CN116000106B - Rolling force setting method in cold continuous rolling speed increasing and decreasing stage - Google Patents
Rolling force setting method in cold continuous rolling speed increasing and decreasing stage Download PDFInfo
- Publication number
- CN116000106B CN116000106B CN202310302713.2A CN202310302713A CN116000106B CN 116000106 B CN116000106 B CN 116000106B CN 202310302713 A CN202310302713 A CN 202310302713A CN 116000106 B CN116000106 B CN 116000106B
- Authority
- CN
- China
- Prior art keywords
- rolling
- speed
- power
- rolled piece
- parameters
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000005096 rolling process Methods 0.000 title claims abstract description 156
- 238000000034 method Methods 0.000 title claims abstract description 52
- 230000003247 decreasing effect Effects 0.000 title claims abstract description 28
- 238000012549 training Methods 0.000 claims abstract description 27
- 238000004519 manufacturing process Methods 0.000 claims abstract description 11
- 238000010008 shearing Methods 0.000 claims abstract description 10
- 230000008569 process Effects 0.000 claims abstract description 9
- 230000007935 neutral effect Effects 0.000 claims description 25
- 229910000831 Steel Inorganic materials 0.000 claims description 16
- 239000010959 steel Substances 0.000 claims description 16
- 230000008859 change Effects 0.000 claims description 14
- 238000009826 distribution Methods 0.000 claims description 11
- 238000005097 cold rolling Methods 0.000 claims description 6
- 230000007423 decrease Effects 0.000 claims description 5
- 239000000839 emulsion Substances 0.000 claims description 3
- 239000013072 incoming material Substances 0.000 claims description 3
- 238000005457 optimization Methods 0.000 claims description 3
- 238000004364 calculation method Methods 0.000 description 11
- 230000007246 mechanism Effects 0.000 description 9
- 238000013528 artificial neural network Methods 0.000 description 4
- 238000004422 calculation algorithm Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 4
- 238000012216 screening Methods 0.000 description 3
- 230000003068 static effect Effects 0.000 description 3
- 238000012937 correction Methods 0.000 description 2
- 230000008878 coupling Effects 0.000 description 2
- 238000010168 coupling process Methods 0.000 description 2
- 238000005859 coupling reaction Methods 0.000 description 2
- 238000013499 data model Methods 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 238000003064 k means clustering Methods 0.000 description 2
- 238000005461 lubrication Methods 0.000 description 2
- 238000007637 random forest analysis Methods 0.000 description 2
- 230000009467 reduction Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 230000000630 rising effect Effects 0.000 description 2
- 238000013459 approach Methods 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000002238 attenuated effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000033228 biological regulation Effects 0.000 description 1
- 230000006835 compression Effects 0.000 description 1
- 238000007906 compression Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000012821 model calculation Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000036961 partial effect Effects 0.000 description 1
- 238000011084 recovery Methods 0.000 description 1
- 230000002829 reductive effect Effects 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/30—Computing systems specially adapted for manufacturing
Landscapes
- Metal Rolling (AREA)
Abstract
The invention relates to a rolling force setting method in a cold continuous rolling speed increasing and decreasing stage, which comprises the following steps: step 1: collecting parameters of rolled pieces, parameters of rolling process and parameters of rollers; step 2: according to the deformation characteristics of the rolled piece, constructing a dynamic speed field and a strain rate field which meet the volume invariant condition and the speed boundary condition, obtaining a rolled piece forming power functional expression, and solving to obtain a functional and minimum values of each forming power; step 3: taking the parameters acquired in the step 1 and the internal plastic deformation power, the shearing power and the tension power obtained in the step 2 as primary selection input characteristics, calculating mutual information entropy between the primary selection input characteristics and rolling force, and eliminating the characteristics with lower mutual information entropy to obtain final input characteristics of the model; step 4: and establishing an LSTM network model, determining optimal super parameters, optimizing the learning rate in the training process to improve the initial stability and training speed of the network, and carrying out network training and rolling force prediction according to actual production data.
Description
Technical Field
The invention belongs to the technical field of automatic production in a rolling process, and relates to a rolling force setting method in a cold continuous rolling speed increasing and decreasing stage.
Background
The rolling force is taken as an important technological parameter in the cold continuous rolling production, not only directly influences the size and shape of the roll gap, but also further influences the thickness precision and the plate shape quality of the finished strip steel due to the fluctuation condition, so that the accurate regulation and control are required. The main factors affecting the rolling force include inter-frame tension, process lubrication and rolling speed in addition to the non-uniformity of properties and thickness fluctuation of the hot rolled stock. In cold continuous rolling production, the speed reduction, the seam passing and the speed rising are indispensable stages, but in the process of the speed rising and falling, the rolling speed can lead to severe fluctuation of rolling force and cause thickness super-tolerance and plate shape defects by influencing the lubrication state of a rolling interface.
