CN112989687A - Forecasting method for continuous casting billet corner transverse crack defect based on metallurgical principle - Google Patents
Forecasting method for continuous casting billet corner transverse crack defect based on metallurgical principle Download PDFInfo
- Publication number
- CN112989687A CN112989687A CN201911307077.2A CN201911307077A CN112989687A CN 112989687 A CN112989687 A CN 112989687A CN 201911307077 A CN201911307077 A CN 201911307077A CN 112989687 A CN112989687 A CN 112989687A
- Authority
- CN
- China
- Prior art keywords
- continuous casting
- casting billet
- model
- crystallizer
- layer
- 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
Links
- 238000009749 continuous casting Methods 0.000 title claims abstract description 59
- 230000007547 defect Effects 0.000 title claims abstract description 38
- 238000013277 forecasting method Methods 0.000 title claims abstract description 8
- 238000000034 method Methods 0.000 claims abstract description 30
- 238000003062 neural network model Methods 0.000 claims abstract description 27
- 238000004519 manufacturing process Methods 0.000 claims abstract description 17
- 238000005272 metallurgy Methods 0.000 claims abstract description 5
- 229910000831 Steel Inorganic materials 0.000 claims description 25
- 239000010959 steel Substances 0.000 claims description 25
- 238000001816 cooling Methods 0.000 claims description 23
- 230000004907 flux Effects 0.000 claims description 20
- 230000008569 process Effects 0.000 claims description 19
- 230000008859 change Effects 0.000 claims description 18
- 238000012937 correction Methods 0.000 claims description 18
- 238000005266 casting Methods 0.000 claims description 12
- 238000012549 training Methods 0.000 claims description 10
- 230000006870 function Effects 0.000 claims description 9
- 239000007921 spray Substances 0.000 claims description 9
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 claims description 9
- 230000035945 sensitivity Effects 0.000 claims description 8
- 238000004364 calculation method Methods 0.000 claims description 6
- 239000002893 slag Substances 0.000 claims description 6
- 230000002950 deficient Effects 0.000 claims description 4
- 230000001133 acceleration Effects 0.000 claims description 3
- 230000003044 adaptive effect Effects 0.000 claims description 3
- 238000004422 calculation algorithm Methods 0.000 claims description 3
- 230000005484 gravity Effects 0.000 claims description 3
- 210000002364 input neuron Anatomy 0.000 claims description 3
- 239000007788 liquid Substances 0.000 claims description 3
- 210000004205 output neuron Anatomy 0.000 claims description 3
- 238000012795 verification Methods 0.000 claims 3
- 238000013528 artificial neural network Methods 0.000 claims 1
- 229910000742 Microalloyed steel Inorganic materials 0.000 description 2
- 229910052799 carbon Inorganic materials 0.000 description 2
- 229910052758 niobium Inorganic materials 0.000 description 2
- 239000010955 niobium Substances 0.000 description 2
- GUCVJGMIXFAOAE-UHFFFAOYSA-N niobium atom Chemical compound [Nb] GUCVJGMIXFAOAE-UHFFFAOYSA-N 0.000 description 2
- 239000000843 powder Substances 0.000 description 2
- 229910000851 Alloy steel Inorganic materials 0.000 description 1
- OKTJSMMVPCPJKN-UHFFFAOYSA-N Carbon Chemical compound [C] OKTJSMMVPCPJKN-UHFFFAOYSA-N 0.000 description 1
- 230000004075 alteration Effects 0.000 description 1
- CFJRGWXELQQLSA-UHFFFAOYSA-N azanylidyneniobium Chemical compound [Nb]#N CFJRGWXELQQLSA-UHFFFAOYSA-N 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000005336 cracking Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000001556 precipitation Methods 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
- 230000001681 protective effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000005096 rolling process Methods 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
- 239000002436 steel type Substances 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Images
Classifications
-
- 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
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- 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)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Continuous Casting (AREA)
Abstract
The invention belongs to the technical field of metallurgical continuous casting, and particularly relates to a method for forecasting a continuous casting billet corner transverse crack defect based on a metallurgical principle. Aiming at the defect of insufficient online prediction of the transverse crack defect of the continuous casting billet by the prior art, the forecasting method of the transverse crack defect of the continuous casting billet based on the metallurgical principle is provided, and comprises the following steps: firstly, establishing a principle model for key factors of the continuous casting billet for generating the transverse angle crack defect through analyzing the metallurgical principle of the continuous casting billet; establishing a BP neural network model by taking the principle model as a node of an input layer; learning the established BP neural network model by using the existing continuous casting billet metallurgy sample data until a verified BP neural network model is obtained; and (3) bringing the sample parameters of the actual production continuous casting billet into a verified BP neural network model to obtain the result of predicting the transverse crack defect of the continuous casting billet, and providing reference for production and manufacturing.
