CN114602998A - Side roller setting method of roller straightener based on side roller neural network and fuzzy technology - Google Patents

Side roller setting method of roller straightener based on side roller neural network and fuzzy technology Download PDF

Info

Publication number
CN114602998A
CN114602998A CN202210195206.9A CN202210195206A CN114602998A CN 114602998 A CN114602998 A CN 114602998A CN 202210195206 A CN202210195206 A CN 202210195206A CN 114602998 A CN114602998 A CN 114602998A
Authority
CN
China
Prior art keywords
edge
roller
fuzzy
roll
residual curvature
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.)
Granted
Application number
CN202210195206.9A
Other languages
Chinese (zh)
Other versions
CN114602998B (en
Inventor
胡鹰
徐杰
胡鹏
宋婷
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Taiyuan University of Science and Technology
Original Assignee
Taiyuan University of Science and Technology
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Taiyuan University of Science and Technology filed Critical Taiyuan University of Science and Technology
Priority to CN202210195206.9A priority Critical patent/CN114602998B/en
Publication of CN114602998A publication Critical patent/CN114602998A/en
Application granted granted Critical
Publication of CN114602998B publication Critical patent/CN114602998B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B21—MECHANICAL METAL-WORKING WITHOUT ESSENTIALLY REMOVING MATERIAL; PUNCHING METAL
    • B21D—WORKING OR PROCESSING OF SHEET METAL OR METAL TUBES, RODS OR PROFILES WITHOUT ESSENTIALLY REMOVING MATERIAL; PUNCHING METAL
    • B21D1/00—Straightening, restoring form or removing local distortions of sheet metal or specific articles made therefrom; Stretching sheet metal combined with rolling
    • B21D1/02—Straightening, restoring form or removing local distortions of sheet metal or specific articles made therefrom; Stretching sheet metal combined with rolling by rollers
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B21—MECHANICAL METAL-WORKING WITHOUT ESSENTIALLY REMOVING MATERIAL; PUNCHING METAL
    • B21C—MANUFACTURE OF METAL SHEETS, WIRE, RODS, TUBES OR PROFILES, OTHERWISE THAN BY ROLLING; AUXILIARY OPERATIONS USED IN CONNECTION WITH METAL-WORKING WITHOUT ESSENTIALLY REMOVING MATERIAL
    • B21C51/00—Measuring, gauging, indicating, counting, or marking devices specially adapted for use in the production or manipulation of material in accordance with subclasses B21B - B21F
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • 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

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Molecular Biology (AREA)
  • Artificial Intelligence (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Mechanical Engineering (AREA)
  • Straightening Metal Sheet-Like Bodies (AREA)

Abstract

The invention relates to the technical field of straightener processes, in particular to a side roller setting method of a roller straightener based on a side roller neural network and a fuzzy technology, which is used for obtaining side roller samples of the existing straightener and constructing a sample library; building a straightening machine edge roller model frame based on an edge roller neural network; training and verifying the edge roller model of the straightening machine by using the samples in the sample library to obtain a trained edge roller model of the straightening machine; acquiring a new straightener side roll sample on line in real time as an input quantity, and predicting the lifting quantity of the side roll through the trained straightener side roll model; if the measured residual curvature of the plate is not 0, acquiring the correction quantity of the edge roll by using an edge roll fuzzy control system; and optimizing the edge roller model of the straightening machine by using the edge roller lifting amount and the edge roller correction amount to obtain an optimized intelligent control system of the edge roller of the straightening machine.

