Method for intelligently predicting planar shape of medium plate
Technical Field
The invention belongs to the field of rolling, and particularly relates to an intelligent prediction method for the plane shape of a medium plate. Mainly relates to the calculation of a patent classification number G06; calculating; the counting G06F electric digital data processing G06F17/00 is particularly suitable for computer-aided design of a function-specific digital computing device or data processing method G06F 17/50.
Background
The planar shape of the final product deviates from a rectangular shape due to the free flow of the metal under three-dimensional deformation in the multi-pass rolling process of the medium plate, as shown in fig. 1, which causes an increase in the amount of cutting loss in the subsequent shearing process, which is one of the important factors affecting the yield. The current common control means is to carry out load reduction plane shape control through a hydraulic system at the last forming pass or the last widening pass, and convert the volume of a defect part into corresponding change of the thickness of the cross section of a plate blank so as to enable the plane shape of a final steel plate to be close to a rectangle and reduce the cutting loss of the head, the tail and the side edges of the steel plate to the maximum extent.
The control precision of the plane shape is determined by the prediction precision of the final plane shape of the steel plate under different process conditions, and a common plane shape prediction method is to obtain a mathematical model through experimental simulation and regression according to actually measured data.
Because the regression model is simple and the actual production conditions are complex, the plane shape prediction curve established under the experimental conditions is difficult to adapt to the actual production, the prediction precision is not high, and the plane shape on-line application effect cannot meet the control requirement of the optimal yield.
Disclosure of Invention
Aiming at the problems that a traditional plane shape prediction model is simple, poor in precision and difficult to realize on-line high-precision calculation, the invention provides an intelligent prediction method for the plane shape of a medium plate
Establishing a three-dimensional finite element explicit dynamics model in the billet rolling process, setting rolling conditions, simulating the billet rolling process, and extracting node coordinates at the edge of a rolled piece of a simulation result;
-generating a metal flow curve corresponding to the current simulation process according to the coordinate fitting of the edge nodes;
repeating the process to obtain different billets and a plurality of metal flow curves corresponding to the rolling process under different rolling conditions;
selecting the inlet thickness H, the width W and the reduction rate of a rolled piece as input parameters; selecting key points on the multiple metal flow curves as the output of an artificial neural network, and training the artificial neural network;
inputting the parameters and rolling conditions of the steel billet to be predicted currently into the trained artificial neural network to generate a prediction result, and completing the forming prediction.
In a preferred embodiment, for the plane shape prediction in the medium plate transverse-longitudinal rolling molding mode, the multi-pass rolling is divided into a billet widening stage, a longitudinal rolling first pass stage and a longitudinal rolling remaining pass rolling stage;
the stretching stage is used for predicting the side shape of each rolling pass corresponding to each stretching stage based on the trained artificial neural network, and a prediction result corresponding to each pass is generated; after all the rolling passes of the widening stage are completed, superposing a predictor result corresponding to each pass to obtain a last pass side prediction result of the widening stage;
-first pass of longitudinal rolling: the rolled piece rotates 90 degrees after being stretched and rolled for longitudinal rolling, and the prediction result of the last secondary side of the stretching stage is used as the initial value of the head and tail shape of the first primary longitudinal rolling pass; calculating the head and tail shape change of the rolled piece caused by setting of the planar shape curve parameters followed by the rolling mill under the condition of load;
calculating the head and tail shape change of the rolled piece caused by setting of the planar shape curve parameters followed by the rolling mill under the condition of load;
directly predicting the first pass of longitudinal rolling through the trained artificial neural network to obtain predicted head and tail shape changes;
superposing the shape changes to obtain a head and tail shape prediction result of the rolled piece in the first-pass longitudinal rolling stage;
-rolling stage of the remaining passes of longitudinal rolling:
and superposing the head and tail shape change caused by the longitudinal normal extension and the head and tail shape change of the longitudinal rolling pass obtained based on the artificial neural network prediction as the final head and tail shape, namely the final prediction result.
