FLATNESS CONTROL USING OPTIMIZER

Abstract

A metal strip is rolled in a roll stand and a control device for the roll stand determines, by means of a working cycle, a number of manipulated variables for flatness actuators of the roll stand and actuates them accordingly. The control device implements an optimizer, which provisionally sets the current correction values, and determines a totality of flatness values. Then, the optimizer minimizes the relationship by varying the current correction variables. When determining the current correction variables (s), the optimizer considers linear ancillary conditions, based at least in part on a vector having the ancillary conditions upheld by the current correction values and a vector having the ancillary conditions upheld by the difference of the current correction values relative to the correction values of the preceding working cycle. The control device determines the manipulated variables for the flatness actuators in consideration of the determined current correction variables.

Claims

1-17. (canceled)

18. An operating method for a roll stand in which a metal strip is rolled, comprising: determining, by a control device for the roll stand in each of a working cycle of the control device, a number of manipulated variables (S) for a corresponding number of flatness actuators of the roll stand; and actuating, in each working cycle by the control device, the flatness actuators according to the manipulated variables (S) determined, wherein the control device implements a first optimizer, which initially sets current correction values (s) in a provisional manner and determines a totality of flatness values (f) according to the relationship:
f(s)=f0+W.Math.(ss) or f(s)=f0+W.Math.s, wherein: s is a totality of the current correction values, f0 are initial flatness values, W is an effectiveness matrix, and s is a totality of the correction values determined in the preceding working cycle, and then minimizes the relationship:
f(s)f*+ss0+ss to determine the current correction values (s) by varying the current correction values (s), wherein: f* is a totality of flatness target values, s0 is a totality of target values for the current correction values (s), and and are weighting factors, wherein the first optimizer considers linear ancillary conditions when determining the current correction values (s), wherein the linear ancillary conditions have the form:
C.Math.sB or the form:
C.Math.sB and |ss|<c, wherein C is a matrix, B is a vector consisting of the ancillary conditions to be upheld by the current correction values (s) and c is a vector consisting of the ancillary conditions to be upheld by the difference between the current correction values (s) and the correction values (s) of the preceding working cycle, and wherein the control device determines the manipulated variables (S) for the flatness actuators taking account of the current correction values (s) determined by the first optimizer.

19. The operating method as recited in claim 18, wherein: a measuring device is used to detect measured flatness values (fM) in a spatially resolved manner over the width of the metal strip, and the measured flatness values (fM) and the associated flatness target values (f*) are supplied to the control device as initial flatness values (f0).

20. The operating method as recited in claim 19, wherein: the control device implements a flatness controller arranged downstream of the first optimizer to which the current correction values (s) determined by the first optimizer are supplied and which flatness controller determines change values for the manipulated variables (S) for the flatness actuators from the current correction values (s), the flatness controller forms a sum of the current correction values (s) weighted with a gain factor (KP) and an output signal (s) from a plant model of the roll stand, the flatness controller determines a provisional signal (S) based on said sum, the flatness controller determines change values for the manipulated variables (S) for the flatness actuators by differentiating the provisional signal (S), and the evenness controller supplies the provisional signal (S) to the plant model of the roll stand as an input signal.

21. The operating method as recited in claim 20, wherein the flatness controller determines the provisional signal (S) by filtering the sum by means of a filter.

22. The operating method as recited in claim 21, wherein the control device dynamically parameterizes the filter.

23. The operating method as recited in claim 19, wherein: the control device implements a second optimizer, which is structured in the same way as the first optimizer, the totality of flatness values (f) for the second optimizer is determined on the basis of a totality of nominal flatness values (fW) and the current correction variables (s) valid for the second optimizer, the nominal flatness values (fW) correspond to a nominal change (FWN) in the rolling force (FW), the weighting factors (, ) for the second optimizer have the value 0, and the control device determines the manipulated variables (S) for the flatness actuators while additionally taking into account an actual change (FW) in the rolling force (FW), the nominal change (FWN) in the rolling force (FW) and the current correction values (s) determined by the second optimizer.