Description

Side roller setting method of roller straightener based on side roller neural network and fuzzy technology
Technical Field
The invention relates to the technical field of straightener processes, in particular to a method for setting side rollers of a roller type straightener.
Background
The edge roller control model of the straightening machine is a key technology of a straightening machine control system, but the process setting model of the edge roller is not mature so far, and currently, a worker mainly adjusts the edge roller on line, namely manually adjusts the edge roller according to the warping degree of the head and the tail of a plate entering the straightening machine, but the mode has low efficiency and poor precision, so that the method is of great importance in researching the edge roller setting method of the roller straightening machine.
This patent adopts limit roller neural network, and simple structure combines with production practice, adopts limit roller fuzzy control to correct simultaneously and goes into, and export limit roller lift volume can improve aligning ability and efficiency to defect panel head and the afterbody.
Disclosure of Invention
The invention provides a setting method for side rollers of a roller straightening machine based on a side roller neural network and a fuzzy technology, which solves the problem of low efficiency and poor precision caused by the current manual adjustment mode of the side rollers of the straightening machine.
The invention adopts the technical scheme that: a roller straightening machine edge roller setting method based on an edge roller neural network and a fuzzy technology comprises the following steps:
obtaining an edge roller sample of the existing straightener and constructing a sample library;
building a straightening machine edge roller model frame based on an edge roller neural network;
training and verifying the edge roller model of the straightening machine by using the samples in the sample library to obtain a trained edge roller model of the straightening machine;
acquiring a new straightener side roll sample on line in real time as an input quantity, and predicting the lifting quantity of the side roll through the trained straightener side roll model;
if the measured residual curvature of the plate is not 0, acquiring the correction quantity of the edge roll by using an edge roll fuzzy control system;
and optimizing the edge roller model of the straightening machine by using the edge roller lifting amount and the edge roller correction amount to obtain an optimized intelligent control system of the edge roller of the straightening machine.
Obtaining the samples of the edge rollers of the existing straightener, wherein each sample comprises the temperature, the thickness, the yield strength, the elastic modulus, the reduction, the warping degree of the plate head or the plate tail and the lifting amount of the corresponding inlet edge roller and outlet edge roller.
In the method for constructing the edge roller model frame of the straightener based on the edge roller neural network, parameters of 6 influences of the straightening edge rollers such as temperature, thickness, yield strength, elastic modulus, rolling reduction and warping degree of a plate head or a plate tail are used as input quantity of the edge roller model of the straightener, lifting quantity of corresponding inlet edge rollers and outlet edge rollers is used as output quantity of the edge roller model of the straightener, and the number of hidden layers is set to be 9 according to experience, namely, a three-layer edge roller neural network model of the edge rollers of the straightener is established, 6 units of an input layer, 9 units of a middle layer and 2 units of an output layer, a Sigmoid function is used for transfer functions of the input layer and the hidden layer of the edge roller model of the straightener, and a linear function is adopted for the output layer. Training and verifying the side roller model of the straightener by using the samples of the sample library, wherein the training and verifying comprises the following steps:
will initiate the weight w(1),w(2)And initial bias weight C(1)，C(2)Set to a random number with an empirical value between-0.4 and 0.4, where w(1),w(2)Each is the weight value of the hidden layer and the output layer of the edge roller neural network, C(1)，C(2)The offsets of the hidden layer and the output layer of the edge roller neural network are respectively, 60% of samples in the sample library are used as training samples, and the rest samples are used for verification;
randomly taking a group of straightener side roller samples, namely a P-th group of straightener side roller samples, from the plate head training samples, and transmitting straightener side roller sample information to an output layer of a straightener side roller model through forward propagation to obtain the output quantity of a side roller neural network, namely the lifting quantity of the inlet side roller and the outlet side roller;
the P group straightener side roll sample input vector of the network: i.e. iP＝[i1P,i2P,....i6P]TAnd the sample output vector of the side roller of the P group of the straightening machines is as follows: y isP＝[y1P,y2P]T；
When the edge roller sample of the p-th group of the straighteners is input, the actual edge roller lifting amount is as follows:
bP＝f1(w(1)iP+C(1))
dP＝f2(w(2)bP+C(2))
wherein i1P,i2P,....i6PRespectively the sample temperature, thickness, yield strength, elastic modulus, rolling reduction, warpage of the plate head or the plate tail, y1P,y2PTheoretical lifting amounts of the inlet edge roll and the outlet edge roll of the sample of the P group of edge rolls of the straightener, bP，dPOutputting the actual value of the layer for the edge roll sample hidden layer of the P-th group of the straightening machines; f1 is a Sigmoid function; f2 is a linear function;
utilizing theoretical lifting amount of sample inlet side roller and outlet side roller of the P-th group of straightener side roller and actual value of sample output layer of the P-th group of straightener side roller, performing back propagation by adopting error square sum, adopting error square sum as performance index, and setting EpTaking an L2 norm as an objective function of the edge roller neural network when the P-th group of samples are input
Figure BDA0003526076180000021
Wherein E isPThe error square sum of the sample of the side roller of the P group of the straightening machines is obtained; d1P，d2PActual lifting amounts of an inlet edge roller and an outlet edge roller of the sample of the edge roller of the P group of the straightening machines are respectively; y is1P，y2PRespectively obtaining theoretical lifting amounts of an inlet side roller and an outlet side roller of the sample of the side roller of the P group of the straightening machines; e1PThe difference value of the actual lifting amount of the edge roll at the sample inlet of the edge roll of the P group of the straightening machines and the theoretical value is obtained; e2PThe difference value of the actual lifting amount of the edge roll at the sample outlet of the edge roll of the P group of the straightening machines and the theoretical value is obtained;
Figure BDA0003526076180000022
Figure BDA0003526076180000023
the edge roll model training method of the straightener adopts a gradient descent method, wherein Wjk 1(n +1) is a weight value obtained by training the plate-oriented model for n +1 times from the kth neuron of the input layer to the jth neuron of the hidden layer, wherein the kth neuron is i in fig. 3kK is 1,2, …,6, wherein the jth neuron is that of FIG. 3
Figure BDA0003526076180000024
j is 1,2, …, 9; wjk 1(n) training the model for the plate header for n times to obtain weights from the kth neuron of the input layer to the jth neuron of the hidden layer; wpq 2(n +1) is the weight value obtained by training the model for n +1 times from the q-th neuron of the hidden layer to the p-th neuron of the output layer, wherein the q-th neuron of the hidden layer is the weight value of the q-th neuron of FIG. 3
Figure BDA0003526076180000025
q is 1,2, …,9, the p-th neuron of the output layer is that of FIG. 3
Figure BDA0003526076180000026
p is 1, 2; wpq 2(n) training the model for n times to obtain a weight value from the q-th neuron of the hidden layer to the p-th neuron of the output layer; taking delta as step length, and taking 0.001;
the maximum number of times of training the edge roll model of the straightener is set to 1000, namely
n≥1000
Performance target set to 1 × e-3I.e. by
EP(n)≤1×e-3
And if the error meets the performance target, the straightening machine side roller model is proved to be trained completely.
If the measured residual curvature of the sheet is not 0, obtaining a side roll correction amount using a side roll fuzzy control system, comprising:
if the measured residual curvature of the plate head or the plate tail is not 0, converting the residual curvature of the plate head or the plate tail and the change rate of the residual curvature from digital quantity into fuzzy quantity, then carrying out fuzzy reasoning on the obtained residual curvature and the fuzzy quantity of the change rate thereof by using a side roller fuzzy rule obtained by depending on practical experience, and finally converting the side roller fuzzy adjustment quantity obtained by a reasoning result into a side roller accurate adjustment quantity, which specifically comprises the following steps:
the method for establishing the edge roller fuzzy control system of the straightener comprises the following steps:
the first step is as follows: select edge roll fuzzy controller
