levels at which they are to be set have been identied, the base design can be obtained. The base design for any fractional factorial design of the form 2 k1 is the full 2 k1 matrix. Numbers within the matrix (below) designate the level at which the factor is set for a particular coded test condition. For example, a number ``1'' in the coded test conditions elements indicates a test in which that factor is set to its higher level. Similarly, ``1'' indicates that a factor has been set at its lower level. The Factor Definition Factor level and specification A Roll 2 wrap angle a High=24
Low=18
B Roll 4 wrap angle High=12:5
Low=6
C Line speed High=100 m=min Low=60 m=min
D Elongation High=0:6% Low=0:3% E Yield stress High=230 MPa Low=190 MPa a See Fig. 1 for roll number specifications. 394 J.W. Morris et al. / Journal of Materials Processing Technology 120 (2002) 385396 letters in the ``test'' column indicate which factor is set to its higher level. The rst test is designated ``(1)'' since all factors at this test condition are set to their lowest level. A.3. Introduction of the remaining factor The nal factorthe design generatoris introduced by setting levels equal to the product of the elements in the coded test conditions. Therefore, test ``(1)'' becomes test ``e'' since the design generator in Eq. (1) requires that the row elements of factors A, B, C and D are multiplied, hence E = (1) 4 = 1. The base design now takes the form A.4. Alias structure Using alias algebra and referring to Eq. (1), we obtain E = ABCD = ABCDE The table above shows all 15 aliased location effects with first-order interactions highlighted in bold. The design gen- erator has been chosen such that no main location effect is aliased with any other main location effect, or with any first- order interaction location effect. Moreover, all first-order interactions are aliased with fourth-order interaction location effects. Hence, the design is the so-called resolution V. A.5. Defining the test matrix By conjugating both the base design matrix and the matrix containing the design-generated terms the test matrix, shown below, is obtained. Aliased location effects A=BCDE B=ACDE C=ABDE D=ABCE E=ABCD AB=CDE AC=BDE AD=BCE AE=BCD BC=ADE BD=ACE BE=ACD CD=ABE CE=ABD DE=ABC Test Coded test conditions A B C D E e 1 1 1 1 1 a 1 1 1 1 1 b 1 1 1 1 1 abe 1 1 1 1 1 c 1 1 1 1 1 ace 1 1 1 1 1 bce 1 1 1 1 1 abc 1 1 1 1 1 d 1 1 1 1 1 ade 1 1 1 1 1 bde 1 1 1 1 1 abd 1 1 1 1 1 cde 1 1 1 1 1 acd 1 1 1 1 1 bcd 1 1 1 1 1 abcde 1 1 1 1 1 Test Coded test conditions A B C D (1) 1 1 1 1 a 1 1 1 1 b 1 1 1 1 ab 1 1 1 1 c 1 1 1 1 ac 1 1 1 1 bc 1 1 1 1 abc 1 1 1 1 d 1 1 1 1 ad 1 1 1 1 bd 1 1 1 1 abd 1 1 1 1 cd 1 1 1 1 acd 1 1 1 1 bcd 1 1 1 1 abcd 1 1 1 1 Test Coded test conditions for full 2 4 E = ABCD Response A B C D (1) 1 1 1 1 a 1 1 1 1 b 1 1 1 1 ab 1 1 1 1 c 1 1 1 1 ac 1 1 1 1 bc 1 1 1 1 abc 1 1 1 1 d 1 1 1 1 ad 1 1 1 1 bd 1 1 1 1 abd 1 1 1 1 cd 1 1 1 1 acd 1 1 1 1 bcd 1 1 1 1 abcd 1 1 1 1 Coded test conditions for fractional 2 51 e 1 1 1 1 1 Y 1 a 1 1 1 1 1 Y 2 b 1 1 1 1 1 Y 3 abe 1 1 1 1 1 Y 4 c 1 1 1 1 1 Y 5 ace 1 1 1 1 1 Y 6 bce 1 1 1 1 1 Y 7 abc 1 1 1 1 1 Y 8 d 1 1 1 1 1 Y 9 J.W. Morris et al. / Journal of Materials Processing Technology 120 (2002) 385396 395 References [1] M. Kobayashi, M. Yamashita, K. Takezawa, Establishment of a process for producing dead-flat thin gauge strips using levellers, Kawasaki Seitetsu Giho 27 (3) (1995) 136142. [2] R.R. Hilsen, N.H. Polakowski, Equipment and process variables in ten- sion levelling of carbon strip, Iron Steel Eng. October 1969, pp. 4756. [3] R.R. Hilsen, B.S. Levy, N.H. Polakowski, in: Mechanical Working and Steel Processing, Vol. VIII, Metallurgical Society of AIME, 1967, pp. 3155. [4] H.M. Williamson, Strain softening during tension levelling of rolled sheet metals, in: Rolling Technology, Institute of Metals, Australia, 1986, pp. 199205. [5] Y. Misaka, T. Masui, Shape correction of steel strip by tension leveller, Trans. ISIJ 18 (1978) 475484. [6] T. Sheppard, J.M. Roberts, On the mechanics of the tension-levelling process, J. Inst. Met. 99 (1971) 293301. [7] T. Sheppard, J.M. Roberts, J. Inst. Met. 99 (1971) 223228. [8] A. Noe, Theoretical and practical investigation on strecher-and-roller levelling, Stahl und Eisen 106 (21) (1986) 6773. [9] M.G. Kinnavy, Recent developments in tension levelling, Iron and Steel Engineer Year Book, November 1972, pp. 6772. [10] R.A. Bland, M.F. Alters, Tension levelling of ferrous and non-ferrous strip, Iron and Steel Engineer Year Book, 1967, pp. 613623. [11] R. Hill, The Mathematical Theory of Plasticity, Oxford Engineering Science Series, Oxford University Press, Oxford, 1950. [12] A. Mendelson, Plasticity: Theory and Application, Collier Macmil- lan, New York, 1968. [13] G. Taguchi, Introduction to quality engineering: designing quality into products and processes, Asian Productivity Organization (ISBN: 9283310845), 1986. [14] J.W. Morris, S.J. Hardy, J.T. Thomas, A.W. Lees, in: Proceedings of the Conference Tension and Roller Levelling of Flat Products, Whitehall, London, March 1999, Paper 5, The Institute of Materials. [15] D.C. Montgomery, Design and Analysis of Experiments, 3rd Edition, Wiley, New York, 1991. Appendix A.5 (Continued ) Test Coded test conditions for full 2 4 E = ABCD Response A B C D ade 1 1 1 1 1 Y 10 bde 1 1 1 1 1 Y 11 abd 1 1 1 1 1 Y 12 cde 1 1 1 1 1 Y 13 acd 1 1 1 1 1 Y 14 bcd 1 1 1 1 1 Y 15 abcde 1 1 1 1 1 Y 16 396 J.W. Morris et al. / Journal of Materials Processing Technology 120 (2002) 385396