Description
Technical Field
The invention belongs to the technical field of metallurgical continuous casting, and particularly relates to a method for forecasting a continuous casting billet corner transverse crack defect based on a metallurgical principle.
Background
With the rapid development of steel enterprises, the hot charging and hot delivery of continuous casting billets and the direct rolling technology make continuous casting steel become an important research field of the steel enterprises, and the continuous casting hot charging and hot delivery not only can greatly reduce the production cost and capital equipment investment of the steel enterprises, but also can obviously improve the product quality and the enterprise competitiveness. However, in the continuous casting process, the brittleness of the casting blank is improved due to precipitation of micro carbon and niobium nitride in the steel, so that the high-temperature mechanical property of the niobium-containing steel is greatly changed, the third brittleness interval of the steel is widened and deepened, the crack sensitivity of the steel is enhanced, and the incidence rate of the transverse corner cracks of the continuous casting slab is far higher than that of other steel types.
Chinese patent document CN108145113A discloses a secondary cooling width cut control method for reducing transverse cracks of niobium-containing microalloyed steel continuous casting slab corners; chinese patent document CN103846401B discloses a secondary cooling process for improving the surface quality of an extra-thick slab. The patent provides a solution to the transverse crack of the casting blank corner from the perspective of the secondary cooling process. Chinese patent document CN104259412A discloses a casting powder for continuous casting of medium-carbon low-alloy steel in an extra-thick plate blank and a preparation method thereof, and explains a method for solving corner transverse cracking from the aspect of crystallizer casting powder. Chinese patent CN 107287487A discloses a method for avoiding the transverse crack of a casting blank corner by adjusting the components of J55-grade micro-alloy steel.
In summary, the prior art attempts to analyze and solve the transverse angle crack of the continuous casting billet on the basis of the principle of metallurgy, but in the production process of the continuous casting billet, the transverse angle crack defect needs to be predicted on line, and the method is expected to play a role in guiding the production.
Disclosure of Invention
The invention aims to solve the technical problem that the prior art method has insufficient online prediction on the transverse crack defect of the continuous casting billet angle and is difficult to guide the production process.
The technical scheme provided by the invention for solving the technical problems is that the invention provides a forecasting method of the continuous casting billet corner transverse crack defect based on the metallurgical principle, which comprises the following steps:
step 1: establishing a principle model for key factors of the continuous casting billet for generating the transverse angle crack defect through analyzing the metallurgical principle of the continuous casting billet;
step 2: establishing a BP neural network model by taking the principle model as a node of an input layer;
and step 3: performing a learning process on the BP neural network model established in the step 2 by using the existing continuous casting billet metallurgy sample data until a verified BP neural network model is obtained;
and 4, step 4: and (3) substituting the sample parameters of the actual production continuous casting billet into the BP neural network model verified in the step (3), and obtaining a result of predicting the transverse crack defect of the continuous casting billet through a judgment process.
Further, the principle model in the step 1 comprises a crack sensitivity index model, a molten steel superheat degree and throwing speed change model, a crystallizer vibration model, a crystallizer cooling uniformity model, a protective slag control model and a secondary cooling control model.
Crack sensitivity index model: x'1=b′1+b1x1+b2x2+...+bnxnWherein x'1Means the probability of crack generation of the continuous casting billet, x1,x2...xnIs the probability of the influence of various elements on the cracks of the continuous casting billet, b'1,b1,b2...bnRefers to the impact weight of various elements.
Molten steel superheat degree and throwing speed change model: x'2=c′1+c1N+c2a, wherein x'2Is the index of the superheat degree of molten steel and the change of the drawing speed, N is the off-line data of the superheat degree of the molten steel, a is the change of the drawing speedΔ v is the amount of change in speed, Δ t is the time required for the change in speed, and the change in throwing speed has an upper limit value A, c'1,c1,c2Refers to the impact weight.
A crystallizer vibration model: x'3=d′1+d1tf+d2Af+d3f, wherein x'3Is the vibration index, t, of the crystallizerfRefers to the negative slip time of the crystallizer under vibration, AfMeans the amount of vibration runout, f means the internal friction force f of the mold is 2 mu rho ghb,mu is a transverse friction coefficient of a billet shell (mu is 0.4-0.6), b is a width of a casting billet, rho is a density of molten steel, g is a gravity acceleration, h is a depth from a liquid level of a crystallizer, and d'1,d1,d2,d3Refers to the impact weight.
Crystallizer cooling uniformity model x'4=e′1+e1×{A1,A2}+e2×Max{B1,B2,B11,B22}+e3×Max{C1,C2X 'therein'4Is an index of cooling uniformity of the crystallizer, A1,A2Respectively the left and right conicity of the crystallizer, B1,B2Respectively refers to the heat flux density of the left and right wide surfaces of the crystallizer, B11,B22Respectively means the left and right heat flux density of the narrow surface of the crystallizer, C1Refers to the heat flux density inside and outside the wide surface of the crystallizer, C2Refers to the internal and external heat flow density of the narrow surface of the crystallizer'1,e1,e2,e3Refers to the impact weight.