Selecting 2 double-input single-output edge roller fuzzy controllers for a plate head, namely a controller I and a controller II, selecting residual curvature rc of the plate head and change rate rce of the residual curvature of the plate head as the input of the edge roller fuzzy controller I for the controller I, and recording the output of the controller I as adjustment quantity of an edge roller at an inlet of the correction plate head as es; for the controller II, selecting the residual curvature rc of the plate head and the change rate rce of the residual curvature of the plate head as the input of the edge roll fuzzy controller II, and outputting the residual curvature rc of the plate head and the change rate rce of the residual curvature of the plate head as the adjustment quantity of the edge roll at the first outlet of the correction plate, and recording the adjustment quantity as os;
selecting 2 double-input single-output edge roller fuzzy controllers which are respectively a controller III and a controller IV aiming at the plate tail, selecting the residual curvature rd of the plate tail and the change rate rde of the residual curvature of the plate tail as the input of the edge roller fuzzy controller III aiming at the controller III, and recording the output of the controller III as the adjustment quantity of the edge roller at the entrance of the correction plate tail, which is em; for the controller IV, selecting the residual curvature rd of the board tail and the change rate rde of the residual curvature of the board tail as the input of the edge roller fuzzy controller IV, and outputting the output as the adjustment quantity of the edge roller at the outlet of the correction board tail, which is recorded as us;
the second step: selecting fuzzy subsets for fuzzification
Selecting the residual curvature rc of the plate head and the residual curvature of the plate head aiming at the edge roller fuzzy controller I and the controller II of the plate headThe ambiguity domains of the rate of change rce of are each [ -3,3]Then the scale factor of the residual curvature rc of the plate head is
Figure BDA0003526076180000031
Is marked as KrcWherein, rcamaxMaximum value of the actual plate head residual curvature, rcaminThe minimum value of the residual curvature of the actual plate head is obtained; if the actual input board head residual curvature is
Figure BDA0003526076180000032
After discourse-domain linear transformation to
Figure BDA0003526076180000033
Then to ircFuzzification with a single point is noted as irc(r)(ii) a The rate of change of residual curvature rce of the board head has a scale factor of
Figure BDA0003526076180000034
Is marked as KrceWherein, rceamaxThe maximum value of the rate of change of the residual curvature of the actual board head, rceaminThe minimum value of the change rate of the actual plate head residual curvature is obtained; if the actually input change rate of the plate head residual curvature is
Figure BDA0003526076180000035
After discourse domain linear transformation into
Figure BDA0003526076180000036
Then to irceFuzzification by a single point is denoted as irce(i)(ii) a Defining 7 fuzzy subsets { NB, NM, NS, Z, PS, PM, PB } for each of the slab-start residual curvature rc and the slab-start residual curvature rate rce, wherein for the slab-start residual curvature, a larger absolute value of the negative portion indicates that the slab is bent downward, and a larger absolute value of the positive portion indicates that the slab is bent upward; adopting triangular membership functions for the residual curvature and the change rate of the plate head;
the knowledge base also has to define membership function for the first inlet edge roll adjustment es and the first outlet edge roll adjustment os of the output value correction plate according to the actual engineering processThe fuzzy domain of es and os is selected as [ -3,3 [ ]]Then the scale factor of es is
Figure BDA0003526076180000037
Is marked as KesWherein es isamaxFor the minimum value of the actual inlet edge roll adjustment, esaminThe minimum value of the adjustment quantity of the actual inlet side roller is obtained; the scaling factor of os is
Figure BDA0003526076180000038
Is marked as KosIn which osamaxFor the maximum value of the actual outlet edge roll adjustment, osaminThe minimum value of the adjustment amount of the edge roller at the actual outlet is obtained. 7 fuzzy subsets { NB, NM, NS, Z, PS, PM, PB } are defined respectively, and output membership functions of es and os all adopt Gaussian membership functions;
the third step: establishing a side roller fuzzy rule base
The rule base is summarized by field experience, and the fuzzy rules in the table 1 can be obtained for the edge roll fuzzy controller I; for the edge roll fuzzy controller II, the fuzzy rules of table 2 can be obtained;
the fourth step: fuzzy inference
The fuzzy inference method adopts a Mamdani algorithm, and for the rule of the third step, the total implication relation is obtained by the following formula:
Figure BDA0003526076180000039
Figure BDA00035260761800000310
the result after edge roller fuzzy reasoning is as follows:
Figure BDA00035260761800000311
Figure BDA00035260761800000312
wherein irc(r)Residual curvature of the plate head after blurring, irce(i)The rate of change of residual curvature of the blunted plate heads, μes(v)Fuzzy output value, mu, of edge roll fuzzy control Ios(v)Fuzzy output values of the edge roller fuzzy control II;
the fifth step: defuzzification
The output obtained by inference of the fuzzy rule of the edge roller is a fuzzy subset, defuzzification is carried out by adopting a gravity center method, and the actual adjustment value of the edge roller is obtained by domain-of-discourse inverse transformation, wherein the specific formula is as follows:
Figure BDA0003526076180000041
Figure BDA0003526076180000042
Figure BDA0003526076180000043
Figure BDA0003526076180000044
wherein Z isesFor precise output values, Z, of edge roll fuzzy control I after centrobaringosFor precise output value, u, of edge roll fuzzy control II after center of gravityesFor the actual plate first entry edge roll adjustment, u, corrected after the fuzzy controller IosActual board tail exit edge roll adjustment, es, corrected by fuzzy controller IIamaxFor the minimum value of the actual inlet edge roll adjustment, esaminFor the minimum value of the actual inlet edge roll adjustment, osamaxFor the maximum value of the actual outlet edge roll adjustment, osaminThe minimum value of the adjustment amount of the edge roller at the actual outlet is obtained.
The second through fifth steps are also adapted to the edge roll fuzzy controller III and the controller IV of the board end, and the residual curvature rd of the board end and the rate of change of the residual curvature rde of the board end are selected instead of the residual curvature rc of the board head and the rate of change of the residual curvature rce of the board head. The invention has the beneficial effects that: the method for setting the edge rollers of the roller straightening machine based on the edge roller neural network and the fuzzy technology solves the problem of low efficiency and poor precision caused by the conventional method for manually adjusting the edge rollers of the straightening machine, and improves the straightening capability and efficiency of the head and the tail of the defective plate.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required in the embodiments will be briefly described below, and for those skilled in the art, other drawings can be obtained according to the drawings without creative efforts.
FIG. 1 is a flow chart of a method of the present invention;
FIG. 2 is a reference drawing of the edge rolls of the leveler of the present invention:
FIG. 3 is a schematic diagram of a warp impact model of a header according to the present invention;
FIG. 4 is a schematic diagram of a warp impact model of the present invention;
FIG. 5 is a view showing a structure of a side roll fuzzy control of a leveler for a plate head according to the present invention;
FIG. 6 is a view showing a fuzzy control structure of edge rolls of a leveler for a plate tail according to the present invention;
FIG. 7 is a graph of membership functions for residual curvature and rate of change of the plate headers of the present invention;
FIG. 8 is a membership function graph of the adjustment amount of the inlet and outlet side rolls of the leveler for the header of the board according to the present invention;
wherein, A is an inlet edge roller, and B is an outlet edge roller.
Detailed description of the preferred embodiments
The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
According to the method, the straightening capability and efficiency of the head and the tail of the defective plate are improved by establishing the edge roller setting method of the roller straightening machine based on the edge roller neural network and the fuzzy technology.
The present invention will be described in further detail with reference to the drawings and detailed description, so that the objects, features and advantages of the invention can be more clearly understood.
As shown in FIG. 1, a method for setting edge rollers of a roller straightening machine based on an edge roller neural network and a fuzzy technology comprises the following steps:
101, obtaining a sample of a side roller of a straightening machine, and constructing a sample library;
102, building a straightener side roller model frame based on a side roller neural network;
103, training and verifying the side roller model of the straightener by using the samples in the sample library;
104, acquiring new input data in real time on line, and predicting the lifting amount of the side roller through the trained edge roller model of the straightener;
