CN102658297B - Self-learning method for improving quality of first band steel plate shape with changed specification - Google Patents

Abstract

A self-study method for improving the shape quality of the first strip steel plate for gauge change. The method includes the steps of: firstly classifying the steel type, width and thickness of the strip steel, and calculating the layer number of the strip steel; The learning calculation value is saved in the corresponding layer number; after the roll is changed, the long-term self-learning calculation of the strip shape is performed, and the learning value is stored in the corresponding layer, and the short-term self-learning is cleared; any strip rolling Before, judge whether the self-study inheritance condition is satisfied, if the condition is met, read the inherited self-study value of the last piece of strip steel; before rolling any strip steel, take the short-term shape self-study amount and long-term self-study amount of the corresponding layer; Combining the three self-learning values to carry out the setting calculation of the shape model. The present invention can remarkably improve the quality control precision of the first roll of the specification change through the combined effects of short-term self-study, long-term self-study and inherited self-study.

Description

A Self-learning Method for Improving the Shape Quality of the First Strip for Changing Specifications

Technical field:

The invention relates to a hot-rolled strip steel thin plate rolling process technology, in particular to a self-learning method for improving the shape quality of the first strip steel plate for changing specifications.

Background technique:

With the development of industrial technology, the use of hot-rolled strips is becoming more and more extensive, ranging from automobiles, electronics, home appliances to energy, aviation, mining and other fields. The dimensional accuracy control of hot-rolled strips is also particularly important . As an important part of the control of the dimensional accuracy of the strip, the shape of the plate is more and more valued by iron and steel enterprises. The production mechanism of the shape of the plate is very complicated. A unique shape control system is also essential. Since the development of flatness control technology, the simple flatness problem has been overcome. At present, there is still a difficult problem in the flatness control system. Common deficiencies in the rest of the control system in quality control.

The shape control model is divided into two parts, one part is the setting calculation model, and the other is the dynamic closed-loop control model. The setting calculation model is mainly to ensure the shape quality of the strip head (the shape quality of the strip has not been detected by the shape meter) , while the dynamic closed-loop control model mainly ensures the shape quality of the full length after the meter detects the shape value. Similar to other control models, the setting calculation model generally adopts a theoretical model or an empirical model. Due to factors such as changes in system characteristics, sub-model forecast errors, and environmental changes, the forecast accuracy is limited. For this reason, a self-learning algorithm is introduced. After detecting the strip shape value of the current strip, correct the set value of the next strip through the self-learning system to ensure the quality of the head shape of the next strip. From the above, it can be seen that effective self-learning The algorithm has positive significance to the improvement of the precision of shape quality control.

Work roll bending and roll shifting are the main means to improve the strip shape in hot continuous rolling. Taking the calculation of roll bending as an example, the calculation model structure of roll bending can be expressed as:

B _f ＝(C _m -k _p Pk _WC C _WC -k _WE C _WE -k _BC C _BC -k _BE C _BE -k _CWR C _CWR -C)/k _f

In the formula, B _f is the set value of bending force, C _m needs to control the strip shape target, P is the set value of rolling force; C _WC is the comprehensive roll shape radius value in the middle of the work roll body; C _WE is C _BC is the comprehensive roll shape radius value of the middle part of the back-up roll body; C _BE is the comprehensive roll shape radius value of the back-up roll body side; C _CWR is the initial roll shape of the work roll _; k _p is the influence coefficient of rolling force; k _f is the influence coefficient of bending force; k _WC is the influence coefficient of roll shape in the middle of work roll; k _WE is the influence coefficient of roll shape in the edge of work roll; Influence coefficient; k _BE is the influence coefficient of the roll shape of the backup roll; k _CWR is the influence coefficient of the initial roll shape of the work roll, and C is the self-learning item. Through the self-learning algorithm, C is corrected in real time to achieve the effect of accurately calculating the bending force.

If the specifications of the rolled strips are the same, the expected purpose can be achieved through a simple self-learning algorithm. As shown in Figure 1, the specifications of the strips from i to i+4 are the same, and the self-learning value of each piece can be Effectively apply to the next strip. However, if the specification of strip steel changes, as shown in Figure 2 (thickness of i+2 strips changes) and Figure 3 (width of i+2 strips changes), due to the different rolling characteristics of various strips, then i+ After the rolling of 1 piece of strip steel is completed, the calculated self-learning amount may not be applicable to i+2 pieces of strip steel, and there may even be cases where the learning value makes the quality of the plate shape worse and worse. However, if the first piece of specification is changed For the self-learning value, due to the inherent error of the model, the setting accuracy cannot be effectively guaranteed.