Aiming at the problem of fluctuation of rolling force in the cold continuous rolling speed increasing and decreasing stage, a plurality of domestic researchers are doing relevant researches. The Chinese patent publication No. CN110802114A discloses a method for calculating the rolling force of a cold-rolled sheet and strip, and discloses a method for improving the calculation accuracy of the rolling force through iteration model coefficients. The Chinese patent application with publication number of CN114722516A is a method for setting rolling force and rolling moment of a cold rolling full deformation zone of a steel strip, fully considers the characteristics of the cold rolling deformation zone, refines the cold rolling deformation zone into an inlet elastic compression zone, a plastic deformation zone and an outlet elastic recovery zone, and iteratively calculates unit rolling force and torque of each zone to obtain set values of the rolling force and the rolling moment of the full deformation zone.
The Chinese patent application with publication number of CN114510864A is a method for forecasting the rolling force of the neural network based on a K-means clustering algorithm, a data center and an expansion constant of a radial basis function of hidden layer nodes are determined by adopting the K-means clustering algorithm, and then the weight of an output layer is calculated by a supervised learning algorithm, so that unreasonable setting of the neural network super-parameters caused by human factors is avoided, and the rolling force calculation precision is improved. The Chinese patent with publication number of CN111790762A discloses a hot rolled strip steel rolling force setting method based on random forests, which is used for providing a rolling force calculation result of a mechanism model as an input characteristic of the random forests and performing model tuning by adopting grid search. In order to more effectively utilize a theoretical model and a big data model, the Chinese patent application with publication number of CN112711867A discloses a rolling force prediction method for fusing the theoretical model and the big data model, and an integrated model with a fused correction coefficient is constructed based on an average error multiplication compensation principle, wherein the correction coefficient is determined by the rolling force prediction errors of the theoretical model and a BP neural network, and the model inherits the structure of the theoretical model, exerts the high-precision characteristic of the neural network and can adapt to various positive and negative deviation conditions.
The above-mentioned deficiencies of research mainly have two aspects: (1) The influence of the rolling speed rise and fall on the rolling force is not reflected in the mechanism model no matter the traditional rolling force calculation formula or the energy method; (2) The currently used artificial intelligence algorithms fail to take into account the time-series characteristics of certain parameters, such as roll roughness decreasing with increasing rolling length, and therefore these models may lack certain key laws, resulting in a reduction in the accuracy of rolling force prediction.
Disclosure of Invention
In order to solve the technical problems, the invention aims to provide a rolling force setting method in a cold continuous rolling speed increasing and decreasing stage, which firstly obtains a mechanism model considering the influence of increasing and decreasing speed on the rolling force by constructing a double-parabola dynamic speed field, then adopts an LSTM network to deeply excavate the time sequence characteristics of a mechanism model calculation result and other input characteristics, and carries out compensation learning on errors of the mechanism model, thereby finally achieving the purpose of improving the rolling force prediction and thickness control precision in the speed increasing and decreasing stage.
The invention provides a rolling force setting method in a cold continuous rolling speed increasing and decreasing stage, which comprises the following steps:
step 1: collecting parameters of rolled pieces, parameters of rolling process and parameters of rollers;
step 2: according to the deformation characteristics of the rolled piece, a dynamic speed field and a strain rate field which meet the volume-unchanged condition and the speed boundary condition are constructed, a rolled piece forming power functional expression is obtained, and the minimum values of functional and internal plastic deformation power, shearing power, tension power and friction power are obtained;
step 3: taking the rolling piece parameters, the rolling process parameters and the roller parameters as well as the internal plastic deformation power, the shearing power and the tension power obtained in the step 2 as initial selection input characteristics, calculating mutual information entropy between the initial selection input characteristics and the rolling force, and eliminating six characteristics with the lowest mutual information entropy to obtain final input characteristics of the model;
step 4: and establishing an LSTM network model, determining optimal super parameters, optimizing the learning rate in the training process to improve the initial stability and training speed of the network, and carrying out network training and rolling force prediction according to actual production data.
In the rolling force setting method in the cold continuous rolling speed increasing and decreasing stage of the present invention, in the step 1:
the rolled piece parameters include: the brand of the strip steel, the width of the strip steel and the thickness of the incoming material; the rolling process parameters comprise: the tension before and after each pass, rolling speed, strip steel outlet/inlet thickness, actual roll gap value and change speed thereof, motor torque and emulsion flow; the parameters of the roller include: the roller diameter, the roughness and the rolling length of each pass roller;
the actual roll gap change speed is obtained by differentiating the actual roll gap value; all remaining product parameters, rolling process parameters and roll parameters are obtained from the cold rolling line.