step 105, if the measured residual curvature of the plate is not 0, utilizing a side roller fuzzy control system to obtain a side roller correction quantity;
and 106, optimizing the edge roller model of the straightener by using the edge roller lifting amount and the edge roller correction amount to obtain an optimized edge roller intelligent system.
Through the steps, the lifting amount of the side roller is adjusted through a side roller setting method of the roller straightening machine based on a side roller neural network and a fuzzy technology, so that the problem of low efficiency and poor precision caused by the conventional method for manually adjusting the side roller of the straightening machine can be solved, and a novel side roller setting method of the roller straightening machine is provided, so that the straightening capacity and efficiency of the head and tail parts of the defective plate can be improved.
Wherein, steps 102 to 104 specifically include:
as shown in figures 3 and 4, the head and tail of the plate have many factors influencing the straightening effect in the straightening process, and the patent takes the main factors influencing the straightening effect: the temperature, the thickness, the yield strength, the elastic modulus, the rolling reduction and the warping degree of the board head or the board tail are controlled by the temperature, the thickness, the yield strength, the elastic modulus, the rolling reduction and the lifting amount of the inlet side roller and the outlet side roller of the board head or the board tail. The input layers of the BP edge roll neural network for the plate head or plate tail are set to 6 and the output layers are set to 2.
The edge roll neural networks designed by the patent for the head and the tail of the plate respectively comprise an implicit layer, and the node number utilizes an empirical formula
Figure BDA0003526076180000051
Giving an estimated value, m, n are the numbers of neurons of the input layer and the output layer, respectively, and a is [0, 10 ]]The constant between the two is taken as 6, the number of hidden layers of the BP edge roller neural network designed by the patent is all
Figure BDA0003526076180000052
The number of hidden layers is 9.
The transfer functions of the input layer and the hidden layer of the edge roller model of the straightener use Sigmoid functions, and the output layer also uses linear functions. The performance indexes of the edge roller model of the straightener adopt the sum of squares of errors.
The edge roll model of the straightener for the plate head is provided with E as shown in figure 3pTaking L2 norm as the target function of the edge roller neural network when the P-th group of samples are input
Figure BDA0003526076180000053
Wherein E isPThe error square sum of the sample of the side roller of the P group of the straightening machines is obtained; d is a radical of1P，d2PActual lifting amounts of an inlet edge roller and an outlet edge roller of the sample of the edge roller of the P group of the straightening machines are respectively; y is1P，y2PRespectively obtaining theoretical lifting amounts of an inlet side roller and an outlet side roller of the sample of the side roller of the P group of the straightening machines; e1PThe difference value of the actual lifting amount of the edge roll at the sample inlet of the edge roll of the P group of the straightening machines and the theoretical value is obtained; e2PThe difference value of the actual lifting amount of the edge roll at the sample outlet of the edge roll of the P group of the straightening machines and the theoretical value is obtained;
straightening machine for board tailSide roll model, as shown in FIG. 4, let vpTaking L2 norm as the target function of the edge roller neural network when the P-th group of samples are input
Figure BDA0003526076180000054
Wherein v isPThe error square sum of the sample of the side roller of the P group of the straightening machines is obtained; (o)1P，o2PActual lifting amounts of an inlet edge roller and an outlet edge roller of the sample of the edge roller of the P group of the straightening machines are respectively; h is1P，h2PRespectively obtaining theoretical lifting amounts of an inlet side roller and an outlet side roller of the sample of the side roller of the P group of the straightening machines; v. of1PThe difference value of the actual lifting amount of the edge roll at the sample inlet of the edge roll of the P group of the straightening machines and the theoretical value is obtained; v. of2PThe difference value of the actual lifting amount of the edge roll at the outlet of the sample of the edge roll of the P group of the straightening machines and a theoretical value is obtained;
and further setting an input layer of the edge roller model of the straightener to the hidden layer, assigning a connecting weight value from the hidden layer to the output layer, and setting the initial weight value and the initial bias weight value as random numbers with empirical values between-0.4 and 0.4. The training method all uses a gradient descent method, i.e.
Figure BDA0003526076180000061
Figure BDA0003526076180000062
Wherein, Wjk 1(n +1) is a weight value obtained by training the model aiming at the beginning of the board for n +1 times from the kth neuron of the input layer to the jth neuron of the hidden layer, wherein the kth neuron is i of fig. 3kK is 1,2, …,6, wherein the jth neuron is that of FIG. 3
Figure BDA0003526076180000063
j is 1,2, …, 9; wjk 1(n) for the headboardTraining the weight value obtained from the kth neuron of the input layer to the jth neuron of the hidden layer for n times by the model; wpq 2(n +1) is the weight value obtained by training the model for n +1 times from the q-th neuron of the hidden layer to the p-th neuron of the output layer, wherein the q-th neuron of the hidden layer is the weight value of the q-th neuron of FIG. 3
Figure BDA0003526076180000064
q is 1,2, …,9, the p-th neuron of the output layer is the neuron of figure 3
Figure BDA0003526076180000065
p is 1, 2; wpq 2(n) training the model for n times to obtain a weight value from the q-th neuron of the hidden layer to the p-th neuron of the output layer; delta is taken as the step length, and 0.001 is taken;
Figure BDA0003526076180000066
Figure BDA0003526076180000067
wherein Q isjk 1(n +1) is the weight value obtained by training the model for the plate tail n +1 times from the kth neuron of the input layer to the jth neuron of the hidden layer, wherein the kth neuron is e in FIG. 4kK is 1,2, …,6, wherein the jth neuron is that of FIG. 4
Figure BDA0003526076180000068
j is 1,2, …, 9; qjk 1(n) training the model aiming at the plate tail for n times to obtain weights from the kth neuron of the input layer to the jth neuron of the hidden layer; qpq 2(n +1) is the weight value obtained by training the model for n +1 times from the q-th neuron of the hidden layer to the p-th neuron of the output layer, wherein the q-th neuron of the hidden layer is the weight value of the q-th neuron of FIG. 4
Figure BDA0003526076180000069
q is 1,2, …,9, the p-th neuron of the output layer is the neuron of figure 4
Figure BDA00035260761800000610
p is 1, 2; qpq 2(n) training the model for n times to obtain a weight value from the q-th neuron of the hidden layer to the p-th neuron of the output layer; taking delta as step length, and taking 0.001;
the maximum number of times of training the edge roll model of the straightener is set to 1000, namely
n≥1000
The performance targets are all set to 1 × e-3I.e. by
EP(n)≤1×e-3
vP(n)≤1×e-3
And if the error meets the performance target, the straightening machine side roller model is proved to be trained completely.
Steps 105 and 106 include:
as shown in fig. 5 and 6, for the edge roll fuzzy control structure of the head and the tail of the plate, the residual curvature and the change rate thereof of the head and the tail of the plate are converted into fuzzy quantities from digital quantities, then the obtained fuzzy quantities of the residual curvature and the change rate thereof are subjected to fuzzy reasoning by an edge roll fuzzy rule base which is obtained by depending on practical experience and aims at the inlet and the outlet of the head and the tail of the plate, and finally the edge roll fuzzy inlet and the outlet edge roll adjustment quantity of the head and the tail of the plate, which are respectively obtained by reasoning results, are converted into accurate quantities. The method comprises the following specific steps in sequence: (a) selecting a side roller fuzzy controller; (b) selecting a fuzzy subset for fuzzification; (c) establishing a side roller fuzzy rule base; (d) fuzzy reasoning; (e) and (4) defuzzification.
Wherein, the method in the step (a) comprises the following steps:
selecting 2 double-input single-output edge roller fuzzy controllers for a plate head, namely a controller I and a controller II, selecting residual curvature rc of the plate head and change rate rce of the residual curvature of the plate head as the input of the edge roller fuzzy controller I for the controller I, and recording the output of the controller I as adjustment quantity of an edge roller at an inlet of the correction plate head as es; the controller II selects the stock nose residual curvature rc and the stock nose residual curvature rate rce as inputs to the edge roll blur controller II and outputs a corrected stock nose edge roll adjustment, denoted as os. Edge roll blur control for the leading sheet is shown in fig. 5.
