JPH0618649B2 - Rolling mill rolling schedule setting method - Google Patents

Description

TECHNICAL FIELD The present invention relates to rolling of a metal material by a plurality of rollings (passes), and particularly to setting of a rolling schedule for a plurality of passes.

The multiple rolling means that the rolled material literally undergoes multiple rolling, and when the material to be rolled is conveyed in one direction and sequentially rolled by the tandem rolling mill, it can be rolled by the reverse type rolling mill. It includes the case where rolling is performed a plurality of times and the case where these are combined.

[Conventional technology]

Conventionally, the power and load required for rolling are applied to each rolling mill (each pass).
Of the rolling schedule that distributes the power and load required for rolling in each pass by considering the crown and shape of the rolled plate with a calculation formula (Japanese Patent Laid-Open No. 59- No. 73108) has been proposed.
However, there is still a large amount of operator intervention from the viewpoint of model accuracy that describes the crown / shape. Further, the amount of change in the rolling schedule for each line is large, which is an obstacle to stable operation. In the realization method, the method that considers the crown shape requires a large amount of calculation for the optimum path search and becomes a factor that hinders productivity. Uncertain factors such as strip stability cannot be measured, and it is difficult to automatically reflect them in the rolling schedule from the next time.

From the above viewpoints, in the actual operation, the rolling schedule has been set corresponding to the rolled material type, and this rolling reduction schedule (setting rolling schedule) is changed or adjusted by the operator intervening in the actual operation. The actual reduction schedule) is reflected by methods such as exponential smoothing. That is, the set rolling schedule is corrected by the learning process for each rolled material classification.

[Problems to be Solved by the Invention]

This rolling schedule is set for each rolling material classification,
In the rolling schedule setting method that learns and corrects this based on actual results, if the number of categories of rolled material types is small, rolling materials with relatively large type differences within the same category are rolled with the same rolling schedule. Therefore, the modification of the rolling schedule by learning is suitable for the rolling material type with the highest execution frequency, and there is a large difference with rolling materials that belong to the same rolling material type category but have high execution frequency. For example, the rolling accuracy of rolled materials that are not frequently executed becomes low.
Accuracy is apt to be insufficient and stable operation is hindered. If the number of categories of rolled material is increased, the variation of rolling frequency for each category will be large and the variation of accuracy for each category will also be large. Therefore, the rolling tends to occur with a large difference in accuracy, which also hinders stable operation.

An object of the present invention is to improve the stable operation by reducing the variation of rolling accuracy for each type of rolling and the variation of rolling accuracy due to crossing of categories.

[Means for Solving the Problems]

In the present invention, (a) for each material to be rolled, each group conformity (Fg; g = 1 to 16) representing the degree of conformity to each of the rolling type groups (Rg; g = 1 to 16) And (b) each rolling load distribution calculation formula [(7) formula] for each group (Rg), each rolling pass (i; i = 1 to N) of the material to be rolled
Of the rolling load distribution (Γg, i) corresponding to each group (Rg) in (c), (c) The rolling load distribution (Γg, i) calculated for each group corresponding to each group (Fg) is used as a weight. The target rolling load distribution (Γi) of each pass (i) of the material to be rolled is calculated by weighted averaging, and the rolling schedule is calculated based on the target rolling load distribution (Γi) of each pass (i). The material to be rolled is rolled on the basis of the standard, and (d) the actual reduction schedule of actually rolling the material to be rolled is inversely converted into the actual rolling load distribution ( Γi ) of each pass (i), and Target rolling load distribution for path (i) (Γi), actual rolling load distribution for each path (i) ( Γ
i ) and the fitness of each group (Fg), the learning parameters (Ag, i, k; k = 0 to 4) of each rolling load distribution calculation formula [equation (7)] addressed to each group are described above. Each group fitness (Fg) is weighted and corrected.

In addition, (a) ~ (d) added to the beginning of the above processing items are processing item display symbols added here, and symbols in parentheses in the processing items correspond to those used in Examples described later. Is a symbol.