A mold flux control model: x'5=f′1+f1Nx+f2R+f3f, wherein x'5Is the adaptive index of the covering slag, NxRefers to the consumption of the mold flux, R refers to the alkalinity of the mold flux, f refers to the viscosity of the mold flux, f'1,f1,f2,f3Refers to the impact weight.
And a second cooling control model: x'6=g′1+g1W+g2Q+g3S, wherein x'6Is a secondary cooling zone influence index, W is water spray intensity, Q is water spray amount, S is water spray area, g'1,g1,g2,g3Refers to the impact weight.
Further, the BP neural network model in the step 2 comprises three layers of structures of an input layer, a hidden layer and an output layer, wherein input neurons of the input layer represent a principle model, the input layer and the hidden layer are related through connection weights of intermediate layers of the input layer and the hidden layer, the hidden layer and the output layer are related through connection weights of the intermediate layers of the hidden layer and the output layer, and 1 output neuron of the output layer represents a result of forecasting the transverse fissure defect of the continuous casting billet.
Further, the learning process in step 3 includes the following steps:
1) network initialization: initializing the number of network input nodes, the number of hidden layer nodes and the number of output nodes; ② initializing weight and threshold (weight ω)ij、ωjAnd a threshold value ajB are respectively assigned with random numbers in a range (-1, 1); selecting K samples as training samples, the input value of each node of each sample is xi(k) I ═ 1, 2 …, n, K ═ 1, 2 …, K; the expected output value is y (k), and the actual output value is O (k); setting local error function of single sample Setting the global error function of all samples as Initializing a calculation precision value epsilon and a maximum iteration time T, and initializing the current iteration time T as 1; the learning rate η is initialized.
2) And in the t iteration, circularly reading each training sample, wherein the input value of the kth training sample is xi(k),i=(1,2…,n),k=(1,2…,K)。
3) Calculating layer by layer, and finally calculating to obtain the actual output value O (k) of each output node of the sample, wherein the calculation process is as follows: calculating the output value H of each node of the hidden layerj:
Calculating an actual output value o (k) of the output node:
4) calculating the error function E (k) of a single sample
5) Correcting network weight and threshold
In the BP algorithm, the correction quantity of the weight is in direct proportion to the partial differential of the error E (k) to the weight, and the correction quantity of the threshold is in direct proportion to the partial differential of the error E (k) to the threshold; thereby calculating the weight correction amount and the threshold correction amount.
ωjcorrected value: omegaj=ωj+Δωj=ωj+ηHj(y(k)-O(k))
b corrected value: b + Δ b + η (y (k) -o (k))
If so: δ ═ y (k) -o (k)
Then:
ωj=ωj+ηHjδ
b=b+ηδ
in the above equation: j ═ 1, 2, …, L;
δj=Hj(1-Hj)δωj
ωij=ωij+ηδjxi(k)
aj=aj+ηδj
in the above equation: 1, 2, …, n; j ═ 1, 2, …, L;
6) if K is less than K, making K equal to K +1, and repeating the steps 3, 4 and 5; otherwise, continuing.
7) After all samples have been trained, the global error E is calculated
Stopping iteration when E is less than epsilon or T is T; otherwise, let k be 1, t be t +1, repeat steps 2), 3), 4), 5), 6), 7).
Further, the determination process in step 4 utilizes the weight ω obtained from the learning process in step 3)ij、ωjAnd a threshold value aiB, predicting an actual production continuous casting billet sample, and specifically comprising the following steps:
1) the input parameter expression of the sample of the actual production continuous casting billet is as follows: x'i(i=1,2,…,n)。
2) Calculating an output value of each node of the BP neural network model hidden layer:
3) calculating an expected output value O of a node of the BP neural network model output layer:
4) when O is less than 0.5, judging that no defect exists; when O > 0.5, it is judged to be defective.
The forecasting method has the beneficial effects that the forecasting method of the transverse crack defect of the continuous casting billet corner based on the metallurgical principle is provided, the defect generation reason is essentially forecasted based on the metallurgical principle by combining a statistical method and a neural network model, the key factor or key control link of the defect generation is positioned, and reference is provided for production and manufacturing or feedforward control.
Drawings
Fig. 1 is a schematic diagram of a three-layer BP network structure.