In the rolling force setting method in the cold continuous rolling speed increasing and decreasing stage of the invention, the step 2 specifically comprises the following steps:
step 2.1: a double parabolic dynamic velocity field is constructed according to the following assumption:
(1) The dynamic speed field is linearly overlapped by a steady-state speed field and a vertical speed additional term, and the steady-state speed field is in a special form when the roll gap change speed is zero;
(2) The horizontal speed difference between the surface and the center of the strip steel gradually increases from the inlet to the neutral plane, and gradually decreases to zero from the neutral plane to the outlet;
(3) The horizontal speed is parabolic on the same vertical section;
step 2.2: obtaining a rolling piece forming power functional expression according to a double-parabola dynamic speed field, wherein the rolling piece forming power comprises the following components: internal plastic deformation power, friction power, shear power, and tension power;
step 2.3: and solving a neutral angle which minimizes the power functional by adopting a search method, and calculating internal plastic deformation power, friction power, shearing power and tension power at the moment.
In the rolling force setting method in the cold continuous rolling speed increasing and decreasing stage, the double-parabola dynamic speed field constructed in the step 2.1 is as follows:
in the formula :
wherein ,xis the distance from the entrance of the rolled piece;yis the distance from the rolling line;v x is the horizontal speed distribution of the rolled piece;v y is the vertical speed distribution of the rolled piece;Usecond volume flow for rolled piece;h x the thickness of the vertical section of the rolled piece is arbitrary;is the derivative of the thickness of any vertical section of the rolled piece;h 0 is the thickness of the inlet of the rolled piece;h 1 is the thickness of the outlet of the rolled piece;h n the thickness of the neutral plane of the rolled piece;x n is the distance between the neutral surface and the inlet of the rolled piece;α n is a neutral angle;v n the horizontal speed of the roller at the neutral plane is set;lis the length of the deformation zone;bis the width of the rolled piece;R 0 is the radius of the roller;v r is the rolling speed;v c (t) The roll gap change speed is the roll gap change speed;ttime is;v 0 a horizontal velocity profile for the inlet of the product;v 1 the horizontal speed distribution of the outlet of the rolled piece;
the corresponding strain rate field is shown as follows:
wherein ,is thatxA directional line strain rate; />Is thatyA directional line strain rate; />Is thatxyIn-plane strain rate.
In the rolling force setting method in the cold continuous rolling speed increasing and decreasing stage of the invention, the rolling piece forming power functional expression in the step 2.2 comprises the following steps:
internal plastic deformation power expression:
wherein ,σ s is the deformation resistance of the rolled piece; according to the deformation characteristics of the rolled piece in the deformation zone, the maximum strain rate can be obtainedAnd minimum strain rate->;VIs the volume of the rolled piece; />Is the strain rate tensor; />Is specific plastic power; z is the distance from the center line of the rolled piece;
friction power expression:
in the formula :
wherein ,mis the coefficient of friction;kshear strength of the rolled piece; deltav x Is thatxThe speed difference between the directional rolling piece and the roller; deltav y Is thatyThe speed difference between the directional rolling piece and the roller;α 0 is the inlet cross-sectional contact angle;α 1 is the outlet cross-section contact angle;α x a contact angle of any vertical interface of a rolled piece;ais thatα 0 、α 1 Or (b)α n ;
Shear power expression:
wherein ,Δv t0 A discontinuous amount of velocity for the inlet cross-sectional workpiece; deltav t1 A discontinuous amount of product velocity for the outlet cross section;is the derivative of the thickness of the inlet of the rolled piece; />Is the derivative of the product outlet thickness;
tension power expression:
wherein ,σ b is the back tension of the rolled piece;σ f is the front tension of the rolled piece.
In the rolling force setting method in the cold continuous rolling speed increasing and decreasing stage, in the step 2.3, a searching method is adopted to solve a neutral angle which minimizes a power functional, and the expression is as follows:
in the formula :
in the rolling force setting method in the cold continuous rolling speed increasing and decreasing stage of the present invention, the step 4 includes:
step 4.1: performing parameter optimization by adopting a grid search method, and determining the optimal layer number, node number and time step of the LSTM network;
step 4.2: dynamically adjusting the learning rate in the training process by adopting a learning rate preheating strategy so as to improve the initial stability and training speed of the network;
step 4.3: and performing network training according to the actual production data, and calculating a rolling force predicted value.
In the rolling force setting method in the cold continuous rolling speed increasing and decreasing stage, the learning rate in the initial training stage in the step 4.2 is linearly increased, and the rolling force setting method shows an exponential decay trend after reaching a certain iteration round.