[Action]

The group suitability (Fg) in (a) above indicates the suitability of the material to be rolled to each of the rolling type groups.
Conventionally, for each rolled material, it has been determined that they fit into one group and not into another group, but according to (a) above, such a binary group dependency No decision is made, and which group the material to be rolled belongs (conforms to) is not specified in one group, and the set of conformity (Fg; g = 1 to 16) to each group (conformance to the group) Distribution).

Then, according to the above (b), the material to be rolled is grouped (Rg)
Of the rolling load distribution (Γg, i) corresponding to each group (Rg) in the case of rolling assuming that the rolling load distribution corresponds to each group (Rg), the rolling load distribution corresponding to each group (Rg) is calculated by the above (c). Since the target rolling load distribution (Γi) is calculated by weighted averaging (Γg, i) with each group fitness (Fg) calculated in (a) above as a weight, the target rolling load distribution (Γi)
Indicates that the target rolling load distribution (Γx, i) has a high contribution rate (Fx) calculated by the load distribution calculation formula addressed to the group (Rx) to which the rolling target material has a high degree of conformity, and the rolling target material is suitable. The degree of contribution (Fz) of the target rolling load distribution (Γz, i) calculated by the load distribution calculation formula addressed to the group (Rz) having a low degree of use becomes low.

That is, the target rolling load distribution (Γi) is not uniquely determined by the target load distribution (eg, Γz, i) defined in one group, but a group (Rx) having a high degree of conformity with the material to be rolled. The target load distribution (Γg, i) determined in each group from the group to the group (Rz) having a low degree of conformity is weighted by the group conformity of the material to be rolled and smoothed over each group. As a result, variations in rolling accuracy due to the above-mentioned conventional crossing of sections are reduced.

Further, by the above (d), the target rolling load distribution (Γi) of each pass (i) of the material to be rolled, the actual rolling load distribution ( Γi ) of each pass (i), and the group suitability (Fg) Based on the above, the learning parameters (Ag, i, k; k = 0-4) of each rolling load distribution calculation formula [equation (7)] addressed to each group are weighted and corrected by each group fitness (Fg). Therefore, according to the rolling results of one rolling target material, the rolling load distribution calculation formulas of substantially all groups are weighted by the group suitability (Fg), learned and corrected, and the rolling target material that has actually been rolled The rolling load distribution formula addressed to the group with a high degree of conformance was learned and modified by reflecting the rolling results of this time at a relatively large ratio, and the rolling load distribution formula addressed to the group with a low degree of conformity was this time. This is a learning correction in which the rolling results of the above are reflected at a relatively low ratio. As a result, the difference between the rolling results for each group and the learning correction based thereon is reduced, and the difference in rolling accuracy between groups is reduced. Therefore, variations in rolling accuracy due to the above-mentioned conventional crossing of sections are further reduced.

As a result of the above, according to the present invention, the difference in rolling precision due to the difference in rolling type of the material to be rolled is reduced, and the variation in rolling precision due to the crossing of sections does not substantially occur.

Other objects and features of the present invention will become apparent from the following description of embodiments with reference to the drawings.

〔Example〕

In one embodiment of the present invention, first, rolled material type (group)
The classification is as follows: steel sheet final target thickness h: thick / thin, steel sheet width W
4 parameters (h, W, C, high / low of hardness C of steel plate and thick / thin of thickness H of steel plate).
H) is divided into two stages, and 2 ⁴ = 16 groups. This group is thick / thin, large / small,
Group division indicators such as high / low and thick / thin are ambiguous expressions.

Taking the target thickness h as an example, “target thickness is thin” is expressed by the degree Sh (0 to 1) as shown in (1) and (2) below, and “target thickness is thick”. Is expressed by the degree Bh (0 to 1).