The reference number, X is the input variable, n is the number of input layer nodes, namely the number of main influence parameters of the casting blank quality defect, Y is the output variable, the number of output nodes is only 1, namely whether the casting blank quality defect exists, H is the output value of the hidden layer, the number of the hidden layer nodes is L, H is the output value of the hidden layerjThe output value of the jth output node representing the hidden layer, j ═ 1, 2, …, L), ωijIs the connection weight between the ith node of the input layer and the jth node of the hidden layer, i ═ 1, 2, …, n, j ═ 1, 2, …, L), ωjThe connection weight between the jth node of the hidden layer and the node of the output layer, j ═ 1, 2, …, L), ajThe threshold of the jth node of the hidden layer is represented by j ═ 1, 2, …, L, and b.
Detailed Description
The detailed description of the specific implementation mode of the invention is combined with the attached drawings, and the forecasting method of the transverse crack defect of the continuous casting billet based on the metallurgical principle comprises the following steps:
step 1: by analyzing the metallurgical principle of the continuous casting billet, a principle model is established for key factors of the continuous casting billet for generating the transverse angle crack defect, and the principle model comprises a crack sensitivity index model, a molten steel superheat degree and throwing speed change model, a crystallizer vibration model, a crystallizer cooling uniformity model, a covering slag control model and a secondary cooling control model.
Crack sensitivity index model: x'1=b′1+b1x1+b2x2+…+bnxnWherein x'1Means the probability of crack generation of the continuous casting billet, x1,x2…xnIs the probability of the influence of various elements on the cracks of the continuous casting billet, b'1,b1,b2…bnRefers to the impact weight of various elements.
Molten steel superheat degree and throwing speed change model: x'2=c′1+c1N+c2a, wherein x'2Is the index of the change of the superheat degree of the molten steel and the drawing speed, N is off-line data of the superheat degree of the molten steel, a is off-line data of the superheat degree of the molten steel, andvariation of withdrawal speedΔ v is the amount of change in speed, Δ t is the time required for the change in speed, and the change in throwing speed has an upper limit value A, c'1,c1,c2Refers to the impact weight.
A crystallizer vibration model: x'3=d′1+d1tf+d2Af+d3f, wherein x'3Is the vibration index, t, of the crystallizerfRefers to the negative slip time of the crystallizer under vibration, AfThe vibration deflection is referred to as f, the internal friction force f of the crystallizer is 2 mu rho ghb, mu is a transverse friction coefficient of a billet shell (mu is 0.4-0.6), b is a casting blank width, rho is molten steel density, g is gravity acceleration, h is a depth from the liquid surface of the crystallizer, and d'1,d1,d2,d3Refers to the impact weight.
Crystallizer cooling uniformity model x'4=e′1+e1×{A1,A2}+e2×Max{B1,B2,B11,B22}+e3×Max{C1,C2In which x4Is an index of cooling uniformity of the crystallizer, A1,A2Respectively the left and right conicity of the crystallizer, B1,B2Respectively refers to the heat flux density of the left and right wide surfaces of the crystallizer, B11,B22Respectively means the left and right heat flux density of the narrow surface of the crystallizer, C1Refers to the heat flux density inside and outside the wide surface of the crystallizer, C2Refers to the internal and external heat flow density of the narrow surface of the crystallizer'1,e1,e2,e3Refers to the impact weight.
A mold flux control model: x'5=f1′+f1Nx+f2R+f3f, wherein x'5Is the adaptive index of the covering slag, NxIs the consumption of the mold flux, R is the alkalinity of the mold flux, f is the viscosity of the mold flux1′,f1,f2,f3Mean the weight of influence。
And a second cooling control model: x'6=g′1+g1W+g2Q+g3S, wherein x'6Is a secondary cooling zone influence index, W is water spray intensity, Q is water spray amount, S is water spray area, g'1,g1,g2,g3Refers to the impact weight.
Step 2: establishing a three-layer BP neural network model shown in figure 1 by taking a principle model as a node of an input layer, wherein X is an input variable, n is the number of nodes of the input layer, namely the number of main influence parameters of casting blank quality defects, Y is an output variable, the number of output nodes is only 1, namely whether the casting blank quality defects exist or not, H is an output value of a hidden layer, the number of nodes of the hidden layer is L, and H is an output value of the hidden layerjThe output value of the jth output node representing the hidden layer, j ═ 1, 2, …, L), ωijIs the connection weight between the ith node of the input layer and the jth node of the hidden layer, i ═ 1, 2, …, n, j ═ 1, 2, …, L), ωjThe connection weight between the jth node of the hidden layer and the node of the output layer, j ═ 1, 2, …, L), ajThe threshold of the jth node of the hidden layer is represented by j ═ 1, 2, …, L, and b. The BP neural network model comprises a three-layer structure comprising a multi-input layer, a single hidden layer and a single output layer, wherein input neurons of the input layer represent a principle model, the input layer and the hidden layer are associated through connection weights of intermediate layers of the input layer and the hidden layer, the hidden layer and the output layer are associated through connection weights of the intermediate layers of the hidden layer and the output layer, and 1 output neuron of the output layer represents a result of forecasting the transverse fissure defect of the continuous casting billet.