The rolling force setting method in the cold continuous rolling speed increasing and decreasing stage has the following beneficial effects:
firstly, by constructing a double-parabola dynamic speed field, the theoretical calculation precision of the rolling force in the lifting speed stage is improved; secondly, the method does not need to give out a moment arm coefficient and an accurate friction coefficient, and effectively avoids the influence of parameters which are difficult to measure on the rolling force prediction; thirdly, the LSTM network provided by the invention can be used for excavating the time sequence characteristics of data, so that the model learns the deep relation of parameters such as force arm coefficients, friction coefficients and the like, and the rolling force prediction precision is further improved; finally, on the basis of solving the theoretical calculation value of the rolling force by the dynamic speed field, the invention adopts the LSTM network to carry out error compensation learning on the theoretical value, thereby effectively improving the interpretation and prediction precision of the model.
Drawings
FIG. 1 is a flow chart of a rolling force setting method in the cold continuous rolling speed increasing and decreasing stage of the invention;
FIG. 2 is a schematic view of a horizontal velocity distribution of a deformation zone according to an embodiment of the present invention;
FIG. 3 shows the mutual information entropy between each feature and rolling force in the embodiment of the invention;
FIG. 4 is a comparison of model performance before and after feature screening in an embodiment of the present invention;
FIG. 5 shows the learning rate as a function of iteration cycles in an embodiment of the present invention;
FIG. 6 is a comparison of performance of the present invention using a wakeup front-back model;
FIGS. 7a-7e are graphs showing rolling force predictions and absolute error comparisons for five models in examples of the present invention; wherein 7a is the comparison of the rolling force prediction result and the absolute error of the Hill formula, 7b is the comparison of the rolling force prediction result and the absolute error of the static speed field model, 7c is the comparison of the rolling force prediction result and the absolute error of the dynamic speed field model, 7d is the comparison of the rolling force prediction result and the absolute error of the ANN model, and 7e is the comparison of the rolling force prediction result and the absolute error of the LSTM model;
FIG. 8 is a comparison of the performance of five models in the examples of the present invention.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is evident that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
In this example, a 1450mm UCM six-roller cold continuous rolling mill set of a certain factory is taken as an example, the rolling force of the 5 th stand is predicted, and the rolling mill rollers are all flat rollers.
As shown in fig. 1, the rolling force setting method in the cold continuous rolling speed increasing and decreasing stage of the invention comprises the following steps:
step 1: collecting rolling piece parameters, rolling process parameters and roller parameters, wherein in the step 1:
the rolled piece parameters include: the brand of the strip steel, the width of the strip steel and the thickness of the incoming material;
the rolling process parameters comprise: the tension before and after each pass, rolling speed, strip steel outlet/inlet thickness, actual roll gap value and change speed thereof, motor torque and emulsion flow;
the parameters of the roller include: roll diameter, roughness and rolling length of each pass of the roller.
The actual roll gap change speed is obtained by differentiating the actual roll gap value; all remaining product parameters, rolling process parameters and roll parameters are obtained from the cold rolling line.
Step 2: according to the deformation characteristics of the rolled piece, a dynamic speed field and a strain rate field which meet the volume-unchanged condition and the speed boundary condition are constructed, a rolled piece forming power functional expression is obtained, and minimum values of functional and internal plastic deformation power, shearing power, tension power and friction power are obtained, wherein the step 2 specifically comprises the following steps:
step 2.1: a double parabolic dynamic velocity field is constructed from the deformation zone horizontal velocity profile schematic of fig. 2 and the following assumption, wherein,α x a contact angle of any vertical interface of the rolled piece, and a unit rad;θis the bite angle in rad.
(1) The dynamic speed field is linearly overlapped by a steady-state speed field and a vertical speed additional term, and the steady-state speed field is in a special form when the roll gap change speed is zero;
(2) The horizontal speed difference between the surface and the center of the strip steel gradually increases from the inlet to the neutral plane, and gradually decreases to zero from the neutral plane to the outlet;
(3) The horizontal speed is parabolic on the same vertical section;
in this embodiment, the constructed double parabolic dynamic velocity field is as follows:
in the formula :
wherein ,xthe distance is the unit m from the entrance of the rolled piece;ythe distance from the rolling line is in m;v x the horizontal speed distribution of the rolled piece is in m/s;v y is the vertical speed distribution of the rolled piece, and the unit is m/s;Uis the second volume flow rate of rolled piece, unit m 3 ;h x The thickness of any vertical section of the rolled piece is given by the unit m;is the derivative of the thickness of any vertical section of the rolled piece;h 0 the thickness of the inlet of the rolled piece is in m;h 1 the thickness of the outlet of the rolled piece is m;h n the thickness of the neutral plane of the rolled piece is in unit of m;x n the distance between the neutral surface and the inlet of the rolled piece is in m;α n is the neutral angle, unit rad;v n the horizontal speed of the roller at the neutral plane is in m/s;lthe unit of the deformation zone length is m;bthe width of the rolled piece is the unit m;R 0 the unit is the radius of the roller, m;v r the rolling speed is the unit m/s;v c (t) The unit is m/s for the roll gap change speed;ttime, unit s;v 0 a horizontal velocity profile for the inlet of the product;v 1 the horizontal speed distribution of the outlet of the rolled piece;
the corresponding strain rate field is shown as follows:
wherein ,is thatxDirectional line strain rate, units/s; />Is thatyDirectional line strain rate, units/s; />Is thatxyIn-plane strain rate, units/s.