24. The operating method as recited in claim 18, wherein: the initial flatness values (f0) correspond to a totality of nominal flatness values (fW), the nominal flatness values (fW) correspond to a nominal change (FWN) in the rolling force (FW), the weighting factors (, ) have the value 0, and the control device determines the manipulated variables (S) for the flatness actuators taking account of an actual change (FW) in the rolling force (FW), the nominal change (FWN) in the rolling force (FW) and the current correction values (s) determined.

25. The operating method as recited in claim 18, wherein the first optimizer varies the current correction values (s) in a plurality of iterations.

26. The operating method as recited in claim 25, wherein the first optimizer stops varying the current correction values (s) as soon as at least one of: the first optimizer has carried out a predetermined number of iterations; the first optimizer has varied the current correction values (s) for a predetermined time; the current correction values (s) only change insignificantly from iteration to iteration; the relationship:
f(s)f*+ss0+ss only changes insignificantly; or another termination criterion is met.

27. The operating method as recited in claim 25, wherein the first optimizer considers the ancillary conditions in each iteration.

28. The operating method as recited in claim 18, wherein the first optimizer determines the correction values (s) according to an interior-point method.

29. The operating method as recited in claim 28, wherein the control device determines the effectiveness matrix (W) automatically on the basis of models of the roll stand.

30. The operating method as recited in claim 29, wherein the control device dynamically redetermines the effectiveness matrix (W) each time immediately before the commencement of the rolling of a respective metal strip.

31. The operating method as recited in claim 30, wherein the control device also dynamically tracks the effectiveness matrix (W) during the rolling of the respective metal strip.

32. A computer program comprising machine code that can be executed by a control device for a roll stand for rolling a metal strip, wherein the execution of the machine code by the control device causes the control device to operate the roll stand according to an operating method as recited in claim 18.

33. A control device for a roll stand for rolling a metal strip, wherein the control device is embodied as a software-programmable control device and programmed with a computer program as recited in claim 32.

34. A roll stand for rolling a metal strip, wherein the roll stand comprises a number of flatness actuators that are able to influence the flatness of the metal strip emerging from the roll stand, wherein the roll stand comprises a control device for the roll stand for rolling the metal strip, the control device being embodied as a software-programmable control device programmed with a computer program, the computer program comprising machine code that can be executed by the control device, and wherein the execution of the machine code by the control device causes the control device to operate the roll stand according to an operating method, by which the flatness actuators of the roll stand are actuated according to the operating method as recited in claim 18.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

[0084] The above-described properties, features and advantages of this invention and the manner in which they are achieved will become clearer and more plainly comprehensible in conjunction with the following description of the exemplary embodiments explained in more detail in conjunction with the drawings. The drawings show in schematic form:

[0085] FIG. 1 a roll stand for rolling a metal strip from the side,

[0086] FIG. 2 the roll stand in FIG. 1 from above,

[0087] FIG. 3 the roll stand in FIG. 1 viewed in the direction of transport of the metal strip,

[0088] FIG. 4 a measuring device from below,

[0089] FIG. 5 the internal structure of a control device,

[0090] FIG. 6 a flowchart,

[0091] FIG. 7 a modification of the internal structure in FIG. 5, and

[0092] FIG. 8 an alternative embodiment of the internal structure in FIG. 5.

DETAILED DESCRIPTION

[0093] According to FIGS. 1 to 3, a roll stand for rolling a metal strip 1 includes a number of rolls 2, 3. Generally, additionally to working rolls 2, the rolls 2, 3 comprise support rolls 3. Frequently, no further rolls are present. In this case, the roll stand is a four-high stand. In some cases, further rolls are also present, for example in the case of a six-high stand an intermediate roll in each case between the two working rolls 2 and the two support rolls 3. Other embodiments are also known, for example a 12-roller roll stand or a 20-roller roll stand.

[0094] The roll stand is controlled by a control device 4. The control device 4 is generally embodied as a software-programmable control device. This is indicated in FIG. 1 by the reference P (for microprocessor) within the control device 4. The control device 4 is programmed with a computer program 5. The computer program 5 comprises machine code 6 that can be executed by the control device 4. The execution of the machine code 6 by the control device 4 causes the control device 4 to operate the roll stand according to an operating method according to the invention.

[0095] The following initially explains a first basic embodiment in greater detail with additional reference to FIGS. 4 and 5.