Furthermore, the specific process of calculating the head and tail shape change of the rolled piece caused by the setting of the plane shape curve parameters followed by the rolling mill under the load pressure is as follows:
-setting control parameters of the curve followed by the rolling mill under load: the length of the stable segment is L
1Length of loaded lower section is L
2The loaded reduction amount is delta h and the central shape compensation amount is d; the width of the steel plate is L, and the thickness of an outlet is initially set to be h; the coordinates of the 7-point plane shape control point are respectively (0, h-a + delta h), (L)
1,h-a+Δh),(L
1+L
2,h-a),(
h-a-d),(L-L
1-L
2,h-a),(L-L
1H-a + Δ h), (L, h-a + Δ h); corresponding thickness thk1 h-a-d, thk2 h-a, thk3 h-a + Δ h;
-obtaining an intermediate variable a ensuring that the rolled length remains constant under control of the plan shape by calculation according to the following formula:
when the plane shape control is not put into use, rolling the steel plate of the last pass of the transverse rolling into a thickness H from the thickness H; and when the plane shape control is put into, the last transverse rolling pass is the plane shape control pass, and at the moment, the steel plate is rolled into a variable thickness shape set by a 7-point curve from the thickness H, and in order to ensure that the rolled length is consistent with the rolled length when the plane shape is not put into, the value of the intermediate variable a is calculated according to the volume invariance principle and plane shape setting parameters.
Furthermore, in the calculation process of the rolling stage of the rest passes of the longitudinal rolling, the shape of the cross section of the inlet thickness is assumed to be a rectangular cross section;
assuming that no width exists in the longitudinal rolling process, calculating the longitudinal normal extension according to the volume invariance principle;
and setting a plurality of tracking points in the half-width direction of the rolled piece according to the plane shape setting parameters, calculating the corresponding normal extension length according to the tracking point position for each pass of longitudinal rolling, and superposing the corresponding normal extension length with the head and tail shapes output by the artificial neural network to obtain the head and tail shapes of the pass of rolling.
As a preferred embodiment, a 4-degree curve is used for the fitting, the 4-degree curve having the form:
y=A2x2+A3x3+A4x4
wherein x is node normalized horizontal axis coordinate, x is node horizontal axis coordinate/1000, for widening rolling, the horizontal axis direction is rolling direction, and for longitudinal rolling, the horizontal axis direction is width direction; y is the output of the fitting curve shape; a. the2、A3And A4Are fitting parameters of the curve.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions of the prior art, the drawings needed to be used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the drawings without creative efforts.
FIG. 1 is a schematic diagram of the shape change of a medium plate rolling process mentioned in the background of the invention
FIG. 2 is a schematic diagram of selecting key training points on a shape fitting curve of a rolled piece according to the invention
FIG. 3 is a schematic diagram of an artificial neuron network according to the present invention
FIG. 4 is a schematic diagram of the calculation process of the side shape of the widening stage according to the present invention
FIG. 5 is a schematic diagram of a 7-point control curve selected according to the planar shape of the present invention
FIG. 6 is a schematic view of the rolling stock planform prediction for the rolling process of the present invention
FIG. 7 is a schematic diagram of the results of finite element simulation calculation of rolled pieces of the same width of the present invention, wherein FIG. 7a is a schematic diagram of the results of finite element simulation calculation of head shape with a reduction rate of 20% and an entrance thickness of 50 mm; FIG. 7b is a schematic diagram showing the results of finite element simulation calculation of head shape with a reduction ratio of 20% and an inlet thickness of 200mm
FIG. 8 is a schematic diagram of an error state when a neural network is trained 10000 times in the embodiment of the present invention
FIG. 9 is a schematic diagram of the change of the head and tail shapes during the rolling process of each longitudinal rolling pass in the embodiment of the invention, wherein FIG. 9a is a schematic diagram of the head change; FIG. 9b is a schematic diagram of a tail variation
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the following describes the technical solutions of the embodiments of the present invention clearly and completely with reference to the accompanying drawings in the embodiments of the present invention:
as shown in fig. 1-9: the technical route adopted by the intelligent prediction method for the plane shape of the medium plate is as follows:
1. simulation calculation of single-pass rolling process by using finite element software
Establishing a three-dimensional finite element explicit dynamic model in the billet rolling process,
setting a roller as a rigid body and a rolled piece as a deformable body to cover initial conditions of blanks and finished products in the production process, including the thickness, the width and the reduction rate of the rolled piece, performing unit division on the rolled piece in finite element software, setting friction conditions, applying parameters such as initial speeds of the roller and the rolled piece, and then performing simulation, and extracting node coordinates at the head, the tail and the side edges of the rolled piece in the result as evaluation calculation bases of metal flow.