And step 3: and (3) carrying out a learning process on the BP neural network model established in the step (2) by using the existing continuous casting billet metallurgy sample data until a verified BP neural network model is obtained, wherein the learning process comprises the following steps:
1) network initialization: initializing the number of network input nodes, the number of hidden layer nodes and the number of output nodes; ② initializing weight and threshold (weight ω)ij、ωjAnd a threshold value ajB are respectively assigned with random numbers in a range (-1, 1); selecting K samples for trainingTraining samples, each sample having an input value of x for each nodei(k) I ═ 1, 2 …, n, K ═ 1, 2 …, K; the expected output value is y (k), and the actual output value is O (k); setting local error function of single sample Setting the global error function of all samples as Initializing a calculation precision value epsilon and a maximum iteration time T, and initializing the current iteration time T as 1; the learning rate η is initialized.
2) And in the t iteration, circularly reading each training sample, wherein the input value of the kth training sample is xi(k),i=(1,2…,n),k=(1,2…,K)。
3) Calculating layer by layer, and finally calculating to obtain the actual output value O (k) of each output node of the sample, wherein the calculation process is as follows: calculating the output value H of each node of the hidden layerj:
Calculating an actual output value o (k) of the output node:
4) calculating the error function E (k) of a single sample
5) Correcting network weight and threshold
In the BP algorithm, the correction quantity of the weight is in direct proportion to the partial differential of the error E (k) to the weight, and the correction quantity of the threshold is in direct proportion to the partial differential of the error E (k) to the threshold; thereby calculating the weight correction amount and the threshold correction amount.
ωjcorrected value: omegaj=ωj+Δωj=ωj+ηHj(y(k)-O(k))
b corrected value: b + Δ b + η (y (k) -o (k))
If so: δ ═ y (k) -o (k)
Then:
ωj=ωj+ηHjδ
b=b+ηδ
in the above equation: j ═ 1, 2, …, L;
δj=Hj(1-Hj)δωj
ωij=ωij+ηδjxi(k)
aj=aj+ηδj
in the above equation: 1, 2, …, n; j ═ 1, 2, …, L;
6) if K is less than K, making K equal to K +1, and repeating the steps 3, 4 and 5; otherwise, continuing.
7) After all samples have been trained, the global error E is calculated
Stopping iteration when E is less than epsilon or T is T; otherwise, let k be 1, t be t +1, repeat steps 2), 3), 4), 5), 6), 7).
And 4, step 4: and (3) substituting the sample parameters of the actual production continuous casting billet into the BP neural network model verified in the step (3), and obtaining a result of predicting the transverse crack defect of the continuous casting billet through a judging process, wherein the judging process comprises the following steps:
1) actual production continuous casting billet sample input parameters: x'i(i=1,2,…,n)。
2) Calculating the output value of each node of the hidden layer of the BP neural network model:
3) calculating an expected output value O of a node of the BP neural network model output layer:
4) when O is less than 0.5, judging that no defect exists; when O > 0.5, it is judged to be defective.
The present invention is not limited to the specific technical solutions described in the above embodiments, and other embodiments may be made in the present invention in addition to the above embodiments. It will be understood by those skilled in the art that various changes, substitutions of equivalents, and alterations can be made without departing from the spirit and scope of the invention.
Claims (5)
1. A forecasting method of continuous casting billet corner transverse crack defects based on a metallurgical principle is characterized by comprising the following steps:
step 1: establishing a key factor principle model for key factors of the continuous casting billet generating the angle transverse crack defect, wherein the key factors comprise crack sensitivity index, molten steel superheat degree, throwing speed change, crystallizer vibration, crystallizer cooling uniformity, covering slag control and secondary cooling control;
step 2: establishing a BP neural network model by taking the principle model as a node of an input layer;
and step 3: performing a learning process on the BP neural network model established in the step 2 by using the existing continuous casting billet metallurgy sample data until a verification BP neural network model is obtained;
and 4, step 4: substituting the actual production continuous casting billet sample parameters into the verification BP neural network model in the step 3, and obtaining a result of predicting the angle transverse crack defect of the continuous casting billet through judgment; the determination is:
when the expected output value O of the node of the verification BP neural network model output layer is less than 0.5, judging that no defect exists; when O > 0.5, it is judged to be defective.