Step 2.2: obtaining a rolling piece forming power functional expression according to a double-parabola dynamic speed field, wherein the rolling piece forming power comprises the following components: internal plastic deformation power, friction power, shear power, and tension power;
in this embodiment, the GA linearized yield criterion is used to obtain the internal plastic deformation power expression as:
wherein ,σ s is the deformation resistance of the rolled piece, and the unit is MPa; according to the deformation characteristics of the rolled piece in the deformation zone, the maximum strain rate can be obtainedAnd minimum strain rate->;VIs the volume of the rolled piece; />Is the strain rate tensor; />Is specific plastic power; z is the distance from the center line of the rolled piece;
the friction power expression is:
in the formula :
wherein ,mis the coefficient of friction;kshear strength of rolled piece, unit MPa; deltav x Is thatxThe speed difference between the directional rolled piece and the roller is in m/s; deltav y Is thatyThe speed difference between the directional rolled piece and the roller is in m/s;α 0 is the entrance cross-sectional contact angle, unit rad;α 1 the unit rad is the outlet cross-sectional contact angle;α x a contact angle of any vertical interface of a rolled piece;ais thatα 0 、α 1 Or (b)α n ;
The shear power expression is:
wherein ,Δv t0 The unit is m/s for the discontinuous amount of the speed of the rolled piece with the inlet section; deltav t1 The unit is m/s, which is the discontinuous amount of the speed of the rolled piece with the outlet section;is the derivative of the thickness of the inlet of the rolled piece; />Is the derivative of the product outlet thickness;
the tension power expression is:
wherein ,σ b is the post-rolling tension of the rolled piece, and is in MPa;σ f is the front tension of the rolled piece, and is unit MPa.
Step 2.3: solving a neutral angle which minimizes a power functional by adopting a search method, and calculating internal plastic deformation power, friction power, shearing power and tension power at the moment;
in this embodiment, the power functional is used for centering anglesAnd solving a neutral angle which minimizes the power functional by adopting a search method, wherein the partial derivative is equal to 0, and the expression is shown as follows:
in the formula :
finally, the internal plastic deformation power, friction power, shear power and tension power at this neutral angle are calculated.
Step 3: taking the rolling piece parameters, the rolling process parameters and the roller parameters as well as the internal plastic deformation power, the shearing power and the tension power obtained in the step 2 as initial selection input characteristics, calculating the mutual information entropy between the initial selection input characteristics and the rolling force, and eliminating the characteristics with lower mutual information entropy to obtain the final input characteristics of the model; in specific implementation, features are removed one by one according to mutual information entropy, and training is performed, so that the prediction effect is best when the last six features are removed in the embodiment.
In this example, the initial input features include a rolling stock parameter of 3 dimensions, a rolling process parameter of 11 dimensions, a rolling roll parameter of 1 and the resulting internal plastic deformation power, shear power and tension power, totaling 18 dimensions, as shown in table 1.
Table 1 feature number and corresponding name
Wherein, confirm the deformation resistance according to the steel brand in step 1.
According to the mutual information theory, the mutual information entropy between the initial selection input characteristics and the rolling force is calculated, and the result is shown in figure 3. And then eliminating the characteristics with mutual information entropy smaller than 0.5 to obtain the final input characteristics of the model.
The performance pairs of the models before and after feature screening are respectively trained for 30 times, as shown in fig. 4, it can be seen that the feature screening can not only reduce the dimension of the input features and further reduce the calculation time consumption, but also improve the correlation coefficient R and the decision coefficient R of the models 2 Meanwhile, the influence of the random initialization of the model parameters on the stability of the calculation result can be better avoided.
Step 4: establishing an LSTM network model, determining optimal super parameters, optimizing the learning rate in the training process to improve the initial stability and training speed of the network, and carrying out network training and rolling force prediction according to actual production data, wherein the step 4 comprises the following steps:
step 4.1: performing parameter optimization by adopting a grid search method, and determining the optimal layer number, node number and time step of the LSTM network;
in this embodiment, the number of layers is in the range of [1,2,3], the number of nodes is in the range of [5,10,15,20,25,30], and the time step is in the range of [5,10,20,25,50]. The average loss obtained for each group of model parameters is shown in tables 2 and 3, and the calculation results show that the model effect is better the more complex the non-network structure is, the larger the time step is. The optimal super parameters are finally determined to be 2 layers, 25 nodes and 10 time steps.