[0096] In the context of the first basic embodiment, a measuring device 7 by means of which measured flatness values fM are detected during the operation of the roll stand is arranged downstream of the roll stand in accordance with FIG. 1. The detection of the measured flatness values fM takes place in a spatially resolved manner over the width b of the metal strip 1. For example, the measuring device 7 can be embodied in accordance with the depiction in FIG. 4 as a segmented measuring roller arranged on the outlet side of the roll stand. Such segmented measuring rollers are generally known to those skilled in the art. Due to the detection of the measured flatness values fM in a spatially resolved manner over the width b of the metal strip 1, the measured flatness values fM do not represent a scalar, but a vector. Therefore, fM designates the totality of the measured flatness values fM and not only one single value detected at a specific location viewed in the width of the metal strip 1. The spatial resolution can be as required. The number of support points for which in each case an individual measured flatness value fM is detected often fluctuates within the upper double-digit range.

[0097] The measured flatness values fM are supplied to the control device 4 as initial flatness values f0 in accordance with FIG. 5. Furthermore, associated flatness target values f* are supplied to the control device 4. Hence, the flatness target values f* also do not entail one single target value, but the totality of the flatness target values f*, i.e. a corresponding vector.

[0098] The control device 4 determines by means of a working cycle T in each case a number of manipulated variables S for a corresponding number of flatness actuators 8 and actuates these according to the manipulated variables S determined. Therefore, the manipulated variables S are determined anew with each working cycle T. They then remain valid until the next determination of the manipulated variables S. The working cycle T is generally in a range of less than 1 second, for example between 0.2 seconds and 0.5 seconds.

[0099] The flatness actuators 8 are used to influence the flatness of the metal strip 1 emerging from the roll stand. For example, corresponding reverse-bending devices can exert reverse bending forces FR on the working rolls 2. Alternatively or additionally, if applicable, the working rolls 2 (or, if present, the intermediate rolls) can be axially displaced in accordance with an axial displacement A. Alternatively or additionally, a corresponding device can exert a segmented temperature influence. For example, a cooling device can be used for a respective local cooling K of the working rolls 2. Other flatness actuators 8 are also possible.

[0100] FIG. 5 shows the internal structure of the control device 4 for the first basic embodiment of the present invention. However, the blocks depicted in FIG. 5 are generally provided not as hardware, but as program modules. Therefore, they are obtained by executing the machine code 6 of the computer program 5.

[0101] In accordance with FIG. 5, the control device 4 implements an optimizer 9. In the following, the optimizer 9 is referred to as a first optimizer. In the context of the embodiment shown in FIG. 5, the first optimizer 9 is the sole optimizer.

[0102] Generally speaking, an optimizer for the purpose of the present invention is an arithmetic block to which certain input variables are supplied. The arithmetic block then determines a target function into which the input variables and the output variables applied by the arithmetic block are entered. The arithmetic block then varies the output variables with aim of optimizing the target function. To this end, the arithmetic block generally carries out a plurality of iterations, wherein it in each case determines in each iteration, in each case on the basis of the input variables and the last output variables applied, the target function and, on the basis of the target function determined, varies the output variables with the aim of optimizing the target function. Such optimizers are generally known to those skilled in the art. Purely by way of example, reference is made to optimizers which work in accordance with the following methods: [0103] continuous optimization methods, such as, for example, simplex methods, interior-point methods, trust-region methods, cubic-overregularization methods, SLP methods and methods of the Gaussian/Newtonian type, for example SQP methods. These methods can be embodied as linear or non-linear as required. [0104] methods of discrete optimization such as, for example, cutting-plane methods, methods of the branch-and-bound type, network optimization methods etc. [0105] methods of mixed-integer optimization, for example as a combination of continuous and discrete methods. [0106] heuristic and metaheuristic optimization methods, for example genetic methods, evolutionary methods, anti-colony optimization methods, swarm methods, simulated annealing and tabu search. [0107] genetic optimization methods.

[0108] If necessary, the above-named optimization methods can be combined with processing in a neural network.