Initial simulation parameters of the rolled piece are as follows:
rolling simulation initial parameter of widening stage
● width of steel blank: 2000mm, 2400mm, 2800mm, 3200mm
● thickness of steel blank: 150mm, 200mm, 250mm, 300mm
● reduction ratio: 5%, 10%, 15%, 20%
Rolling simulation initial parameter in longitudinal rolling stage
● width of steel blank: 1500mm, 2000mm, 2400mm, 2850mm, 3300mm
● thickness of steel blank: 10mm, 50mm, 100mm, 150mm, 200mm, 250mm, 300mm
● reduction ratio: 5%, 10%, 15%, 20%, 30%
2. Establishing a metal flow fitting curve of a rolling process
To simplify the calculation, for the widening rolling, the side metal flow within 800mm from the head to the tail was followed; for longitudinal rolling, the head-to-tail metal flow is tracked over the half width of the product.
Obtaining node coordinates in a simulation result of finite element software, fitting by adopting a curve for 4 times after processing, wherein the curve for 4 times adopts the following form, so that parameter fitting is facilitated.
y=A2x2+A3x3+A4x4(1)
Wherein x is node normalized horizontal axis coordinate, x is node horizontal axis coordinate/1000, for widening rolling, the horizontal axis direction is rolling direction, and for longitudinal rolling, the horizontal axis direction is width direction; y is the output of the fitting curve shape; a. the2、A3And A4Are fitting parameters of the curve.
In order to ensure the precision of artificial neural network training on the fitting curve, the invention selects 3 key points P on the fitting curve for 4 times at equal intervals1、P2And P3Height h of1、h2And h3As an output of the neural network, a neural network is selected,
as shown in fig. 2. Assuming that the half width of a rolled piece is w, the selection of the coordinates of the transverse axis of 3 points is respectively as follows: x is the number of1=w/3,x2=w*2/3,x3=w,h1、h2、h3The corresponding height on the fitted curve.
3. Training single-pass three-dimensional rolling metal flow by using artificial neural network
Selecting the inlet thickness H, the width W and the reduction rate of a rolled piece as input parameters to fit the height H of 3 key points selected on the curve1、h2、h3As output parameterAnd (3) training a plurality of groups of results of the finite element software three-dimensional simulation to obtain the optimal parameters of the training network, wherein the trained neural network can calculate the metal flow curve of the rolled piece with any specification in the single-pass rolling process, and the structure of the neural network is shown in figure 3.
Height h of 3 key points output by trained artificial neuron network1、h2And h3The curve parameter A can be obtained by substituting 4 fitting curve equations (1)2、A3And A4。
The fitted curve parameters are therefore:
4. method for predicting head and tail shapes of rolled pieces in multi-pass rolling process in superposition mode
(1) Broadening phase shape prediction
In the widening stage, the billet needs to be rolled by rotating 90 degrees, the head and the tail of the original billet are changed into sides, the shape of the sides in the widening process is also changed and needs to be tracked,
and finally, the shape change of each pass in the broadening stage is subjected to superposition processing based on the artificial neuron network prediction.
The change of the shape of the side edge in the widening process is mainly concentrated on the part close to the head and the tail, and the length of 800mm away from the head and the tail is selected as the calculation range of the shape of the side edge by the invention, as shown in figure 4.
And obtaining the final shape of the side edge of the last widening pass after multi-pass superposition calculation, and taking the final shape as the initial head-tail shape of the longitudinal rolling.