There are many research results on the self-learning method of the strip rolling model, but there are few documents on how to effectively use the rolling model self-learning value between the strips, such as document 1 (model and control of strip hot rolling, metallurgical Industrial Press, 2002) mentioned the growth memory recursive least squares method and exponential smoothing method, but did not mention how to apply the calculated self-learning value to the subsequent rolled strip; the rolling model self-learning value There are few documents on how to effectively use strips, such as document 2 (Development and application of rapid self-learning of plate shape for mixed rolling of stainless steel and carbon steel, 2006 National Steel Rolling Production Technology Conference Proceedings) mentioning the consideration of strip properties Layer refinement is carried out to solve the plate shape problem of mixed rolling of stainless steel and carbon steel, but it does not consider the self-learning algorithm of adjacent strips in different layers and the learning method after changing rolls in the same layer; such as literature 3( The self-learning algorithm optimization of the hot-rolled strip rolling force model, Beijing University of Science and Technology Journal, V32, No6) also mentions the use of layer refinement and judging the variation coefficient of thickness and width between strips and strips. The literature has a certain practicability, if the steel strength changes between adjacent strips and the self-learning of the same layer after the roll is changed, there are deficiencies; for example, literature 4 (Design and application of flatness control system , Proceedings of the 29th China Control Conference) mentioned the combined effect of short-term self-learning and long-term self-learning, but did not mention the use of strip shape self-learning for changing specifications.

Invention content:

Through the above technical background and literature analysis, it can be seen that the existing self-learning methods cannot effectively guarantee the flatness of the first coil of the regular plan arrangement, width cross-rolling, and cross-rolling of steel types, resulting in defects in the quality control of the flatness . In order to solve the above problems, the present invention relates to a plate shape self-learning strategy. Through the combination of short-term self-learning, long-term self-learning and inherited self-learning, the purpose of improving the efficiency of self-learning is achieved, and the first block of changing specifications under various working conditions is improved. Strip shape quality. The specific implementation steps are as follows:

(a) First classify the steel type, width, and thickness of the strip steel, and determine the layer number; the calculation of the layer number is related to the steel type, width, and thickness of the strip steel. It is stipulated that the steel type has a total of GNum files, and the width has a total of BNum files, the thickness has a total of HNum files. Then the total number of layers is: GNum×BNum×HNum. If the gear of a certain strip steel grade is iG, the width gear is iB, and the thickness gear is iH, the calculation method of the layer number N is:

N=iG×BNum×HNum+iB×HNum+iH+1

Taking the 1780mm hot rolling mill as an example, the steel grade iG is divided according to the strength of the strip steel. Define the strip yield strength σ _s <= 200Mpa to define iG=0, 200<σ _s <=300Mpa to define iG=1, and so on. The width gear iB is divided according to the strip width, if the product outline width range is 800mm-1700mm, when defining the strip width B<=800mm, iB=0; 800<B<=900mm, iB=1, and so on. The thickness level iH is divided according to the thickness of the steel strip. If the thickness range of the product outline is 1.2mm-18mm, when the defined thickness h<=1.2mm, iH=0; 1.2<H<=2.0mm, iH=1, and so on;

In the computer memory, GNum×BNum×HNum records are opened, and each record stores the short-term self-learning and long-term self-learning values of the strip shape. The program stores or reads through the layer number N index of the strip.

(b) After the rolling of any strip steel is completed, short-term self-learning calculation of strip shape is carried out, and the calculated value is saved in the corresponding layer number; for any layer number, the short-term self-learning of the shape adopts the following algorithm:

$\mathrm{<span\; class="notranslate">\; \&Delta;</span>}{\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; the\; s</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; \&Delta;</span>}{\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; the\; s</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; \&Delta;</span>}{\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; the\; s</span>}}^{\mathrm{<span\; class="notranslate">\; R</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; \&Delta;</span>}{\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; the\; s</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span>$

(i)为根据第i块带钢板形实测值反算出来的板形短期自学习调整系数；k₁为板形短期自学习增益系数，取值范围为0.3-0.7。In the above formula: ΔC _s (i+1) is the short-term self-study amount of the i+1 strip steel plate with the same layer number; ΔC _s (i) is the short-term self-study of the i-th strip steel plate with the same layer number quantity;

(i) is the short-term self-learning adjustment coefficient of strip shape back-calculated based on the measured value of the i-th steel strip; k ₁ is the short-term self-learning gain coefficient of the strip shape, and the value range is 0.3-0.7.