Target thickness h is thin: Sh = 1 (h ≦ h ₁ ) = (hh ₁ ) / (h ₂ −h ₁ ) (h ₁ <h ≦ h ₂ ) = 0 (h ₂ <h) ・ ・ ・ (1 ) Target thickness h is thin: Bh = 0 (h ≦ h ₁ ) = (hh ₂ ) / (h ₁ -h ₂ ) (h ₁ <h ≦ h ₂ ) = 1 (h ₂ <h) ... ( 2) h ₁ and h ₂ are values that delimit the target plate thickness between “thin” and “thick”, respectively. Here, the normal boundary value that divides the rolled material into “thin material”, “normal material”, and “thick material” is used, and h ₁ is the boundary value between normal “thin material” and “normal material”, h 1 ₂
Is a boundary value between “normal material” and “thick material”, and has a relationship of “thin material” <h ₁ ≦ “normal material” ≦ h ₂ <“thick material”.

Similarly, the steel plate width W is narrow degree S _W, a wide degree B _W, hardness C higher degree S _C of the steel sheet, low degree B _C, and the degree original plate thickness H of the steel plate is thick S _H, thin degree B Express _H.

The degree of conformity of a certain rolling material to each of the 16 groups, that is, the group conformity is Fg; g = 1 to 1
If 6, then each group suitability Fg of the material to be rolled is calculated as follows.

For example, from h, W, C and H of the material to be rolled, the above (1)
Degree Sh based on equation like, calculates S _W, S _C and S _H, when determining these minimum values, the minimum value denote the first group fitness F ₁ of the rolling target material.

F ₁ indicates the degree to which the material to be rolled conforms to the first group in which the target thickness h is thin, the width W is narrow, the hardness C is low, and the thickness H of the original plate is thin, that is, the first group compatibility.

F ₂ indicates the degree of conforming to the second group, that is, the degree of conforming to the second group, that is, the target thickness is thin, the width is narrow, the hardness is low, and the thickness of the original plate is thick.

F ₃ indicates the degree of conforming to the third group, that is, the degree of conformity of the third group, that is, the target thickness is thin, the width is narrow, the hardness is high, and the thickness of the original plate is thin.

.. F ₁₆ indicates the degree of conforming to the 16th group, that is, the target thickness is thick, the width is wide, the hardness is high, and the thickness of the original plate is thick, that is, the degree of conforming to the 16th group.

In this manner F ₁ is a first group, the F ₂ is the second group, the is F ₃ a third group, ..., that F ₁₆ represents the degree conforming to the 16 groups, group fitness It is based on the above definition of Fg, g = 1 to 16.

In this way, the first to sixteenth group suitability F _{1 to} F ₁₆ are calculated for each material to be rolled.

Next, the rolling schedule calculation formula for the group will be described.

The rolling schedule calculation parameter needs to have a clear relationship between the rolling reduction amount and the required capacity of each pass i, i = 1 to N. Therefore, the logarithmic rolling reduction represented by the following formula (4), which has a linear correlation with the rolling load, is used. The ratio Γi is used.

Γi = log (hi / Hi) / log (h / H) (4) where hi: i-pass entry side thickness, Hi: i-pass entry side thickness, h: target thickness, H: original plate Thickness and N: number of passes.

The logarithmic reduction ratio Γi has the following properties and can be easily converted into the pass outlet plate thickness (reduction schedule).

Hereinafter, Γi is called rolling load distribution (load balance).

Next, it is necessary to have a load balance calculation formula (rolling load distribution calculation formula) for each group, but it is not necessary to use a non-linear formula. This is because the non-linear part can be expressed by the group division by the ambiguous expression and each group fitness Fg, g = 1 to 16, and when the reduction schedule is uniform in the group, the schedule may have a fixed value. It is possible. In this embodiment, the grouping parameters (h,
W, C, H) linear regression equation
The load balance Γg, i of the i-th pass of group g is
It is calculated by equation (7).

Γg, i ＝ Ag, i, ₀ + h ・ Ag, i, ₁ + W ・ Ag, i, ₂ + C ・ Ag, i, ₃ + H ・ Ag, i, ₄
(7) Here, Ag, i, k; k = 0 to 4 are coefficients of the regression equation (coefficients corrected by learning).

The load balance Γg, i of the group g (g = 1 to 16) obtained by the equation (7) is weighted and averaged with the group suitability Fg of the rolling target material of the group g being used as a weight to calculate i of the rolling target material. The target load balance Γi of the pass is calculated.

The rolling schedule is determined using this target load balance Γi and Eq. (6).

On the basis of the reduction schedule determined in this way, the material to be subjected to the above-described calculation is rolled. In this rolling, there may be intervention (change or modification of the rolling schedule) depending on the operator's judgment.

As described above, the reduction schedule independent of only a single group (single rolled material classification) is calculated and provided for rolling. Next, a method for learning and correcting the coefficient Ag, i, k of the above load balance calculation formula (7) to reflect the operator's judgment (intervention: actual result) in the actual rolling in the subsequent rolling schedule calculation. To do.

At the time of actual end of rolling, the actual rolling reduction schedule with operator's judgment is converted back to the load balance Γi . The underline means that it is a track record. Γi = log ( hi / Hi ) / log ( h / H ) (9) Note that this is the rolling schedule in Equation (4) replaced with the actual value.

Based on this actual load balance Γi , the coefficient A of the load balance calculation formula (7) for the group with a goodness of fit of a certain value or more is determined according to the goodness of fit Fg of each group of the material to be rolled which has obtained this result.
Modify g, i, k. In this embodiment, the weighted recursive least squares method is used in consideration of convergence, stability, calculation amount, and the like. If a vector expression is used to simplify the description of the calculation formula, the formula (7) is expressed by the following formula (10).

Γg, i = Ag, iX ... (10) Ag, i = (Ag, i, ₀ , Ag, i, ₁ , Ag, i, ₂ , Ag, i, ₃ , Ag, i, ₄ ) ^T X = (1, h, W, C, H) ^T Note that the double-width character represents a vector, and ^T represents a transposed matrix.

In this formula (10), the coefficient Ag, i of each pass i in the rolling load distribution calculation formula (7) of the group having a goodness of fit Fg of a certain value (for example, 0.01) or more is corrected and updated.

The coefficient Ag, i and the load balance Γi multiplied by the fitness Fg as a weight are multiplied by F to the tail of the symbols.
It is indicated by adding. Ag, iF = Fg × Ag, i ・ ・ ・ (11-1) X F = Fg × X ・ ・ ・ (11-2) Γg, iF = Fg × Γg, i ・ ・ ・ (11-3) Γg , I F = Fg × Γi (11-4) Coefficient updating is performed using the recursive least squares method (12-1) to (12-4). The number of conforming groups x number of passes is calculated as follows A g, iFA = Ag, iF + Kg, i (Γg, i F-Γg, iF) ·
.. (12-1) Kg, i = Pg, iXF (1 + XF ^T Pg, iXF) ^-1 ... (12-
2) Pg, iA = (1−Kg, iXF ^T ) Pg, i / λg, i (12-3) λg, i = 1−G ( Γg, i F−Γg, iF) ² (1 + X ^T Pg, iX
F) ^-1 ... (12-4) where Kg, i: modified gain vector (5 × 1), Pg, i: error covariance matrix (5 × 5), λg, i: forgetting factor, and , G: adjustment coefficient.

The coefficient after updating was represented by adding A to the end of the symbol. After updating, these are stored for use in the next calculation and learning calculation.

Pg, i includes past data for subsequent calculations of equations (12-2) to (12-4), and the forgetting factor λg, i is the reflection rate of the latest data when the past data becomes huge. The adjustment coefficient G is a coefficient for adjusting excessively past data to a reasonably small amount in order to prevent the load balance tracking accuracy from becoming low and deteriorating, and the adjustment coefficient G is a coefficient for adjusting the reflection rate.