[0109] The initial flatness values f0 and the flatness target values f* are supplied to the first optimizer 9. The first optimizer 9 determines correction values s, specifically a separate value in each case for each flatness actuator 8. The correction values s are valid for the current working cycle T and are therefore referred to below as current correction values s. Similarly to the initial flatness values f0 and the flatness target values f*, the reference character s therefore also stands for the totality of the current correction values. Therefore, here once againat least generallythis is not a scalar, but a vector. However, it is possible in individual cases that only one single flatness actuator 8 will be present. In this case, the vector s degenerates into a scalar. The meaning of the current correction values s will become apparent from later explanations.

[0110] The current correction values s are supplied to the first optimizer 9 again. However, they are first delayed by one (1) working cycle T in a time-delay element 10. Therefore, the correction values s supplied to the first optimizer 9 in a specific working cycle T correspond to the correction values for the previous working cycle T. Therefore, in the following, they are referred to as delayed correction values and given the reference character s.

[0111] The first optimizer 9 determines the current correction values s by minimizing the relationship

f(s)f*+ss0+ss(5)

[0112] Therefore, the first optimizer 9 varies the current correction values s until it has minimized this relationship. In other words: the first optimizer 9 initially applies the current correction values s as provisional values. Using the provisionally applied values for the correction values s, the first optimizer 9 then minimizes the above relationship by varying the current correction values s. The valid correction values s for the respective working cycle T are then the most recently determined or last varied current correction values s.

[0113] In said relationship, f is a totality of flatness values, i.e. once again a vector. The flatness values f are determined by the first optimizer 9 on the basis of the initial flatness values f0 and the current correction values s. For example, the first optimizer 9 can determine the flatness values f according to the relationship

f(s)=f0+W.Math.(ss)(6)

[0114] W is an effectiveness matrix. It specifies individually what influence a specific individual correction value s has on which of the flatness values f.

[0115] s0 is a totality of target values for the correction values s. The target values s0 can, for example, be determined such that the associated flatness actuators 8 are stressed as little as possible, for example actuated as little as possible. The target values s0 can be specified to the control device 5 as fixed. Alternatively, they can be specified to the control device 5 as variables or as parameters.

[0116] and are weighting factors. In each case, they have a non-negative value. They are generally greater than 0. They can be specified to the first optimizer 9 as fixed or can be parameterizable.

[0117] When determining current correction values s, the first optimizer 9 considers ancillary conditions. The ancillary conditions comprise linear ancillary conditions. Preferably, the ancillary conditions even comprise exclusively linear ancillary conditions.

[0118] In particular, the first optimizer 9 in each case considers linear ancillary conditions with the form

C.Math.sB(7)

[0119] Herein, C is a matrix. B is a vector having the ancillary conditions to be upheld by the current correction values s. In addition, the first optimizer 9 can consider further linear ancillary conditions with the form

|ss|c(8)

[0120] Herein, c is a vector having the ancillary conditions to be upheld by the difference between the current correction values s and the delayed correction values s.

[0121] Suitable optimizers are known to those skilled in the art per se. Therefore, the first optimizer 9 can be implemented as required. Preferably, the first optimizer 9 determines the correction values s according to an interior-point method.

[0122] The current correction values s determined by means of the first optimizer 9i.e. the current correction values s after the variation of the correction values srepresent the basis on which the control device 4 determines the manipulated variables S for the flatness actuators 8.

[0123] Generally, the first optimizer 9 varies the current correction values s in a plurality of iterations. Therefore, it attempts gradually to determine increasingly better current correction values s. In this case, the first optimizer 9 stops varying the current correction values as soon as at least one of the following termination criteria is met: [0124] the first optimizer 9 has carried out a predetermined number of iterations [0125] the first optimizer 9 has varied the current correction values s for a predetermined time. [0126] the current correction values s have now only changed insignificantly compared to the preceding iteration. To this end, the term

ss

[0127] can be compared with a predetermined threshold. If said term falls below this threshold, the first optimizer 9 detects a now only insignificant change. [0128] The relationship

f(s)f*+ss0+ss

[0129] as a whole has now only changed insignificantly compared to the preceding iteration. To this end, said relationship can be compared with a predetermined threshold. If said term falls below this threshold, the first optimizer 9 detects a now only significant change.

[0130] Alternatively or additionally, it is also possible for the first optimizer 9 to check whether another termination criterion is met. The decisive factor is that the resulting termination criterion ensures that only a finite number of iterations is performed.