TABLE 2 influence of network architecture on loss
TABLE 3 influence of time step on loss
Step 4.2: adopting a wakeup strategy, namely a learning rate preheating strategy to dynamically adjust the learning rate in the training process so as to improve the initial stability and training speed of the network;
random initialization of model parameters may fall into a locally optimal solution, resulting in the model taking multiple iteration runs to tune back. Therefore, a small learning rate is used in the prior art, and after the model has rough knowledge of the data distribution, the large learning rate is adopted to accelerate convergence to an optimal solution. After training to a certain iteration round, the attenuation learning rate is adopted to slowly approach to the optimal solution. In this embodiment, the variation of the learning rate with the iteration round is shown in fig. 5. The learning rate in the initial training stage is linearly increased, and the learning rate is exponentially attenuated after reaching a certain iteration round.
Fig. 6 shows the comparison of model performance before and after the wakeup strategy is adopted, so that the model performance is greatly improved after the method is adopted, and meanwhile, the influence of random initialization of model parameters on the stability of a calculation result is greatly avoided.
Step 4.3: and performing network training according to the actual production data, and calculating a rolling force predicted value.
In the embodiment, 1200 actual production data are obtained, wherein 1000 training sets and 200 testing sets are obtained; the model super parameters are the layer number 2, the node number 25 and the time step 10 described in the step 4.1; the learning rate reaches a maximum value of 0.015 at the 50 th iteration round, and the total iteration round is 1500. The rolling force prediction results of five models, namely Hill equation, static speed field, dynamic speed field, ANN and LSTM, are compared, wherein the ANN and LSTM adopt the same network layer number and node number, and the results are shown in figures 7a-7e and 8.
The rolling force prediction result and absolute error pair are shown in fig. 7a-7e, where 7a is Hill formula, 7b is static velocity field, 7c is dynamic velocity field, 7d is ANN model, and 7e is LSTM model, it can be seen that: from the perspective of the predicted values of the rolling forces, the fluctuation of the conventional rolling force calculation formula in fig. 7a is the largest, followed by the mechanism model in fig. 7b and 7c, and the fluctuation of the mechanism and data coupling model in fig. 7d and 7e is the smallest; from the angle of absolute error, the errors of the five models are slightly different in the stable rolling stage, but gradually decrease in sequence, the errors of the models are sequentially reduced in the lifting speed stage, the mechanism and data coupling model decrease in amplitude is maximum, and the LSTM model error is minimum.
In order to further compare the prediction effect, the five models are evaluated by selecting normalized standard deviation, root mean square error and correlation coefficient which are widely applied to the performance evaluation of the fitting model. FIG. 8 is a Taylor plot of three indices that can be compared simultaneously, where circumferential is the correlation coefficient, radial is the normalized standard deviation, and distance from the measured value is the root mean square error. It is not difficult to find that, among the five models, LSTM is closest to the actual measurement value, so that the model is best in overall performance.
In summary, the rolling force setting method for the cold continuous rolling speed increasing and decreasing stage of the LSTM network, which is coupled with the mechanism model based on the dynamic speed field and considers the data time sequence characteristics, has a good prediction effect.
The foregoing description of the preferred embodiments of the invention is not intended to limit the scope of the invention, but rather to enable any modification, equivalent replacement, improvement or the like to be made without departing from the spirit and principles of the invention.
Claims (7)
1. The rolling force setting method for the cold continuous rolling speed increasing and decreasing stage is characterized by comprising the following steps:
step 1: collecting parameters of rolled pieces, parameters of rolling process and parameters of rollers;
step 2: according to the deformation characteristics of the rolled piece, a dynamic speed field and a strain rate field which meet the volume-unchanged condition and the speed boundary condition are constructed, a rolled piece forming power functional expression is obtained, and the minimum values of functional and internal plastic deformation power, shearing power, tension power and friction power are obtained;
step 3: taking the rolling piece parameters, the rolling process parameters and the roller parameters as well as the internal plastic deformation power, the shearing power and the tension power obtained in the step 2 as initial selection input characteristics, calculating mutual information entropy between the initial selection input characteristics and the rolling force, and eliminating six characteristics with the lowest mutual information entropy to obtain final input characteristics of the model;
step 4: establishing an LSTM network model, determining optimal super parameters, optimizing the learning rate in the training process to improve the initial stability and training speed of the network, and carrying out network training and rolling force prediction according to actual production data;
the step 2 specifically comprises the following steps:
step 2.1: a double parabolic dynamic velocity field is constructed according to the following assumption:
(1) The dynamic speed field is linearly overlapped by a steady-state speed field and a vertical speed additional term, and the steady-state speed field is in a special form when the roll gap change speed is zero;
(2) The horizontal speed difference between the surface and the center of the strip steel gradually increases from the inlet to the neutral plane, and gradually decreases to zero from the neutral plane to the outlet;
(3) The horizontal speed is parabolic on the same vertical section;
step 2.2: obtaining a rolling piece forming power functional expression according to a double-parabola dynamic speed field, wherein the rolling piece forming power comprises the following components: internal plastic deformation power, friction power, shear power, and tension power;
step 2.3: and solving a neutral angle which minimizes the power functional by adopting a search method, and calculating internal plastic deformation power, friction power, shearing power and tension power at the moment.