(c) After the roll change, carry out the long-term self-study calculation of the strip shape, store the learned value in the corresponding layer, and clear the short-term self-study; Long-term self-study of the plate shape is carried out for all layer numbers. For any layer number, the following algorithm is adopted:

ΔC _l (j+1)＝ΔC _l (j)+k ₂ ΔC _s (j)

In the above formula: ΔC _l (j+1) is the long-term self-learning amount of the steel plate shape of the j+1th roll changing unit with the same layer number; ΔC _l (j) is the strip steel plate of the jth roll changing unit with the same layer number ΔC _s (j) is the short-term self-study amount of strip shape according to unit j; k ₂ is the gain coefficient of long-term self-study of strip shape, and the value range is 0.3-0.7. After the long-term self-study is completed, the short-term self-study amount ΔC _s (j) of the j roll changing unit is cleared.

(d) Before any strip rolling, judge whether the inherited self-learning meets the conditions. If the conditions are met, take the inherited self-learning value; after the arbitrary strip rolling is completed, calculate the inherited self-learning amount. The formula used is:

$\mathrm{<span\; class="notranslate">\; \&Delta;</span>}{\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; a</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; 3</span>}}\mathrm{<span\; class="notranslate">\; \&Delta;</span>}{\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; the\; s</span>}}^{\mathrm{<span\; class="notranslate">\; R</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span>$

为根据第i块带钢板形实测值反算出来的板形短期自学习调整系数；k₃为板形短期自学习增益系数，取值范围为0.3-0.7。其中，i和i+1块带钢层别号并不一定相同，若满足继承条件，则ΔC_a(i+1)对i+1块带钢起作用，否则不起作用。In the above formula: ΔC _a (i+1) is the shape inheritance self-learning amount of the i+1th strip steel plate;

is the short-term self-learning adjustment coefficient of strip shape based on the measured value of the i-th steel strip; k ₃ is the short-term self-learning gain coefficient of the strip shape, and the value range is 0.3-0.7. Wherein, i and i+1 steel strip layer numbers are not necessarily the same, if the succession condition is satisfied, then ΔC _a (i+1) works on i+1 steel strip, otherwise it does not work.

Whether the inheritance conditions are satisfied is judged as follows:

1) The strip steel for judging the inheritance conditions must meet the following requirements: the judgment starts from the second piece of steel strip after roll change and start rolling; the specification of the strip steel has changed; the strip steel has not been rolled during the roll change cycle .

2) If the front and back strips iG and iB are equal, and the difference in the number N of the front and back strip layers is within the range of [-2, +2], the inheritance condition is met;

3) If the front and rear strips iB and iH are equal, and the difference N of the front and rear strip layers is within the range of [BNum×HNum-2, BNum×HNum+2], the inheritance condition is satisfied;

4) If the front and rear iG and iH are equal, and the difference N of the front and rear strip layer numbers is within the range of [HNum-2, HNum+2], the inheritance condition is met;

5) Other conditions do not meet the inheritance conditions.

(e) Before any strip rolling, take the short-term shape self-learning amount and the long-term self-learning amount of the corresponding layer, and if the inheritance condition is met, read the inherited self-learning amount of the previous strip.

Therefore, in the roll bending calculation formula, if the inheritance conditions are met, the self-learning value C _i+1 of the i+1 piece of strip steel after the specification is changed is expressed as:

C _i+1 ＝ΔC _s (i+1)+ΔC _l (i+1)+ΔC _a (i+1)

If the inheritance condition is not satisfied, the self-learning value C _i+1 of the i+1 piece of steel strip after changing the specification is expressed as:

C _i+1 ＝ΔC _s (i+1)+ΔC _l (i+1)

(f) Combined with the shape self-learning value, the setting calculation of the shape model is performed.

The beneficial effects of the present invention are: due to the adoption of the above technical proposal, the present invention can significantly improve the control accuracy of the first strip shape after changing specifications, and this self-learning scheme has good adaptability to cross-rolling and traditional rolling plan arrangement . The present invention can achieve the effect by modifying the self-learning mode in the secondary computer, without equipment and system transformation, easy realization and high feasibility. In order to verify the effectiveness of the control, an experimental comparison was carried out in a 1780mm hot continuous rolling mill. The experimental period was two months. The comparison data show that the hit rate of the first strip steel strip in the specification is increased by 58% when the latter self-learning method is adopted.

Description of drawings:

Fig. 1 Logic block diagram of rolling plan arrangement (strip steel grades have the same width and thickness).

Fig. 2 Logic block diagram of rolling plan arrangement (i+2 strip thickness changes).

Fig. 3 Logic block diagram of rolling plan arrangement (i+2 strip width changes).

Fig. 4 Short-term self-study flow chart of strip steel A.

Fig. 5 Long-term self-study flow chart of strip steel A.

Figure 6 The overall flow chart of self-study of strip steel shape.

Detailed ways:

The present invention will be further described below in conjunction with accompanying drawing and specific embodiment:

1) Determination of stall division and layer number. As shown in Table 1, the steel grade of a 1780mm hot rolling mill is divided into 8 grades in total, GNum=8; the width is divided into 10 grades, BNum=10; the thickness is divided into 12 grades, HNum=12; the total layer record is 8× 10×12=960. A data area with 960 records is opened in the computer, and long-term self-learning and short-term self-learning are placed in each data area for access and reading.

Define the specification of strip steel A: Q235B, width 1250mm, thickness 3.0mm;

Define the specification of strip steel B: Q235B, width 1250mm, thickness 2.3mm;

Define strip C specification: Q235B, width 1000mm, thickness 3.0mm;

Define strip steel D specification: Q235B, width 1000mm, thickness 2.0mm;

Table 1 Steel grade width and thickness classification table

Then strip steel A: iG=2, iB=5, iH=3, its layer number N=2×10×12+5×12+3+1=304. Both the long-term self-learning and short-term self-learning of the strip steel of this specification are stored in the layer number of 304.

Then strip steel B: iG=2, iB=5, iH=2, its layer number N=2×10×12+5×12+2+1=303. Both the long-term self-learning and short-term self-learning of the strip steel of this specification are stored in the layer number of 303.

Then strip steel C: iG=2, iB=2, iH=3, its layer number N=2×10×12+2×12+3+1=268. Long-term self-study and short-term self-study of this specification are stored in the layer number of 268.

Then strip steel D: iG=2, iB=2, iH=2, its layer number N=2×10×12+2×12+2+1=267. Both the long-term self-learning and short-term self-learning of the strip steel of this specification are stored in the layer number of 267.

2) Short-term self-study. Take strip steel A as an example, if strip steel A is rolled after changing rolls, and strip steel A is rolled for the first time, there is no short-term self-learning amount at this time. After the rolling of A is completed, the short-term self-learning amount is calculated. When the second strip A is rolled, the short-term self-learning amount is taken from before setting the bending force to the 304th floor. After the rolling is completed, the short-term self-learning amount is corrected. And save it to layer 304 for use in the subsequent rolling of strip steel A. As shown in Figure 4, it is the short-term self-learning flow chart of strip steel A.

3) Long-term self-study. Taking strip steel A as an example, if strip steel A has been rolled in a certain rolling unit, after changing the rolls, according to the long-term self-learning calculation formula, take out the short-term self-learning amount of 304 layers, calculate the long-term self-learning amount, and then Update the long-term self-learning value of the 304 layer for the reading of the next rolling unit strip A, and clear the short-term self-learning value of the 304 layer. Figure 5 is the long-term self-learning flow chart of strip steel A.

4) Inherited self-learning. According to what is mentioned in the summary of the invention, four steel strips A, B, C, and D are taken as examples to illustrate the process of inherited self-learning.

Assumption 1: After rolling strip A, roll strip B (the first rolling of B). After the rolling of strip steel A is completed, the inheritance self-learning calculation is carried out. Compared with strip steel A, the steel type and width of strip B remain unchanged, and the thickness changes, but the layer number changes to 1, which meets the inheritance conditions. The steel reads the inherited self-learning value of strip steel A, and the long-term self-learning value and the inherited self-learning value of strip steel A are read from layer 303 when setting the shape of strip steel B.

Hypothesis 2: If strip D is rolled after strip C is rolled (the first rolling of D), then the same reason is met, and the succession condition is satisfied.

Hypothesis 3: If strip A is rolled, strip C is rolled (the first rolling of C), the steel type and thickness remain unchanged, the width changes, and the layer change is 36, exceeding [10, 12] If the inheritance condition is not met, then when strip C is rolled, the shape setting value only takes the long-term self-learning value of the layer where strip C is located, and does not take the inherited self-learning value of strip A.