2. The method for forecasting the transverse crack defect of the continuous casting billet based on the metallurgical principle as claimed in claim 1, wherein the key factor principle model comprises a crack sensitivity index model, a molten steel superheat degree and throwing speed change model, a crystallizer vibration model, a crystallizer cooling uniformity model, a mold flux control model and a secondary cooling control model;
the crack sensitivity index model is expressed by the following formula: x'1=b′1+b1x1+b2x2+...+bnxnWherein x'1Means the probability of crack generation of the continuous casting billet, x1,x2...xnIs the probability of the influence of various elements on the cracks of the continuous casting billet, b'1,b1,b2...bnRefers to the impact weight of various elements;
the molten steel superheat degree and throwing speed change model is expressed by the following formula: x'2=c′1+c1N+c2a, wherein x'2Is the index of the superheat degree of molten steel and the change of the drawing speed, N is the off-line data of the superheat degree of the molten steel, a is the change of the drawing speedDeltav is the speed variation, Deltat is the time required by the speed variation, the throwing speed variation has an upper limit value A,c′1,c1,c2refers to the impact weight;
the crystallizer vibration model is expressed by the following formula: x'3=d′1+d1tf+d2Af+d3f, wherein x'3Is the vibration index, t, of the crystallizerfRefers to the negative slip time of the crystallizer under vibration, AfThe vibration deflection is referred to as f, the internal friction force f of the crystallizer is 2 mu rho ghb, mu is a transverse friction coefficient of a billet shell (mu is 0.4-0.6), b is a casting blank width, rho is molten steel density, g is gravity acceleration, h is a depth from the liquid surface of the crystallizer, and d'1,d1,d2,d3Refers to the impact weight;
the crystallizer cooling uniformity model is expressed by the following formula:
x'4=e′1+e1×{A1,A2}+e2×Max{B1,B2,B11,B22}+e3×Max{C1,C2x 'therein'4Is an index of cooling uniformity of the crystallizer, A1,A2Respectively the left and right conicity of the crystallizer, B1,B2Respectively refers to the heat flux density of the left and right wide surfaces of the crystallizer, B11,B22Respectively means the left and right heat flux density of the narrow surface of the crystallizer, C1Refers to the heat flux density inside and outside the wide surface of the crystallizer, C2Refers to the internal and external heat flow density of the narrow surface of the crystallizer'1,e1,e2,e3Refers to the impact weight;
the mold flux control model is expressed by the following formula: x'5=f1′+f1Nx+f2R+f3f, wherein x'5Is the adaptive index of the covering slag, NxIs the consumption of the mold flux, R is the alkalinity of the mold flux, f is the viscosity of the mold flux1′,f1,f2,f3Refers to the impact weight.
The secondary cooling control model is expressed by the following formula: x'6=g′1+g1W+g2Q+g3S, wherein x'6Is a secondary cooling zone influence index, W is water spray intensity, Q is water spray amount, S is water spray area, g'1,g1,g2,g3Refers to the impact weight.
3. The method for forecasting the transverse crack defect of the continuous casting billet based on the metallurgical principle as claimed in claim 1, wherein the BP neural network model comprises a three-layer structure of a multi-input layer, a single hidden layer and a single output layer, the input neurons of the input layer represent a principle model, the input layer and the hidden layer are associated through the connection weights of the two intermediate layers, the hidden layer and the output layer are associated through the connection weights of the two intermediate layers, and 1 output neuron of the output layer represents the result of forecasting the transverse crack defect of the continuous casting billet.
4. The method for forecasting the transverse crack defect of the continuous casting billet based on the metallurgical principle as claimed in claim 1, wherein the learning process comprises the following steps:
1) network initialization: initializing the number of input nodes, the number of hidden layer nodes and the number of output nodes of the BP neural network; ② initializing weight and threshold (weight ω)ij、ωjAnd a threshold value ajB are respectively assigned with random numbers in a range (-1, 1); selecting K samples as training samples, the input value of each node of each sample is xi(k) I ═ 1, 2 …, n, K ═ 1, 2.., K); the expected output value is y (k), and the actual output value is O (k); setting local error function of single sample Setting the global error function of all samples as Initializing a calculation precision value epsilon and a maximum iteration time T, and initializing the current iteration time T as 1; initializing a learning rate eta;
2) and in the t iteration, circularly reading each training sample, wherein the input value of the kth training sample is xi(k),i=(1,2…,n),k=(1,2…,K);
3) Calculating layer by layer, and finally calculating to obtain an actual output value O (k) of each output node of the training sample, wherein the calculation process is as follows: calculating the output value H of each node of the hidden layerj:
Calculating an actual output value o (k) of the output node:
4) calculating error function E (k) of single training sample
5) Correcting network weight and threshold
In the BP algorithm, the correction quantity of the weight is in direct proportion to the partial differential of the error E (k) to the weight, and the correction quantity of the threshold is in direct proportion to the partial differential of the error E (k) to the threshold; thereby calculating a weight correction and a threshold correction:
ωjcorrected value: omegaj=ωj+Δωj=ωj+ηHj(y(k)-O(k))
b corrected value: b + Δ b + η (y (k) -o (k))
If so: δ ═ y (k) -o (k)
Then:
ωj=ωj+ηHjδ
b=b+ηδ
in the above equation: j ═ 1, 2, …, L;
δj=Hj(1-Hj)δωj
ωij=ωij+ηδjxi(k)
aj=aj+ηδj
in the above equation: 1, 2, …, n; j ═ 1, 2, …, L;
6) if K is less than K, making K equal to K +1, and repeating the steps 3, 4 and 5; otherwise, continuing;
7) after all samples have been trained, the global error E is calculated
Stopping iteration when E is less than epsilon or T is T; otherwise, let k be 1, t be t +1, repeat steps 2), 3), 4), 5), 6), 7).