FIG. 5 shows the structure of a hot tandem rolling line for carrying out the above-described embodiment of the present invention. Here, the material 1 to be rolled is sent from left to right and sequentially rolled by seven tandem rolling mills 2. The actual rolling load of each rolling mill is detected by load detectors (load cells) 3-1 to 3-7. The timing of load detection is 1-2 seconds after biting for each rolling mill. From the detected load, the actual load balance calculation device 4 calculates the actual plate thickness in consideration of the mill elongation and the like to calculate the actual load balance Γi . Model coefficient updating device 5
However, Ag, iFA, Pg, and iA are calculated from (11-1) to (11-4) from the goodness of fit Fg calculated by the ambiguous expression conversion device 10 and the coefficient placed in the actual load balance and storage device 6. , (12-1) ~ (12-
It is calculated by the equation (4), and these are updated and stored in the storage device 6.

The reduction schedule calculation device 7 is calculated by the fuzzy expression conversion device 10 from the coefficient Ag, i stored in the storage device 6 and the rolled material type information (h, W, C, H) input by the input device 9. Calculate the reduction schedule based on the goodness of fit Fg.
The reduction gap calculation device 8 calculates the roll gap to be set based on this, and outputs it to the rolling mill gap setting device 11.

FIG. 1 shows an arithmetic processing process for one rolling in the rolling line shown in FIG. The calculation process will be described. First, when a strip preparation instruction is given, the ambiguous expression conversion device 10 converts the rolled material information (h, W, C, H) input by the input device 9 into (1) to (). 3) Formula Fg for each group
And the rolling coefficient data vector X and the latest coefficient vector Ag, i written in the memory 6, the rolling reduction schedule calculation device 7 calculates the rolling reduction schedule by the equations (7) and (8). Steps 1-5).

Next, the reduction gap calculation device 8 calculates the roll gaps to be set in each rolling mill based on the calculated reduction schedule (step 6), and the rolling mill gap setting devices 11-1 to 11-7 calculate these roll gaps. Set (step 7). When the rolling is started, the detectors 3-1 to 3-7 give the actual load balance calculation device 4 a rolling load detection value 1 to 2 seconds after the rolled material is caught in each rolling mill (step 8, 9).

The actual load balance calculation device 4 calculates the actual plate thickness from the rolling load detection value and the set roll gap, calculates the actual load balance, and gives this to the model coefficient updating device 5 (step 10).

The model coefficient updating device 5 uses the calculated value Γg, i, the actual value Γi, and the group suitability Fg to perform the above (11-1) to (11-4), (12-1).
~ In the equation (12-4), the coefficient vector Ag, i and the operation data Pg,
i is calculated, and these are updated and written in the memory 6 (steps 11 and 12).

[Implementation result]

A method that has a fixed schedule according to the type of rolled material and that smooths the operator's intervention exponentially (table lookup: a conventional method) and a method according to the present invention (fuzzy adaptive control: the method of the present invention) Was tested and run online to compare actual reduction schedules.

The actual rolling load distribution (load balance) relating to the strip width and strip thickness of the material to be rolled is shown in FIG. 2a for the method of the present invention and in FIG. 2b for the conventional method. Conventional method (2b
In the figure), the rolling load distribution changes relatively large and abruptly with respect to the relatively small difference between the strip width and the strip thickness, and the unevenness due to the variation in the progress of learning for each rolling material type classification is large. As described above, stable rolling is difficult because of large variations in rolling accuracy and large differences in accuracy due to division straddling, but according to the method of the present invention (Fig. 2a), rolling load with respect to difference in strip width and strip thickness is difficult. Since the change in distribution is small and changes smoothly, and the learning correction is smoothly distributed and unevenness is smoothed, the variation in rolling accuracy is small, and the difference in accuracy due to section straddling is extremely small, and stable operation is easy. As a result, as shown in FIG. 3b, the stability of the load balance setting during repeated rolling is low in the conventional method, but the stability of the load balance setting is high in the method of the present invention as shown in FIG. 3a. . The amount of operator intervention is shown in FIG. 4 as an index of stable stripability. One point (black circle: conventional method, white circle: method of the present invention) is a unit of rolling about 40 to 100 coils, and the degree of operator intervention (frequency × amount) may be halved by applying the present invention. I understand.

〔The invention's effect〕

As described above, according to the rolling schedule setting method of the present invention, to which group the material to be rolled belongs (conforms) is a set of conformity (Fg; g = 1 to 16) to each group (for the group). The distribution of the goodness of fit), the target rolling load distribution (Γi) of the material to be rolled is the goodness of fit (Fg) of each group.
With the weight as the weight, rolling load distribution (Γg,
i) is a weighted average, so rolling type (h, W, C, H)
Has a high degree of similarity in the distribution of rolling load between rolling target materials, which reduces variations in rolling accuracy due to conventional cross-section.

Further, based on the target rolling load distribution (Γi) of each pass (i) of the material to be rolled, the actual rolling load distribution ( Γi ) of each pass (i), and each group conformity (Fg), Learning parameters (Ag, i,
k; k = 0 to 4) is weighted by each group fitness (Fg) and learned and corrected, so that the rolling load distribution applied to the rolling target material with the highest weight according to the rolling record of one rolling target material. Not only the calculation formulas but also other rolling load distribution calculation formulas applied with high weight to the rolling target material having a high similarity to the rolling target material are learned and modified with considerable weighting to determine the strip running stability and strip running shape. The intervention based on the operator's judgment in consideration is efficiently, smoothly and stably reflected in the learning, which reduces the difference between the rolling results of each group and the learning correction based on it, and reduces the difference in rolling accuracy between groups. The amount of operator intervention is reduced. Variations in rolling accuracy due to conventional crossing of sections are further reduced.

As a result of the above, according to the present invention, the difference in rolling accuracy due to the difference in rolling type of the material to be rolled is reduced, the intervention amount of the operator is reduced, and the variation in rolling accuracy due to the division straddling does not substantially occur. Through-passability is improved, product accuracy is stabilized, and productivity is improved.

[Brief description of drawings]

FIG. 1 is a flowchart showing an execution process of setting a rolling reduction schedule according to an embodiment of the present invention. 2a and 2b show constant hardness (carbon equivalent 1
0) is a graph showing the relationship between the target thickness and width of the seventh pass (7th stand) and the load balance in rolling with a constant original plate thickness (32 mm), and FIG. FIG. 2b shows a conventional method. 3a and 3b show all paths (1 to 1) for each coil.
6 stand) is a graph showing the load balance,
FIG. 3a shows the one in which the present invention was carried out, and FIG. 3b shows the one in which the conventional method was carried out. FIG. 4 is a graph showing the relationship between the target thickness change amount (average change amount (μ)) for each rolling unit and the operator's intervention amount (frequency × amount), and the white circles in the figure embody the present invention. The black circles indicate the ones obtained by the conventional method. FIG. 5 is a block diagram showing a configuration outline of a rolling line for carrying out the present invention in one mode. 1 ... Rolled material (rolling target material), 2 ... tandem rolling mill (7 units) 3-1 to 3-7 ... Load detector, 4 ... Actual load balance calculation device 5 ... Model coefficient updating device, 6 ... Storage device 7 ... Reduction schedule calculation device, 8 ... Reduction gap calculation device 9 ... Input device, 10 ... Ambiguous expression conversion device 11-1 to 11-7 ... Rolling mill gap setting device

Claims (1)

[Claims] 1. A rolling target material is calculated by calculating a degree of conformity of each group, which represents a degree of conformity with each rolling type group, and using each rolling load distribution calculation formula for each rolling type group. Rolling load distribution for each group of each rolling pass is calculated, and the rolling load distribution calculated for each group is weighted and averaged using each group conformity as a weight, and the target rolling load for each pass of the rolling target material is calculated. The distribution is calculated and a rolling schedule is calculated based on the target rolling load distribution of each pass, and the rolling target material is rolled based on this, and the actual rolling reduction schedule of the actual rolling of the rolling target material is performed for each pass. Based on the target rolling load distribution of each of the passes of the material to be rolled, the actual rolling load distribution of each of the passes, and the degree of conformity of each group, which is converted back to the actual rolling load distribution.
A rolling mill rolling schedule setting method, in which the learning parameters of each rolling load distribution calculation formula addressed to each group are weighted and corrected by the group suitability.