(2) Head and tail shape prediction for first pass rolled piece of longitudinal rolling
The last pass of the widening stage is used as a control pass of a plane shape, a hydraulic system of the rolling mill of the last pass can perform load reduction according to a set curve, and the control parameters of a common 7-point set curve comprise: l is1、L2Δ h and d, the plan view control diagram is shown in fig. 5:
in FIG. 5, L1、L2Δ h and d are plane shape control parameters; l is the width of the steel plate; h is the inlet set thickness; thk1, thk2, and thk3 are thicknesses corresponding to the control points; a is an intermediate variable to be determined.
In the last pass of the widening stage, namely the plane shape control pass, in order to ensure the final control width of the steel plate, the relation between the inlet thickness H parameters a of the rolled piece is determined according to the volume invariance principle:
when the plane shape control is not put into use, rolling the steel plate of the last pass of the transverse rolling into a thickness H from the thickness H; and when the plane shape control is put into, the last transverse rolling pass is the plane shape control pass, and at the moment, the steel plate is rolled into a variable thickness shape set by a 7-point curve from the thickness H, and in order to ensure that the rolled length is consistent with the rolled length when the plane shape is not put into, the value of the intermediate variable a is calculated according to the volume invariance principle and plane shape setting parameters.
The predictive calculation of the shape of the head and tail of the first pass of the longitudinal rolling therefore involves a superposition of three parts:
the shape change of the side edge of the widening stage, the head and tail shape change caused by setting of a plane shape curve and the head and tail shape change of the first longitudinal rolling pass obtained based on the artificial neural network prediction.
(3) Prediction of head and tail shapes of rolled pieces in residual passes of longitudinal rolling
Starting from the second pass of longitudinal rolling, the shape of the inlet thickness section is assumed to be a rectangular section, and the change of the head and tail shape comprises the superposition of the following two parts: the shape change of the head and the tail caused by the longitudinal normal extension and the shape change of the head and the tail of the longitudinal rolling pass obtained based on the artificial neural network prediction.
And if no broadening exists in the longitudinal rolling process, the longitudinal normal extension amount can be calculated according to the volume invariance principle.
In order to accurately calculate the head and tail shapes, a plurality of tracking points are arranged in the half-width direction of a rolled piece according to plane shape set parameters, for each pass of longitudinal rolling, the corresponding normal extension length is calculated according to the position of the tracking points, and the corresponding normal extension length is superposed with the head and tail shapes output by the artificial neural network, so that the head and tail shapes after the pass of rolling are obtained.
And (4) taking the final extension length of the superposed tracking points of the pass as the initial length of the next pass, and sequentially and circularly calculating until all the passes are calculated to obtain the final head and tail shapes. A schematic diagram of the prediction of the planform of the rolled product according to the rolling process schedule is shown in fig. 6.
Examples
In this embodiment, the plan shape of a rolling process procedure of a medium plate is predicted, and the process parameters of the rolled product are as follows:
● steel grade: q235
● blank specification: 220mm × 2000mm × 300mm
● specification of finished product: 55mm X2200 mm
● widening last pass 7-point plane shape setting parameters: l is1=150mm,L2=177.57mm,Δh=3.1mm,d=1.0mm
The rolling schedule and precalculated data are shown in table 1.
TABLE 1 Rolling schedule Table
The rolling schedule is totally 8 times, and a transverse-longitudinal rolling mode is adopted, wherein the first 2 times are subjected to widening rolling, the rest 6 times are subjected to longitudinal rolling, and the last widening time adopts 7-point plane shape control. And 3, predicting the plane shape of the head and the tail of the final rolled piece by considering the following influence factors: the shape of the side edge of the rolled piece during the widening rolling pass, the parameter setting of the plane shape and the shape of the head and the tail during the longitudinal rolling pass are set. And the head and tail shape prediction takes the extension at different positions in the width direction during rolling as a calculation basis, and the final head and tail shapes of the steel plate can be obtained through multi-pass superposition of different extension amounts.