Selecting 2 double-input single-output edge roller fuzzy controllers which are respectively a controller III and a controller IV aiming at the board tail, selecting the residual curvature rd of the board tail and the change rate rde of the residual curvature of the board tail as the input of the edge roller fuzzy controller III aiming at the controller III, and recording the output of the controller III as the adjustment quantity of the edge roller at the entrance of the correction board tail as em; for controller IV, the residual curvature rd of the tail and the rate of change of the residual curvature of the tail rde are selected as the input to edge roll fuzzy controller IV, and the output is the corrected tail exit edge roll adjustment, denoted us. Edge roll blur control for the plate tail is shown in fig. 6.
The method in the step (b) comprises the following steps:
for edge roll fuzzy controller I and controller II of board head, fuzzy domain of residual curvature rc of board head and change rate rce of residual curvature of board head are respectively [ -3,3]Then the scale factor of the residual curvature rc of the plate head is
Figure BDA0003526076180000071
Is marked as KrcWherein rc isamaxIs the maximum value of the actual bow residual curvature, rcaminIs the minimum value of the residual curvature of the actual plate head. If the actual input board head residual curvature is
Figure BDA0003526076180000072
After discourse domain linear transformation into
Figure BDA0003526076180000073
Then to ircFuzzification by a single point is denoted as irc(r). The rate of change of residual curvature rce of the board head has a scale factor of
Figure BDA0003526076180000074
Is marked as KrceWherein, rceamaxThe maximum value of the rate of change of the residual curvature of the actual board head, rceaminThe minimum value of the change rate of the residual curvature of the actual plate head. If the actually input change rate of the plate head residual curvature is
Figure BDA0003526076180000075
After discourse-domain linear transformation to
Figure BDA0003526076180000076
Then to irceFuzzification by a single point is denoted as irce(i). For the residual curvature rc of the lead and the rate of change rce of the residual curvature of the lead, 7 fuzzy subsets { NB, NM, NS, Z, PS, PM, PB } are each defined, wherein for the residual curvature of the lead, a larger absolute value for a negative portion indicates a more downward bending of the sheet, and a larger absolute value for a positive portion indicates a more upward bending of the sheet. The membership functions of the residual curvature rc of the plate head and the change rate rce thereof obtained from practical experience are shown in fig. 7, and the patent adopts triangular membership functions for the residual curvature of the plate head and the change rate thereof.
For edge roll fuzzy controller III and controller IV of the strip tail, fuzzy domains of the residual curvature rd of the strip tail and the change rate rde of the residual curvature of the strip tail are selected to be [ -3,3]Then the scale factor of the residual curvature rd of the plate tail is
Figure BDA0003526076180000077
Is marked as KrdWherein, rdamaxIs the maximum value of the residual curvature of the actual plate tail, rdaminIs the minimum value of the residual curvature of the actual plate tail. If the actually input board tail residual curvature is
Figure BDA0003526076180000078
After discourse domain linear transformation into
Figure BDA0003526076180000079
Then to irdFuzzification by a single point is denoted as ird(r). The rate of change of residual curvature rde of the board tail has a scale factor of
Figure BDA00035260761800000710
Is marked as KrdeWherein, rdeamaxRde, which is the maximum value of the rate of change of the residual curvature of the actual plate tailaminThe minimum value of the change rate of the residual curvature of the actual plate tail is obtained. If the actually input board tail residual curvature change rate is
Figure BDA00035260761800000711
After discourse domain linear transformation into
Figure BDA00035260761800000712
Then to irdeFuzzification by a single point is denoted as irde(i). 7 fuzzy subsets NB, NM, NS, Z, PS, PM, PB are defined for each of the residual curvature rd of the plate tail and the rate of change rde of the residual curvature of the plate head, and the membership functions of the residual curvature rd of the plate tail and its rate of change rde are shown in fig. 7 from practical experience.
The knowledge base also has to carry out membership function definition on the first inlet edge roll adjustment quantity es and the first outlet edge roll adjustment quantity os of the output value correction plate, and the fuzzy domain of es and os is selected to be [ -3,3 ] according to the actual engineering experience]Then the scale factor of es is
Figure BDA0003526076180000081
Is marked as KesWherein es isamaxFor the maximum value of the adjustment of the edge roll at the first entrance of the actual plate, esaminThe minimum value of the adjustment amount of the edge roller at the first inlet of the actual plate is obtained. The scaling factor of os is
Figure BDA0003526076180000082
Is marked as KosIn which osamaxFor the maximum value of the adjustment of the edge rolls at the initial outlet of the actual plate, osaminAnd the adjustment amount of the edge roller at the first outlet of the actual plate is the minimum. And 7 fuzzy subsets { NB, NM, NS, Z, PS, PM, PB } are defined respectively, and output membership functions of es and os are shown in FIG. 8 and adopt Gaussian membership functions. Meanwhile, membership function definition must be carried out on the adjustment quantity em of the edge roller at the inlet of the output value correction plate tail and the adjustment quantity us of the edge roller at the outlet of the correction plate tail, the fuzzy domain of em and us is selected to be [ -3,3]then em has a scale factor of
Figure BDA0003526076180000083
Is marked as KemTherein, emamaxFor the maximum value of the adjustment amount of the edge roll at the entrance of the actual plate tail emaminAnd the minimum value of the adjustment quantity of the edge roller at the inlet of the actual plate tail is obtained. us has a scale factor of
Figure BDA0003526076180000084
Is marked as KusWherein, usamaxFor the maximum value of the adjustment quantity us of the edge roll at the outlet of the actual plate tailaminThe minimum value of the adjustment quantity of the edge roller at the outlet of the actual plate tail is obtained. And each defines 7 fuzzy subsets NB, NM, NS, Z, PS, PM, PB, with the output em and us membership functions also shown in fig. 8.
The rule base of step (c) is summarized by field experience, and the following rules can be obtained:
for the edge roll fuzzy controller I, the following fuzzy rule table is provided:
TABLE 1 fuzzy rule table of edge roll fuzzy controller
Figure BDA0003526076180000085
For edge roll fuzzy controller II, the following fuzzy rule table is as follows:
TABLE 2 fuzzy rule Table of edge roll fuzzy controller
Figure BDA0003526076180000086
Figure BDA0003526076180000091
For edge roll fuzzy controller III, the following fuzzy rule table is as follows:
TABLE 3 fuzzy rule table of edge roll fuzzy controller III
Figure BDA0003526076180000092
For the edge roll fuzzy controller IV, the following fuzzy rule table is as follows:
TABLE 4 fuzzy rule table of edge roll fuzzy controller VI
Figure BDA0003526076180000093
Figure BDA0003526076180000101
And (d) adopting a Mamdani algorithm as the fuzzy inference method in the step (d). For the rule of step c, the total implication is found by the following equation:
Figure BDA0003526076180000102
Figure BDA0003526076180000103
Figure BDA0003526076180000104
Figure BDA0003526076180000105
the result after edge roller fuzzy reasoning is as follows:
Figure BDA0003526076180000106
Figure BDA0003526076180000107
Figure BDA0003526076180000108
Figure BDA0003526076180000109
wherein irc(r)Residual curvature of the blunted plate head, irce(i)Is the rate of change of residual curvature of the blunted plate head, ird(r)Residual curvature of the plate tail after blurring irde(i)The rate of change of residual curvature, μ, of the plate tail after blurringes(v)Fuzzy output value, mu, of edge roll fuzzy control Ios(v)Fuzzy output value mu of edge roller fuzzy control IIem(v)Is the fuzzy output value, mu, of the edge roll fuzzy control IIIus(v)The fuzzy output value of the edge roller fuzzy control IV.
And (e) the output obtained by the fuzzy rule reasoning of the edge roller is a fuzzy subset, the fuzzy subset is defuzzified by adopting a gravity center method, and the actual output value of the edge roller is obtained through domain inverse transformation. The concrete formula is as follows:
Figure BDA00035260761800001010
Figure BDA00035260761800001011
Figure BDA00035260761800001012
Figure BDA0003526076180000111
Figure BDA0003526076180000112
Figure BDA0003526076180000113
Figure BDA0003526076180000114
Figure BDA0003526076180000115
wherein, ZesFor precise output values, Z, of edge roll fuzzy control I after centrobaringosPrecise output value, Z, of edge roll fuzzy control II after center of gravityemFor precise output values, Z, of edge roll fuzzy control III after centringusFor precise output value, u, of edge roll fuzzy control IV after centrobaringesFor the actual plate first entry edge roll adjustment, u, corrected after the fuzzy controller IosActual plate initial exit edge roll adjustment u corrected by fuzzy controller IIemFor the actual plate tail inlet edge roll adjustment, u, corrected after the fuzzy controller IIIusActual board tail exit edge roll adjustment, es, corrected by fuzzy controller IVamaxFor the minimum value of the actual inlet edge roll adjustment, esaminFor the minimum value of the actual inlet edge roll adjustment, osamaxFor the maximum value of the actual outlet edge roll adjustment, osaminThe minimum value of the adjustment amount of the edge roller is the actual outlet.