Hypothesis 4: If after rolling strip A, roll strip B, and then roll strip A, because A has been rolled before rolling B, it already has a short-term self-learning amount and does not meet the inheritance conditions, the second When rolling A for the first time, the strip shape setting reads the long-term self-learning value and short-term self-learning value of the layer where strip steel A is located, and does not take the inherited self-learning value of strip steel B.

The same reason can introduce the use of three self-learning methods under any combination of rolling plans.

Figure 6 shows the overall flow chart of self-study of arbitrary strip shape.

Claims (3)

1. a self-learning method for first belt plate shape quality of specification is changed in improvement, it is characterized in that: described method comprises following steps:

(a) first band steel steel grade, width, thickness are classified, determine the layer alias at place; The layer alias of wherein mentioning calculates relevant with band steel steel grade, width, thickness, and regulation steel grade is total GNum shelves always, and width is total BNum shelves always, thickness is total HNum shelves always, if a certain band steel steel grade gear of living in is iG, width gear is iB, thickness gear is iH, and layer alias N computational methods are:

The identical band steel of layer alias, its short-term self study, long-term self study all deposit identical data area in after having calculated, the data area difference that layer alias difference deposits in;

(b) after belt steel rolling completes arbitrarily, carry out the short-term self study of plate shape and inheritance self study and calculate, short-term self study calculated value is saved in equivalent layer alias;

(c), after roll change finishes, carry out the long-term self study of plate shape and calculate, and by learning value be deposited into corresponding layer not among, by the other short-term self study of equivalent layer zero clearing;

(d), before any belt steel rolling, judge that whether self study succession condition satisfies condition, if satisfy condition, reads lastblock band steel inheritance self study value; Belt steel rolling completes arbitrarily, calculates inheritance self study amount, and formula is:

In above formula:

it is i+1 piece belt plate shape inheritance self study amount;

for adjusting coefficient according to the plate shape short-term self study out of i piece belt plate shape measured value inverse; k ₃for plate shape short-term self study gain coefficient, span is 0.3-0.7;

Wherein, i and i+1 piece band steel layer alias might not be identical, if meet succession condition,

i+1 piece band steel is worked, otherwise inoperative;

Whether meet succession condition judgment as follows:

1) inherit the band steel of condition judgment, need to meet following requirement: second block of band steel starts judgement from roll change open rolling; There is variation in the specification with steel; Band steel did not carry out rolling within this roll change cycle;

2) if front and back band steel iG and iB are all equal, front and back band steel layer alias N difference, in [2 ,+2] scope, meets succession condition;

3) if front and back band steel iB and iH are all equal, front and back band steel layer alias is poor in [BNum × HNum-2, BNum × HNum+2] scope, meets succession condition;

4) if front and back iG and iH are all equal, front and back band steel layer alias is poor in [HNum-2, HNum+2] scope, meets succession condition;

5) all the other situations do not meet succession condition;

(e), before any belt steel rolling, read other short-term plate shape self study amount of equivalent layer and long-term self study amount; (f), in conjunction with three kinds of plate shape self study values, carry out the setting of shape models and calculate.

2. the self-learning method of first belt plate shape quality of specification is changed in a kind of improvement according to claim 1, it is characterized in that: in described step (b), the self study of plate shape short-term is not divided and carried out self study according to layer:

In above formula:

for the identical i+1 piece belt plate shape short-term self study amount of layer alias;

for the identical i piece belt plate shape short-term self study amount of layer alias;

for adjusting coefficient according to the plate shape short-term self study out of i piece belt plate shape measured value inverse;

K ₁for plate shape short-term self study gain coefficient, span is 0.3-0.7.

3. the self-learning method of first belt plate shape quality of specification is changed in a kind of improvement according to claim 1, it is characterized in that: described step (c), after work roll changing, all layer alias for rolling in the roll change cycle carry out the long-term self study of plate shape, according to layer, do not divide and carry out self study, adopt following algorithm:

In above formula:

for the identical long-term self study amount of j+1 roll change unit belt plate shape of layer alias;

for the identical long-term self study amount of j roll change unit belt plate shape of layer alias;

for according to j unit belt plate shape short-term self study amount;

K ₂for the long-term self study gain coefficient of plate shape, span is 0.3-0.7;

After long-term self study completes, by j roll change unit short-term self study amount

zero clearing.

Images