1. Finite element three-dimensional simulation calculation metal flow in single-pass rolling process
The finite element simulation calculation of the rolling process is carried out in advance by setting initial parameters covering blanks and finished products in the production process, and unit node coordinates corresponding to different initial setting conditions are obtained and used as a rule representing metal flow. FIG. 7 shows the results of finite element simulation calculations for rolled pieces of different widths, 50mm and 100mm thick, corresponding to a reduction of 20% respectively.
2. Metal flow curve fitting for single pass rolling process
In order to be able to represent the change in head-to-tail shape in the form of a curve, curve parameters were obtained by fitting 4 times the curve based on the finite element simulation results. For widening rolling, only fitting the shape of the side edge which is close to the head and tail within the length range of 800 mm; for longitudinal rolling, the head and tail shapes in the half width range are fitted. One set of initial parameters for finite element simulation calculations: the inlet thickness is 100mm, the reduction rate is 20%, the width of a rolled piece is 2400mm, and a head fitting curve obtained by finite element rolling simulation is as follows: 0.7348x2+0.0516x3+9.67823x4The corresponding heights of the 3 key points are respectively as follows: 0.36863, 4.4609, and 21.216; the tail fitting curve is: 0.8206x2+0.057x3+10.6x4The corresponding heights of the 3 key points are respectively as follows: 0.406, 4.894, and 23.251.
3. Training of artificial neural networks on simulated data
An artificial neural network is established, and the input layer has 3 variables: inlet thickness, reduction rate and rolled piece width; the output layer has 3 variables corresponding to the height h of 3 key points selected on the metal flow curve1、h2And h3Specific neural network parameters are set as follows:
● neural network architecture: layer 3 BP neural network
● inputting the number of layer units 3; 20 hidden layer units and 3 output layer units
● learning rate eta is 0.4
● the momentum factor alpha is 0.3
FIG. 8 illustrates the convergence of the error of the neural network training process. Training finite element simulation resultsTraining, after 32000 times of training, the error of the network output is less than 1.0 × 10-6And the result meeting the precision requirement is obtained, and the trained neural network can predict the shape of the metal flow of the rolled piece with any specification after single-pass rolling.
4. Intelligent prediction of rolled piece plane shape according to rolling process procedures
For the 7-point plane shape control in the above process, the present invention divides 19 shape tracking points in half width according to the setting curve of the plane shape, wherein L1 is divided into 5 parts in length, L2 is divided into 5 parts in length, and the rest is divided into 8 parts. In the widening stage, overlapping the output curves of the side edges of the previous 2 times to obtain 4 times of curve parameters which are 800mm away from the head and the tail; considering plane shape control implemented at the last pass of the broadening stage, considering three influences of initial side shape, normal extension caused by plane shape setting and metal free flow output by a neural network when the outlet shape is predicted at the first pass of longitudinal rolling, and overlapping; starting from the 2 nd pass of longitudinal rolling, the variation factors influencing the plane shape of the rolled piece only comprise two superposition items of normal extension and head-tail metal free flow in each pass of reduction process.
And (3) overlapping the free flow results of head and tail metals of 19 tracking points in the width direction with the normal extension calculation values to obtain the head and tail shapes of the rolled piece after each pass of rolling. And (3) repeatedly and iteratively calculating according to the rolling schedule until the final calculation is finished, so that the final head and tail shapes of the rolled piece are obtained, and the change of the head and tail shapes in the rolling process of each longitudinal rolling pass is shown in a figure 9.
After multi-pass superposition calculation, the deviation between the curve height calculation result at the half-width position of the rolled piece and an actual measurement value is less than 5%, and because the width of the rolled piece is not considered in the calculation process, the shape prediction and the actual measurement close to the edge of the rolled piece have some differences, but the evaluation of the plane shape control parameters is not influenced. The intelligent prediction method developed above can be used for rapid prediction of the plane shape in the production process, and provides support for online optimization of plane shape parameters.
The above description is only for the preferred embodiment of the present invention, but the scope of the present invention is not limited thereto, and any person skilled in the art should be considered to be within the technical scope of the present invention, and the technical solutions and the inventive concepts thereof according to the present invention should be equivalent or changed within the scope of the present invention.