5. The method for forecasting the transverse crack defect of the continuous casting slab based on the metallurgical principle as claimed in claim 1, wherein the judging process utilizes the weight ω obtained in the learning process in the step 3ij、ωjAnd a threshold value ajB, predicting an actual production continuous casting billet sample, and specifically comprising the following steps:
1) actual production continuous casting billet sample input parameters: x'i(i=1,2,…,n);
2) Calculating an output value of each node of the BP neural network model hidden layer:
3) calculating an expected output value O of a node of the BP neural network model output layer:
4) when O is less than 0.5, judging that no defect exists; when O > 0.5, it is judged to be defective.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201911307077.2A CN112989687B (en) | 2019-12-18 | 2019-12-18 | Method for forecasting transverse crack defect of continuous casting billet angle based on metallurgical principle |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201911307077.2A CN112989687B (en) | 2019-12-18 | 2019-12-18 | Method for forecasting transverse crack defect of continuous casting billet angle based on metallurgical principle |
Publications (2)
Publication Number | Publication Date |
---|---|
CN112989687A true CN112989687A (en) | 2021-06-18 |
CN112989687B CN112989687B (en) | 2023-11-07 |
Family
ID=76343773
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201911307077.2A Active CN112989687B (en) | 2019-12-18 | 2019-12-18 | Method for forecasting transverse crack defect of continuous casting billet angle based on metallurgical principle |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN112989687B (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114603090A (en) * | 2022-03-11 | 2022-06-10 | 北京海卓博尔科技有限公司 | Crystallizer vibration driving device, control method and control system |
Citations (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101299032A (en) * | 2008-06-06 | 2008-11-05 | 重庆交通大学 | Structural crack bionic monitoring system and monitoring method thereof |
CN101995436A (en) * | 2010-10-26 | 2011-03-30 | 江苏大学 | Neural network based drawing part crack monitoring method |
CN102319883A (en) * | 2011-10-09 | 2012-01-18 | 北京首钢自动化信息技术有限公司 | Method for controlling on-line prediction of continuous casting blank quality |
CN102653835A (en) * | 2012-05-09 | 2012-09-05 | 东北大学 | Method for reducing transverse cracks at corner of continuous casting billet of boron-steel-containing wide-thick plate |
CN102937784A (en) * | 2012-10-30 | 2013-02-20 | 中冶南方工程技术有限公司 | Artificial neural network based method for controlling online prediction of casting billet quality |
CN103920859A (en) * | 2013-01-14 | 2014-07-16 | 中冶南方工程技术有限公司 | Continuous casting sheet billet internal crack online prediction method |
CN104889358A (en) * | 2014-03-05 | 2015-09-09 | 鞍钢股份有限公司 | Method for controlling surface crack of continuous cast slab |
CN105328155A (en) * | 2015-10-08 | 2016-02-17 | 东北电力大学 | Steel leakage visualized characteristic forecasting method based on improved neural network |
CN105911095A (en) * | 2016-05-04 | 2016-08-31 | 东北电力大学 | Visual recognition method of continuous casting billet surface longitudinal cracks |
CN106077555A (en) * | 2016-08-12 | 2016-11-09 | 湖南千盟物联信息技术有限公司 | A kind of continuous casting coordinating and optimizing control method |
CN107609647A (en) * | 2017-10-16 | 2018-01-19 | 安徽工业大学 | One kind is based on BP neural network roll alloy mechanical property Forecasting Methodology |
CN110202106A (en) * | 2019-06-04 | 2019-09-06 | 甘肃酒钢集团宏兴钢铁股份有限公司 | The method for controlling CSP thin sheet continuous casting machine production medium carbon alloy steel slab surface cracks |
-
2019
- 2019-12-18 CN CN201911307077.2A patent/CN112989687B/en active Active
Patent Citations (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101299032A (en) * | 2008-06-06 | 2008-11-05 | 重庆交通大学 | Structural crack bionic monitoring system and monitoring method thereof |
CN101995436A (en) * | 2010-10-26 | 2011-03-30 | 江苏大学 | Neural network based drawing part crack monitoring method |
CN102319883A (en) * | 2011-10-09 | 2012-01-18 | 北京首钢自动化信息技术有限公司 | Method for controlling on-line prediction of continuous casting blank quality |
CN102653835A (en) * | 2012-05-09 | 2012-09-05 | 东北大学 | Method for reducing transverse cracks at corner of continuous casting billet of boron-steel-containing wide-thick plate |
CN102937784A (en) * | 2012-10-30 | 2013-02-20 | 中冶南方工程技术有限公司 | Artificial neural network based method for controlling online prediction of casting billet quality |
CN103920859A (en) * | 2013-01-14 | 2014-07-16 | 中冶南方工程技术有限公司 | Continuous casting sheet billet internal crack online prediction method |
CN104889358A (en) * | 2014-03-05 | 2015-09-09 | 鞍钢股份有限公司 | Method for controlling surface crack of continuous cast slab |
CN105328155A (en) * | 2015-10-08 | 2016-02-17 | 东北电力大学 | Steel leakage visualized characteristic forecasting method based on improved neural network |
CN105911095A (en) * | 2016-05-04 | 2016-08-31 | 东北电力大学 | Visual recognition method of continuous casting billet surface longitudinal cracks |
CN106077555A (en) * | 2016-08-12 | 2016-11-09 | 湖南千盟物联信息技术有限公司 | A kind of continuous casting coordinating and optimizing control method |
CN107609647A (en) * | 2017-10-16 | 2018-01-19 | 安徽工业大学 | One kind is based on BP neural network roll alloy mechanical property Forecasting Methodology |
CN110202106A (en) * | 2019-06-04 | 2019-09-06 | 甘肃酒钢集团宏兴钢铁股份有限公司 | The method for controlling CSP thin sheet continuous casting machine production medium carbon alloy steel slab surface cracks |
Non-Patent Citations (2)
Title |
---|
江中块等: "梅钢连铸机板坯角横裂缺陷解决实践", 《 2014年全国炼钢—连铸生产技术会论文集》, pages 548 - 552 * |
郭贤利等: "基于神经网络的连铸板坯质量在线诊断系统", 《冶金自动化》, vol. 37, no. 3, pages 16 - 22 * |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114603090A (en) * | 2022-03-11 | 2022-06-10 | 北京海卓博尔科技有限公司 | Crystallizer vibration driving device, control method and control system |
CN114603090B (en) * | 2022-03-11 | 2023-06-16 | 北京海卓博尔科技有限公司 | Crystallizer vibration driving device, control method and control system |
Also Published As
Publication number | Publication date |
---|---|
CN112989687B (en) | 2023-11-07 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108637020B (en) | Self-adaptive variation PSO-BP neural network strip steel convexity prediction method | |
CN114329940A (en) | Continuous casting billet quality prediction method based on extreme learning machine | |
Zheng et al. | Optimization of high-pressure die-casting process parameters using artificial neural network | |
CN102937784A (en) | Artificial neural network based method for controlling online prediction of casting billet quality | |
WO2023130666A1 (en) | Strip steel plate convexity prediction method based on data-driving and mechanism model fusion | |
CN105363794B (en) | A kind of finish rolling energy-saving control method based on mechanical properties forecast and required power model | |
CN101306438A (en) | Accelerated stream cooling device cooling technique integration method | |
CN113343576B (en) | Prediction method of calcium yield in calcium treatment process based on deep neural network | |
KR102356461B1 (en) | Production method of crack-stopping steel for container ships | |
CN105911095A (en) | Visual recognition method of continuous casting billet surface longitudinal cracks | |
Jiang et al. | BP neural network of continuous casting technological parameters and secondary dendrite arm spacing of spring steel | |
JP7176507B2 (en) | Hot rolling warpage prediction method, warpage control method, hot rolled steel sheet manufacturing method, warpage prediction model generation method, and hot rolling equipment | |
CN104451118B (en) | A kind of belt plate shape being suitable for continuous annealing process develops forecasting procedure piecemeal | |
CN112989687A (en) | Forecasting method for continuous casting billet corner transverse crack defect based on metallurgical principle | |
CN113849020B (en) | Billet heating curve design method and device based on artificial intelligence algorithm | |
CN115608793B (en) | Finish rolling temperature regulation and control method for mechanism fusion data | |
CN107716560B (en) | A kind of Hot Strip Rolling load distribution method | |
CN115238581A (en) | Method for predicting HVOF coating performance optimization spraying parameters based on HML | |
CN109240203B (en) | Multi-model-based continuous casting billet constant weight control method | |
CN109047698B (en) | Continuous casting billet fixed weight and fixed length online prediction method | |
JP3561401B2 (en) | State estimation method for manufacturing process | |
CN116738518A (en) | Numerical simulation verification method for crack position under continuous casting light reduction and internal quality control method | |
CN101947645B (en) | Method for indentifying steel ladle slag entrapment process | |
Wang | Study on influence factors of zinc layer thickness via response surface method, Taguchi method and genetic algorithm | |
KR102075245B1 (en) | Prediction apparatus for iron loss reduction of electric steel sheet |
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 |