2. The method for setting rolling force in the cold continuous rolling speed increasing and decreasing stage according to claim 1, wherein in the step 1:
the rolled piece parameters include: the brand of the strip steel, the width of the strip steel and the thickness of the incoming material; the rolling process parameters comprise: the tension before and after each pass, rolling speed, strip steel outlet/inlet thickness, actual roll gap value and change speed thereof, motor torque and emulsion flow; the parameters of the roller include: the roller diameter, the roughness and the rolling length of each pass roller;
the actual roll gap change speed is obtained by differentiating the actual roll gap value; all remaining product parameters, rolling process parameters and roll parameters are obtained from the cold rolling line.
3. The rolling force setting method of the cold continuous rolling speed increasing and decreasing stage according to claim 1, wherein the double parabolic dynamic speed field constructed in the step 2.1 is as follows:
in the formula :
wherein ,xis the distance from the entrance of the rolled piece;yis the distance from the rolling line;v x is the horizontal speed distribution of the rolled piece;v y is the vertical speed distribution of the rolled piece;Usecond volume flow for rolled piece;h x the thickness of the vertical section of the rolled piece is arbitrary;is the derivative of the thickness of any vertical section of the rolled piece;h 0 is the thickness of the inlet of the rolled piece;h 1 is the thickness of the outlet of the rolled piece;h n the thickness of the neutral plane of the rolled piece;x n is the distance between the neutral surface and the inlet of the rolled piece;α n is a neutral angle;v n the horizontal speed of the roller at the neutral plane is set;lis the length of the deformation zone;bis the width of the rolled piece;R 0 is the radius of the roller;v r is the rolling speed;v c (t) The roll gap change speed is the roll gap change speed;ttime is;v 0 a horizontal velocity profile for the inlet of the product;v 1 the horizontal speed distribution of the outlet of the rolled piece;
the corresponding strain rate field is shown as follows:
4. The rolling force setting method in the cold continuous rolling speed increasing and decreasing stage according to claim 3, wherein the rolling piece forming power functional expression in step 2.2 specifically includes:
internal plastic deformation power expression:
wherein ,σ s is the deformation resistance of the rolled piece; according to the deformation characteristics of the rolled piece in the deformation zone, the maximum strain rate can be obtainedAnd minimum strain rate->;VIs the volume of the rolled piece; />Is the strain rate tensor; />Is specific plastic power; z is the distance from the center line of the rolled piece;
friction power expression:
in the formula :
wherein ,mis the coefficient of friction;kshear strength of the rolled piece; deltav x Is thatxThe speed difference between the directional rolling piece and the roller; deltav y Is thatyThe speed difference between the directional rolling piece and the roller;α 0 is the inlet cross-sectional contact angle;α 1 is the outlet cross-section contact angle;α x a contact angle of any vertical interface of a rolled piece;ais thatα 0 、α 1 Or (b)α n ;
Shear power expression:
wherein ,Δv t0 A discontinuous amount of velocity for the inlet cross-sectional workpiece; deltav t1 A discontinuous amount of product velocity for the outlet cross section;is the derivative of the thickness of the inlet of the rolled piece; />Is the derivative of the product outlet thickness;
tension power expression:
wherein ,σ b is the back tension of the rolled piece;σ f is the front tension of the rolled piece.
6. the rolling force setting method in the cold continuous rolling speed increasing and decreasing stage according to claim 1, wherein the step 4 includes:
step 4.1: performing parameter optimization by adopting a grid search method, and determining the optimal layer number, node number and time step of the LSTM network;
step 4.2: dynamically adjusting the learning rate in the training process by adopting a learning rate preheating strategy so as to improve the initial stability and training speed of the network;
step 4.3: and performing network training according to the actual production data, and calculating a rolling force predicted value.