[0131] The most recently determined current correction values s must adhere to the ancillary conditions according to inequality (7) or according to the inequalities (7) and (8). This is not mandatorily the case with the correction values s, which are determined in the meantime and will be further varied later. However, preferably, the first optimizer 9 considers the ancillary conditions on every iteration. This is in particular the case when the first optimizer 9 works in accordance with a continuous optimization method, in particular according to an interior-point method.

[0132] For the final determination of the manipulated variables S, in the embodiment in accordance with FIG. 5, the control device preferably implements an flatness controller 11. The flatness controller 11 is arranged downstream of the first optimizer 9. The current correction values s determined by the first optimizer 9 are supplied to the flatness controller 11. From these, it determines the manipulated variables S for the flatness actuators 8.

[0133] In principle, the flatness controller 11 can be embodied in various ways, for example as a conventional PI controller. However, according to the depiction in FIG. 5, the flatness controller 11 is embodied as controller according to the observer principle.

[0134] Consequently, the flatness controller 11 initially multiplies the current correction values s with a gain factor KP by means of a multiplier 12. The gain factor KP is always positive. It is generally less than 1. If possible, the gain factor KP should be chosen as high as possible. The output signal s of plant model 14 of the roll stand is added to the weighted current correction values determined in this way in a nodal point 13.

[0135] The flatness controller 11 determines a provisional signal S on the basis of the sum formed in this way. Similarly to the manipulated variables S and the correction values s, s, the provisional signal S is a vector. In the simplest case, the provisional signal S is identical to the sum formed. However, generally the sum formed is filtered in a filter 15 for the determination of the provisional signal S. The filter 15 can in particular be embodied as a low-pass filter. It is possible for the filter 15 only to be set in the context of the commissioning of the roll stand. However, preferably, the control device 4 can also re-specify parameters P to the filter 15 at later time point and as a result dynamically parameterize the filter 15. The flatness controller 11 then parameterizes the provisional signal S in a differentiator 16.

[0136] The differentiated signal is then integrated in an integrator 17. The output signal from the integrator 17 corresponds to the manipulated variables S orif the manipulated variables S are obtained as a sum of a plurality of summandsone of the summands. The manipulated variables S are output to the flatness actuators 8. It is possible for the integrator 17 to be part of the flatness controller 11. Alternatively, it can be arranged outside the flatness controller 11.

[0137] The flatness controller 11 supplies the provisional signal S to the plant model 14 of the roll stand as an input signal. The plant model 14 models the effect of the flatness actuators 8 from the point of view of the control device 4. In particular, the plant model 14 models the temporal transition behavior with which measuring device 7 detects an flatness error that has occurred in the roll stand nip. The model parameters required for this can generally be readily determined from the system geometry. This is known to those skilled in the art.

[0138] Therefore, if an actuation y of a specific flatness actuator 8 takes place at the time point x, the plant model 14 reflects which effect of the actuation y is displayed at which time point t in the detection of the measured values. The plant model 14 takes account of the dynamic behavior of the respective flatness actuator 8. The plant model 14 furthermore takes account of any downtimes, for example the transport time, which (viewed in the direction of transport of the metal strip 1) elapses between the action of the respective flatness actuator 8 on a specific point of the metal strip 1 and the detection of the measured flatness values fM for this point by the measuring device 7. The plant model 14 also takes account of any delay times in the detection of the measured values.

[0139] The control device 4 is often able to access models 18 of the roll stand. For example, the models 18 can be integrated in the control device 4. The models 18 model the behavior of the roll stand in operation. The models 18 can, for example, comprise a rolling force model, a bending model, a flattening model, a roll nip model, a model for modelling the thermal and wear-induced camber of rolls 2, 3 of the roll stand and further models. In accordance with the depiction in FIG. 6, at least in the context of the commissioning of the roll stand, the control device 4 preferably retrieves the models 18 in a step S1 and automatically determines the effectiveness matrix W on the basis of the models 18 in a determining device 19. Only then does the rolling of the metal strip 1 or, if applicable, a plurality of metal strips 1, take place in a step S2.

[0140] Preferably steps S3 and S4 are additionally available. In this case, the control device 4 checks in the step S3 whether a new metal strip 1 is to be rolled. If this is the case, the control device 4 proceeds to the step S4. In the step S4as in the step S1the control device retrieves the models 18 and automatically determines the effectiveness matrix W on the basis of the models 18. Therefore, the steps S3 and S4 also implement the dynamic re-determination of the effectiveness matrix W by the control device 4 in each case immediately before the commencement of the rolling of a respective metal strip 1.

[0141] It is even possible for the control device 4 also to track the effectiveness matrix W dynamically during the rolling of the respective metal strip 1. This is also the case with the embodiments explained latter in accordance with FIGS. 7 and 8. If tracking is to take place, in the case of the embodiments in FIGS. 5 and 7, this can be achieved by steps S5 to S7, for example. In this case, the control device determines expected flatness values f1 in the step S5. The expected flatness values f1 can, for example, be determined with the aid of the plant model 14 and/or the models 18 of the roll stand. In the step S6, the control device 4 evaluates the expected flatness values f1 and the measured flatness values fM. On the basis of the evaluation of the step S6, in the step S7, the control device 4 can then track the effectiveness matrix W. In the case of the embodiment in accordance with FIG. 8, the dynamic tracking can be implemented in another way.

[0142] The following explains a modification of the embodiment FIG. 5 in conjunction with FIG. 7. In the context of the embodiment shown in FIG. 7, a further optimizer 9 is present. Therefore, in the following the optimizer 9 is referred to as a second optimizer for differentiation from the first optimizer 9. The second optimizer 9 is constructed in the same way as the first optimizer 9. The following explains the mode of operation of the second optimizer 9 in conjunction with FIG. 7. Reference is made to the fact that although the same variables are used in the context of the following explanation, the variables are independent of the variables used for the first optimizer 9. Therefore, the values can be different even though the same reference characters are used.

[0143] The second optimizer 9 also determines current correction values s by minimizing the relationship

f(s)f*+ss0+ss(9)

[0144] Therefore, the second optimizer 9 varies the current correction values s that are valid for it until it has minimized this relationship. The second optimizer 9 preferably takes account of the same ancillary conditions as the first optimizer 9. Furthermore, the second optimizer 9 preferably also determines the correction values s according to an interior-point method. However, the weighting factors , have the value 0 for the second optimizer 9. Therefore, as a result, the second optimizer 9 optimizes the relationship

f(s)f*(10)

[0145] For this reason, the value of the correction values s of the previous working cycle T is irrelevant for the second optimizer 9.

[0146] Similarly to the case with the first optimizer 9f is a totality of flatness values, i.e. once again a vector. The flatness values f are determined by the optimizer 9 on the basis of the initial flatness values f0 and the valid current correction values s for the second optimizer 9. Similarly to the case with the first optimizer 9the second optimizer 9 determines the flatness values f according to the relationship

f(s)=f0+W.Math.s(11)

[0147] However, with the second optimizer 9 initial flatness values f0 do not correspond to measured flatness values fM, but to a totality of nominal flatness values fW. These in turn correspond to a nominal change FWN in the rolling force FW. As before, W is an effectiveness matrix. It specifies individually what influence a specific individual correction values s has on which of the flatness values f. In accordance with the depiction in FIG. 7, this can in particular involve the same effectiveness matrix W as that used for the first optimizer 9.

[0148] In the case of the embodiment in accordance with FIG. 7, the control device 4 determines the manipulated variables S for the flatness actuators 8 while additionally taking account of an actual change FW in the rolling force FW, the nominal change FWN in the rolling force FW and the current correction values s determined by means of the second optimizer 9. In particular, the control device 4 scales the current correction values s determined by means of the second optimizer 9 with a factor F, wherein the factor F is obtained from the quotient of the actual change FW in the rolling force FW and the nominal change FWN in the rolling force FW:

F=FW/FWN(12)

[0149] If applicable, furthermore, smoothing in a filter can be performedbefore or after the scaling with the factor F.

[0150] As already mentioned, in the case of a software-based implementation, the individual blocks of the internal structure of the control device 4 are program modules. Therefore, in the case of the embodiment in accordance with FIG. 7 it is possible to use one and the same program modulenamely the implementation of the optimizerfor both the first and second optimizer 9, 9. In particular for this reason, the time-delay element 10 is also present in FIG. 7. The two program modules only have to be parameterized individually for the respective intended application. For, example the weighting factors , can be set to values different from 0 for the first optimizer 9 and to 0 for the second optimizer 9, 9.

[0151] In the context of the embodiment shown in FIG. 7, the first and the second optimizer 9, 9, i.e. both the optimizer 9 whose current correction values s are supplied to the flatness controller 11 and the optimizer 9 whose current correction values s are only multiplied with the factor F. However, according to the depiction in FIG. 8, which depicts the second basic embodiment of the present invention it is also possible for only one single optimizer 9 to be present that operates as described above in conjunction with FIG. 7 for the second optimizer 9. Therefore, in this case, the optimizer 9 whose current correction values s are only multiplied with the factor F is the sole optimizer and hence the first optimizer 9 within the meaning of the present invention.

[0152] An independent separate explanation of the mode of operation of the first optimizer 9 in FIG. 8 is not necessary here since the first optimizer 9 in FIG. 8 operates in exactly the same way as the second optimizer 9 in FIG. 7.

[0153] The present invention has numerous advantages. The online-optimization enables optimum manipulated variables S to be determined in each working cycle T. Due to the embodiment of the flatness controller 11 in accordance with the observer principle, it is able to react as quickly as possible to disturbing influences or other changes. The models 14, 18 required for the implementation of the present invention are usually already present in the control device 4. As a result, there is no additional expenditure on the creation of such models 14, 18. The filter 15 enables efficient adaptation of the flatness controller 11 to the response to the interference specifically for the respective system. Furthermore, separation into the first optimizer 9 on the one hand and the flatness controller 11 on the other causes the flatness error analysis (which takes place in the first optimizer 9) to be decoupled from the dynamic behavior of the controller (which takes place in the flatness controller 11). This results in a simple, modular configuration of the control device 4 for virtually any type of stand. This simplifies parameterization of the control device 4 in the engineering phase and reduces commissioning times. Furthermore, system-specific adaptations can be made in a targeted and simple manner. Furthermore, decoupling simplifies implementation and thus errors are avoided in both the engineering phase and the commissioning phase. Optimum control extends the lifetime of the mechanical components of the roll stand. Compensation of flatness errors resulting from changes in the rolling force FW by precontrol further increases productivity due to very quick reaction to faults. The economic efficiency of the operation is increased because there are fewer rejects. Furthermore, the optimization takes account of the setting limits of the flatness actuators 8 in the context of the determination of the manipulated variables S.

[0154] Although the invention has been illustrated and described in greater detail by the preferred exemplary embodiment, the invention is not restricted by the disclosed examples and other variants can be derived herefrom by the those skilled in the art without departing from the scope of protection of the invention.

LIST OF REFERENCE CHARACTERS

[0155] 1 Metal strip [0156] 2 Working rolls [0157] 3 Support rolls [0158] 4 Control device [0159] 5 Computer program [0160] 6 Machine code [0161] 7 Measuring device [0162] 8 Flatness members [0163] 9, 9 Optimizer [0164] 10, 10 Time-delay elements [0165] 11 Flatness controller [0166] 12 Multiplier [0167] 13 Nodal point [0168] 14 Plant model [0169] 15 Filter [0170] 16 Differentiator [0171] 17 Integrator [0172] 18 Models of the roll stand [0173] 19 Determining device [0174] A Axial displacement [0175] B, c Vectors [0176] C Matrix [0177] B Width [0178] F Factor [0179] F Flatness values [0180] f0 Initial flatness values [0181] fM Measured flatness values [0182] fW Nominal flatness values [0183] f* Flatness target values [0184] FR Reverse bending forces [0185] FW Rolling force [0186] FWN Nominal change in the rolling force [0187] K Local cooling [0188] KP Gain factor [0189] P Parameter [0190] S Manipulated variables [0191] S Provisional signal [0192] S1 to S7 Steps [0193] s Correction values for the current working cycle [0194] s Correction values for the previous working cycle [0195] s Output signal from the model of the roll stand [0196] T Working cycle [0197] W Effectiveness matrix [0198] , Weighting factors [0199] FW Actual change in the rolling force