Claims (6)

1. A roller straightening machine edge roller setting method based on an edge roller neural network and a fuzzy technology is characterized by comprising the following steps:
obtaining an edge roller sample of the existing straightener and constructing a sample library;
building a straightening machine edge roller model frame based on an edge roller neural network;
training and verifying the edge roller model of the straightening machine by using the samples in the sample library to obtain a trained edge roller model of the straightening machine;
acquiring a new straightener side roll sample on line in real time as an input quantity, and predicting the lifting quantity of the side roll through the trained straightener side roll model;
if the measured residual curvature of the plate is not 0, acquiring the correction quantity of the edge roll by using an edge roll fuzzy control system;
and optimizing the edge roller model of the straightening machine by using the edge roller lifting amount and the edge roller correction amount to obtain an optimized intelligent control system of the edge roller of the straightening machine.
2. The method for setting edge rolls of a roll leveler as set forth in claim 1, wherein: obtaining the samples of the edge rollers of the existing straightener, wherein each sample comprises the temperature, the thickness, the yield strength, the elastic modulus, the reduction, the warping degree of the plate head or the plate tail and the lifting amount of the corresponding inlet edge roller and outlet edge roller.
3. The method for setting edge rolls of a roll leveler as set forth in claim 2, wherein: in the method for constructing the edge roller model frame of the straightener based on the edge roller neural network, parameters of 6 influences of the straightening edge rollers such as temperature, thickness, yield strength, elastic modulus, rolling reduction and warping degree of a plate head or a plate tail are used as input quantity of the edge roller model of the straightener, lifting quantity of corresponding inlet edge rollers and outlet edge rollers is used as output quantity of the edge roller model of the straightener, and the number of hidden layers is set to be 9 according to experience, namely, a three-layer edge roller neural network model of the edge rollers of the straightener is established, 6 units of an input layer, 9 units of a middle layer and 2 units of an output layer, a Sigmoid function is used for transfer functions of the input layer and the hidden layer of the edge roller model of the straightener, and a linear function is adopted for the output layer.
4. The method for setting edge rolls of a roll leveler as set forth in claim 3, wherein training and verifying the side roll model of the leveler with samples of the sample library comprises:
will initiate the weight w(1),w(2)And initial bias weight C(1)，C(2)Set to an empirical value of-0A random number between 4 and 0.4, wherein w(1),w(2)Each is the weight value of the hidden layer and the output layer of the edge roller neural network, C(1)，C(2)Respectively offsetting a hidden layer and an output layer of the edge roller neural network, taking 60% of samples in the sample library as training samples, and using the rest samples for verification;
randomly taking a group of straightener side roller samples, namely a P-th group of straightener side roller samples, from the plate head training samples, and transmitting straightener side roller sample information to an output layer of a straightener side roller model through forward propagation to obtain the output quantity of a side roller neural network, namely the lifting quantity of the inlet side roller and the outlet side roller;
sample input vectors for the P-th set of straightener side rolls of the network: i.e. iP＝[i1P,i2P,....i6P]TAnd the sample output vector of the side roller of the P group of the straightening machines is as follows: y isP＝[y1P,y2P]T；
When the edge roller sample of the p-th group of the straighteners is input, the actual edge roller lifting amount is as follows:
bP＝f1(w(1)iP+C(1))
dP＝f2(w(2)bP+C(2))
wherein i1P,i2P,....i6PRespectively the sample temperature, thickness, yield strength, elastic modulus, rolling reduction, warpage of the plate head or the plate tail, y1P,y2PTheoretical lifting amounts of the inlet edge roll and the outlet edge roll of the sample of the P group of edge rolls of the straightener, bP，dPOutputting the actual value of the layer for the edge roll sample hidden layer of the P-th group of the straightening machines; f1 is Sigmoid function; f2 is a linear function;
utilizing theoretical lifting amount of sample inlet side roller and outlet side roller of the P-th group of straightener side roller and actual value of sample output layer of the P-th group of straightener side roller, performing back propagation by adopting error square sum, adopting error square sum as performance index, and setting EpTaking an L2 norm as an objective function of the edge roller neural network when the P-th group of samples are input
Figure FDA0003526076170000021
Wherein E isPThe error square sum of the sample of the side roller of the P group of the straightening machines is obtained; d1P，d2PActual lifting amounts of an inlet edge roller and an outlet edge roller of the sample of the edge roller of the P group of the straightening machines are respectively; y is1P，y2PRespectively obtaining theoretical lifting amounts of an inlet side roller and an outlet side roller of the sample of the side roller of the P group of the straightening machines; e1PThe difference value of the actual lifting amount of the edge roll at the sample inlet of the edge roll of the P group of the straightening machines and the theoretical value is obtained; e2PThe difference value of the actual lifting amount of the edge roll at the sample outlet of the edge roll of the P group of the straightening machines and the theoretical value is obtained;
Figure FDA0003526076170000022
Figure FDA0003526076170000023
the edge roll model training method of the straightener adopts a gradient descent method, wherein Wjk 1(n +1) is a weight value obtained by training the model aiming at the beginning of the board for n +1 times from the kth neuron of the input layer to the jth neuron of the hidden layer, wherein the kth neuron is i of fig. 3kK is 1,2, …,6, wherein the jth neuron is that of FIG. 3
Figure FDA0003526076170000031
j is 1,2, …, 9; wjk 1(n) training the model for the plate header for n times to obtain weights from the kth neuron of the input layer to the jth neuron of the hidden layer; w is a group ofpq 2(n +1) is the weight value obtained by training the model for n +1 times from the q-th neuron of the hidden layer to the p-th neuron of the output layer, wherein the q-th neuron of the hidden layer is a graph3 is
Figure FDA0003526076170000032
q is 1,2, …,9, the p-th neuron of the output layer is that of FIG. 3
Figure FDA0003526076170000033
p is 1, 2; wpq 2(n) training the model for n times to obtain a weight value from the q-th neuron of the hidden layer to the p-th neuron of the output layer; taking delta as step length, and taking 0.001;
the maximum number of times of training of the edge roll model of the straightener is set to 1000, namely
n≥1000
Performance target set to 1 × e-3I.e. by
EP(n)≤1×e-3
And if the error meets the performance target, the straightening machine side roller model is proved to be trained completely.
5. The method for setting edge rolls of a roll leveler as set forth in claim 4, wherein: if the measured residual curvature of the sheet is not 0, obtaining a side roll correction using a side roll fuzzy control system, comprising:
if the measured residual curvature of the plate head or the plate tail is not 0, converting the residual curvature of the plate head or the plate tail and the change rate of the residual curvature from digital quantity into fuzzy quantity, then carrying out fuzzy reasoning on the obtained residual curvature and the fuzzy quantity of the change rate by using a side roller fuzzy rule obtained by depending on practical experience, and finally converting the side roller fuzzy adjustment quantity obtained by a reasoning result into a side roller accurate adjustment quantity, which specifically comprises the following steps:
the method for establishing the edge roller fuzzy control system of the straightener comprises the following steps:
the first step is as follows: fuzzy controller for edge roller selection
Selecting 2 double-input single-output edge roller fuzzy controllers for a plate head, namely a controller I and a controller II, selecting residual curvature rc of the plate head and change rate rce of the residual curvature of the plate head as the input of the edge roller fuzzy controller I for the controller I, and recording the output of the controller I as adjustment quantity of an edge roller at an inlet of the correction plate head as es; for the controller II, selecting the residual curvature rc of the plate head and the change rate rce of the residual curvature of the plate head as the input of the edge roll fuzzy controller II, and outputting the residual curvature rc of the plate head and the change rate rce of the residual curvature of the plate head as the adjustment quantity of the edge roll at the first outlet of the correction plate, and recording the adjustment quantity as os;
selecting 2 double-input single-output edge roller fuzzy controllers which are respectively a controller III and a controller IV aiming at the plate tail, selecting the residual curvature rd of the plate tail and the change rate rde of the residual curvature of the plate tail as the input of the edge roller fuzzy controller III aiming at the controller III, and recording the output of the controller III as the adjustment quantity of the edge roller at the entrance of the correction plate tail, which is em; for the controller IV, selecting the residual curvature rd of the board tail and the change rate rde of the residual curvature of the board tail as the input of the edge roller fuzzy controller IV, and outputting the residual curvature rd and the change rate us as the adjustment quantity of the edge roller at the outlet of the correction board tail;
the second step: selecting fuzzy subsets for fuzzification
For edge roll fuzzy controller I and controller II of the board head, fuzzy domains of residual curvature rc of the board head and change rate rce of the residual curvature of the board head are selected to be [ -3,3]Then the scale factor of the residual curvature rc of the plate head is
Figure FDA0003526076170000041
Is marked as KrcWherein rc isamaxIs the maximum value of the actual bow residual curvature, rcaminThe minimum value of the residual curvature of the actual plate head is obtained; if the actual input board head residual curvature is
Figure FDA0003526076170000042
After discourse domain linear transformation into
Figure FDA0003526076170000043
Then to ircFuzzification with a single point is noted as irc(r)(ii) a The rate of change of the residual curvature of the plate head rce has a scale factor of
Figure FDA0003526076170000044
Is marked as KrceWherein, rceamaxMaximum value of the rate of change of the residual curvature of the actual plate head, rceaminThe minimum value of the change rate of the actual plate head residual curvature is obtained; if the actually input change rate of the plate head residual curvature is
Figure FDA0003526076170000045
After discourse domain linear transformation into
Figure FDA0003526076170000046
Figure FDA0003526076170000047
Then to irceFuzzification with a single point is noted as irce(i)(ii) a Defining 7 fuzzy subsets { NB, NM, NS, Z, PS, PM, PB } for each of the slab-start residual curvature rc and the slab-start residual curvature rate rce, wherein for the slab-start residual curvature, a larger absolute value of the negative portion indicates that the slab is bent downward, and a larger absolute value of the positive portion indicates that the slab is bent upward; adopting triangular membership functions for the residual curvature and the change rate of the plate head;
the knowledge base also has to carry out membership function definition on the first inlet edge roll adjustment quantity es and the first outlet edge roll adjustment quantity os of the output value correction plate, and the fuzzy domain of es and os is selected to be [ -3,3 ] according to the actual engineering experience]Then the scale factor of es is
Figure FDA0003526076170000048
Is marked as KesWherein es isamaxFor the minimum value of the actual inlet edge roll adjustment, esaminThe minimum value of the adjustment quantity of the actual inlet side roller is obtained; the scaling factor of os is
Figure FDA0003526076170000049
Is marked as KosIn which osamaxFor the maximum value of the actual outlet edge roll adjustment, osaminFor actual outlet edge roll adjustmentThe minimum value of the quantity. 7 fuzzy subsets { NB, NM, NS, Z, PS, PM, PB } are defined respectively, and output membership functions of es and os all adopt Gaussian membership functions;
the third step: establishing a side roller fuzzy rule base
The rule base is summarized by field experience, and the fuzzy rules in the table 1 can be obtained for the edge roll fuzzy controller I; for the edge roll fuzzy controller II, the fuzzy rules of table 2 can be obtained;
the fourth step: fuzzy inference
The fuzzy inference method adopts a Mamdani algorithm, and for the rule of the third step, the total implication relation is obtained by the following formula:
Figure FDA0003526076170000051
Figure FDA0003526076170000052
the result after edge roller fuzzy reasoning is as follows:
μes(v)＝(irc(r)×irce(i))°R(rc,rce,es)
μos(v)＝(irc(r)×irce(i))°P(rs,rce,os)
wherein irc(r)Residual curvature of the blunted plate head, irce(i)The rate of change of residual curvature of the blunted plate heads, μes(v)Fuzzy output value, mu, of edge roll fuzzy control Ios(v)Fuzzy output values of the edge roller fuzzy control II;
the fifth step: defuzzification
The output obtained by inference of the fuzzy rule of the edge roller is a fuzzy subset, defuzzification is carried out by adopting a gravity center method, and the actual adjustment value of the edge roller is obtained by domain-of-discourse inverse transformation, wherein the specific formula is as follows:
Figure FDA0003526076170000053
Figure FDA0003526076170000054
Figure FDA0003526076170000055
Figure FDA0003526076170000056
wherein Z isesFor precise output values, Z, of edge roll fuzzy control I after centrobaringosFor precise output value, u, of edge roll fuzzy control II after center of gravityesFor the actual plate first entry edge roll adjustment, u, corrected after the fuzzy controller IosActual board tail exit edge roll adjustment, es, corrected by fuzzy controller IIamaxFor the minimum value of the actual inlet edge roll adjustment, esaminFor the minimum value of the actual inlet edge roll adjustment, osamaxFor the maximum value of the actual outlet edge roll adjustment, osaminThe minimum value of the adjustment amount of the edge roller at the actual outlet is obtained.
6. The method for setting edge rolls of a roll leveler as set forth in claim 5, wherein: the second through fifth steps are also adapted to the edge roll fuzzy controller III and the controller IV of the board end, and the residual curvature rd of the board end and the rate of change of the residual curvature rde of the board end are selected instead of the residual curvature rc of the board head and the rate of change of the residual curvature rce of the board head.
CN202210195206.9A 2022-03-01 2022-03-01 Edge roll setting method of roll straightener based on edge roll neural network and fuzzy technology Active CN114602998B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210195206.9A CN114602998B (en) 2022-03-01 2022-03-01 Edge roll setting method of roll straightener based on edge roll neural network and fuzzy technology

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210195206.9A CN114602998B (en) 2022-03-01 2022-03-01 Edge roll setting method of roll straightener based on edge roll neural network and fuzzy technology

Publications (2)

Publication Number Publication Date
CN114602998A true CN114602998A (en) 2022-06-10
CN114602998B CN114602998B (en) 2023-08-15

Family

ID=81862048

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210195206.9A Active CN114602998B (en) 2022-03-01 2022-03-01 Edge roll setting method of roll straightener based on edge roll neural network and fuzzy technology

Country Status (1)

Country Link
CN (1) CN114602998B (en)

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104368634A (en) * 2014-10-21 2015-02-25 南京钢铁股份有限公司 Method for expanding straightening capacity of straightening machine through side rollers
CN108380700A (en) * 2018-03-09 2018-08-10 太原科技大学 Optimize the roll straightening processing parameter setting method of head-tail aligning
CN108500083A (en) * 2018-03-09 2018-09-07 太原科技大学 The more stock board product roll leveller processing parameter setting methods of small lot
CN111611699A (en) * 2020-05-14 2020-09-01 太原科技大学 Intelligent straightening machine for acquiring residual stress distribution characteristics of plates
US20210268564A1 (en) * 2020-02-28 2021-09-02 Honda Motor Co., Ltd. Methods and systems for automatically adjusting operational parameters of one or more leveling machines

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104368634A (en) * 2014-10-21 2015-02-25 南京钢铁股份有限公司 Method for expanding straightening capacity of straightening machine through side rollers
CN108380700A (en) * 2018-03-09 2018-08-10 太原科技大学 Optimize the roll straightening processing parameter setting method of head-tail aligning
CN108500083A (en) * 2018-03-09 2018-09-07 太原科技大学 The more stock board product roll leveller processing parameter setting methods of small lot
US20210268564A1 (en) * 2020-02-28 2021-09-02 Honda Motor Co., Ltd. Methods and systems for automatically adjusting operational parameters of one or more leveling machines
CN111611699A (en) * 2020-05-14 2020-09-01 太原科技大学 Intelligent straightening machine for acquiring residual stress distribution characteristics of plates

Also Published As

Publication number Publication date
CN114602998B (en) 2023-08-15

Similar Documents

Publication Publication Date Title
Wang et al. Prediction of bending force in the hot strip rolling process using artificial neural network and genetic algorithm (ANN-GA)
CN108637020B (en) Self-adaptive variation PSO-BP neural network strip steel convexity prediction method
US20210149349A9 (en) Networked control system time-delay compensation method based on predictive control
CN101391268B (en) Reverse optimization method of steel plate rolling and cooling controlling-process temperature institution
CN100520786C (en) Processing parameter setting method of tension straightening machine set correct roller
WO2023130666A1 (en) Strip steel plate convexity prediction method based on data-driving and mechanism model fusion
CN110347192B (en) Glass furnace temperature intelligent prediction control method based on attention mechanism and self-encoder
CN103745101A (en) Improved neural network algorithm based forecasting method of set value of rolling force of medium plate
CN106825069B (en) A kind of cold-strip steel high precision plates shape surface roughness on-line intelligence control method
CN113102516B (en) Hot continuous rolling strip steel head width prediction method integrating rolling mechanism and deep learning
CN115608793B (en) Finish rolling temperature regulation and control method for mechanism fusion data
CN111079926B (en) Equipment fault diagnosis method with self-adaptive learning rate based on deep learning
CN106991242A (en) A kind of control method of plate property optimization
CN111241750A (en) BP network cold-rolled strip steel mechanical property prediction method combining genetic algorithm
CN114580286A (en) Neural network model optimization method based on stainless steel ultra-thin strip annealing process
CN111290282B (en) Predictive control method for thermal power generating unit coordination system
CN114192587B (en) Cold-rolled plate shape control parameter setting method based on error back propagation algorithm
CN114602998A (en) Side roller setting method of roller straightener based on side roller neural network and fuzzy technology
CN116689502A (en) Roll bending force parameter setting method of finishing mill group process of hot continuous rolling mill based on SA-SCSO-SCN neural network
CN116343360A (en) Continuous prediction method for longitudinal thickness of cold-rolled strip steel based on data driving
CN112613603B (en) Neural network training method based on amplitude limiter and application thereof
CN115689001A (en) Short-term load prediction method based on pattern matching
CN112767380B (en) Prediction method for end shape of wide and thick plate
Wang et al. A deep learning model for the accurate prediction of the microstructure performance of hot rolled steel
CN113110061A (en) Intelligent irrigation fuzzy control method and system based on improved particle swarm optimization

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