7. The method for setting rolling force in the cold continuous rolling speed increasing and decreasing stage according to claim 6, wherein the initial training learning rate in step 4.2 increases linearly and becomes exponentially decaying after reaching a certain iteration cycle.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310302713.2A CN116000106B (en) | 2023-03-27 | 2023-03-27 | Rolling force setting method in cold continuous rolling speed increasing and decreasing stage |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310302713.2A CN116000106B (en) | 2023-03-27 | 2023-03-27 | Rolling force setting method in cold continuous rolling speed increasing and decreasing stage |
Publications (2)
Publication Number | Publication Date |
---|---|
CN116000106A CN116000106A (en) | 2023-04-25 |
CN116000106B true CN116000106B (en) | 2023-06-16 |
Family
ID=86033995
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202310302713.2A Active CN116000106B (en) | 2023-03-27 | 2023-03-27 | Rolling force setting method in cold continuous rolling speed increasing and decreasing stage |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN116000106B (en) |
Families Citing this family (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116984386B (en) * | 2023-09-26 | 2023-12-08 | 太原理工大学 | Method and device for determining force energy parameters in TRB thinning rolling process |
CN116984385B (en) * | 2023-09-26 | 2024-01-05 | 太原理工大学 | Method and device for determining rolling force in rolling of cold continuous rolling strip steel from thin gauge to thick gauge |
Family Cites Families (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US4261190A (en) * | 1979-07-30 | 1981-04-14 | General Electric Company | Flatness control in hot strip mill |
JPH0747171B2 (en) * | 1988-09-20 | 1995-05-24 | 株式会社東芝 | Rolling mill setting method and device |
CN101604144B (en) * | 2008-06-13 | 2011-08-31 | 中国科学院金属研究所 | Modeling method of plate rolling in online control model |
CN104898430B (en) * | 2015-06-03 | 2017-07-18 | 北京首钢自动化信息技术有限公司 | Single stand cold mill rolling force model parameter optimization method based on data mining |
CN106623443B (en) * | 2016-12-28 | 2019-01-22 | 苏州大学 | A kind of slab rough rolling step draught pressure forecast method |
CN113020287B (en) * | 2021-03-01 | 2023-08-18 | 太原理工大学 | Setting method of sine roll type corrugated roll gap of metal composite plate |
CN114602980B (en) * | 2022-04-24 | 2024-06-04 | 太原科技大学 | Method for measuring equivalent strain of central part of medium plate after same-diameter and different-speed rolling |
CN115815345A (en) * | 2022-11-16 | 2023-03-21 | 天铁热轧板有限公司 | Mechanism collaborative prediction method and system for predicting mechanical property of full-flow hot-rolled strip steel |
-
2023
- 2023-03-27 CN CN202310302713.2A patent/CN116000106B/en active Active
Also Published As
Publication number | Publication date |
---|---|
CN116000106A (en) | 2023-04-25 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN116000106B (en) | Rolling force setting method in cold continuous rolling speed increasing and decreasing stage | |
CN108637020B (en) | Self-adaptive variation PSO-BP neural network strip steel convexity prediction method | |
CN107729943B (en) | Missing data fuzzy clustering algorithm for optimizing estimated value of information feedback extreme learning machine and application thereof | |
WO2023130666A1 (en) | Strip steel plate convexity prediction method based on data-driving and mechanism model fusion | |
CN112749505B (en) | Hot rolled strip steel cross-sectional shape prediction method based on mechanism fusion data | |
CN103962390B (en) | Tension force and draught pressure synthetic setting method in VC roller planisher wet jetting piles process | |
CN114722516A (en) | Method for setting rolling force and rolling moment of steel strip cold rolling full deformation area | |
CN108817103B (en) | Steel rolling model steel family layer classification optimization method | |
CN112711867B (en) | Rolling force prediction method integrating theoretical model and big data model | |
WO2022116571A1 (en) | Lstm-based roll-bending force prediction method for hot rolling | |
CN100520786C (en) | Processing parameter setting method of tension straightening machine set correct roller | |
Chen et al. | Determination of formability considering wrinkling defect in first-pass conventional spinning with linear roller path | |
CN115608793B (en) | Finish rolling temperature regulation and control method for mechanism fusion data | |
CN111241750A (en) | BP network cold-rolled strip steel mechanical property prediction method combining genetic algorithm | |
CN102172630A (en) | Roll groove design method for seamless steel pipe stretch reducing machine set | |
CN117655118B (en) | Strip steel plate shape control method and device with multiple modes fused | |
Li et al. | Modeling and validation of bending force for 6-high tandem cold rolling mill based on machine learning models | |
CN110773580B (en) | Method and system for calculating rolling force based on XGboost algorithm | |
Bu et al. | Rolling schedule multi-objective optimizationbased on influence function for thin gauge steel strip in tandem cold rolling | |
CN110227722A (en) | Optimization method based on roll profile in roll changing period | |
CN110910528A (en) | Method and device for predicting tensile strength of paper sheet | |
CN111570532B (en) | Method for predicting influence of hot rolling coiling temperature and finish rolling temperature on flattening friction coefficient | |
Wang et al. | Quantitative study on the relationship between the transverse thickness difference of cold-rolled silicon strip and incoming section profile based on the mechanism-intelligent model | |
CN110193523B (en) | Method for predicting thickness of oil film on surface of roller in secondary cold rolling process | |
CN108555033B (en) | Steel pipe thickens the control method of segment length minimization when a kind of stretch-reducing mill rolls |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |