CN115270397A - Neural network vermicular graphite cast iron vermicular rate prediction method based on Tensorflow - Google Patents

Neural network vermicular graphite cast iron vermicular rate prediction method based on Tensorflow Download PDF

Info

Publication number
CN115270397A
CN115270397A CN202210680749.XA CN202210680749A CN115270397A CN 115270397 A CN115270397 A CN 115270397A CN 202210680749 A CN202210680749 A CN 202210680749A CN 115270397 A CN115270397 A CN 115270397A
Authority
CN
China
Prior art keywords
neural network
vermicular
cast iron
neurons
rate
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.)
Pending
Application number
CN202210680749.XA
Other languages
Chinese (zh)
Inventor
杨湘杰
杨明浩
叶寒
顾嘉
李全
白宇轩
姜乐付
杨福玲
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.)
Nanchang University
Original Assignee
Nanchang University
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 Nanchang University filed Critical Nanchang University
Priority to CN202210680749.XA priority Critical patent/CN115270397A/en
Publication of CN115270397A publication Critical patent/CN115270397A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • 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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
    • G06F2119/02—Reliability analysis or reliability optimisation; Failure analysis, e.g. worst case scenario performance, failure mode and effects analysis [FMEA]

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Health & Medical Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Data Mining & Analysis (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Computational Linguistics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Computer Hardware Design (AREA)
  • Geometry (AREA)
  • Investigating Or Analyzing Materials Using Thermal Means (AREA)

Abstract

The invention discloses a neural network vermicular cast iron vermicular rate prediction method based on Tensorflow, which is characterized in that through the network structure design of a neural network prediction model of the vermicular cast iron vermicular rate and the acquisition of training data, under the condition that the number of input neurons, output neurons, a nerve activation function and hidden layer neurons of a neural network are determined, preprocessed sample data are adopted to train the preliminarily established neural network prediction model, and further the neural network prediction model for detecting the vermicular cast iron vermicular rate is obtained. Compared with the traditional prediction model, the method has the advantages that the creep rate is rapidly predicted through the neural network construction algorithm model, due to the characteristics of the neural network, the method has self-learning capability and self-adaption capability, when production raw materials or production conditions change, only enough data samples need to be provided again to retrain the original model, the network model can be updated automatically, and the method has great guiding significance on the creep quality in the production process.

Description

Neural network vermicular graphite cast iron vermicular rate prediction method based on Tensorflow
Technical Field
The invention belongs to the technical field of vermicular cast iron vermicular rate prediction, and particularly relates to a neural network vermicular cast iron vermicular rate prediction method based on Tensorflow.
Background
Due to the influence of production conditions and raw material quality, the creep rate and the tissue morphology predicted by the multivariate linear regression model constructed by the traditional thermal analysis technology are difficult to adapt to the current strict requirements of industrial production, and higher accuracy and adaptability are lacked, so that a formula for fitting a regression equation often brings larger errors. There is a need for an adaptive, self-training model.
The development of the artificial neural network opens a new way for predicting the texture and the performance of the vermicular cast iron. It consists of many computational elements operating in parallel, through connections with variable weights, and adjusted in a learning process to achieve the target expectation. The method has the characteristics of strong nonlinear mapping capability, self-adaptive capability, robustness and the like, and can approach any nonlinear system. Therefore, the project provides a vermicular cast iron rate prediction model based on a neural network, and the analysis is carried out by combining the metallographic image processing data, so that the accurate prediction of the vermicular cast iron rate is realized.
Disclosure of Invention
Aiming at the defects of the prior art, the invention provides a neural network vermicular cast iron creep rate prediction method based on Tensorflow, through the network structure design of a neural network prediction model of the vermicular cast iron creep rate and the acquisition of training data, under the condition that the number of input neurons, output neurons, a neural activation function and hidden layer neurons of the neural network is determined, the preliminarily established neural network prediction model is trained by adopting preprocessed sample data, and further the neural network prediction model for detecting the vermicular cast iron creep rate is obtained.
In order to achieve the purpose, the invention adopts the following technical scheme:
a neural network vermicular cast iron vermicular rate prediction method based on Tensorflow comprises the following steps:
s1, recording the change of the temperature of the molten iron of the eutectic vermicular graphite cast iron along with time through a computer integrated temperature data acquisition card, drawing a cooling curve, and selecting a proper characteristic value from the cooling curve to be used for determining an input neuron and an output neuron of a neural network for detecting the vermicular graphite cast iron vermicularity rate;
s2, selecting a proper neural network neuron activation function to ensure smooth network training and ensure the quality of a model;
s3, according to an empirical formula
Figure BDA0003696213570000021
Calculating the number of the neurons of the hidden layer of the neural network;
s4, establishing a preliminary neural network prediction model according to the number of the neural network input neurons, the number of the neural network output neurons, the neural activation function and the number of the neurons of the hidden layer, which are determined in the step;
and S5, preprocessing input and output sample data, and training the preliminary neural network prediction model established in the step S4 by using the processed sample data to obtain the neural network prediction model for detecting the vermicular cast iron vermicularity rate.
In the step S1, a proper characteristic value is selected from the cooling curve and used for determining an input neuron and an output neuron of a neural network for detecting the vermicular cast iron rate, wherein the input neuron comprises a minimum temperature TEU of vermicular cast iron eutectic growth, a eutectic ending temperature TS and a eutectic regeneration temperature TER, and the output neuron is the vermicular cast iron rate.
And S2, selecting a proper neural network neuron activation function, wherein the neuron activation function is a Relu function.
Said empirical formula is used in step S3
Figure BDA0003696213570000022
And calculating the number of the neurons of the hidden layer of the neural network, wherein m is the number of the neurons of the input layer, n is the number of the neurons of the output layer, and c is a constant and has a value range of 4-13.
And S5, preprocessing the input and output sample data in a normalization mode, and converting all the sample data into data between 0 and 1.
Compared with the prior art, the invention has the beneficial effects that:
compared with a traditional linear regression-constructed model, the method has the advantages that the creep rate is rapidly predicted through the neural network construction algorithm model, due to the characteristics of the neural network, the neural network has self-learning capability and self-adaptive capability, when production raw materials or production conditions change, only enough data samples need to be provided again to retrain the original model, the network model can be updated automatically, and the method has great guiding significance on the creep quality in the production process.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to these drawings without creative efforts.
FIG. 1 is a flow chart of a neural network vermicular cast iron vermicular rate prediction method based on Tensorflow in the invention;
FIG. 2 is a schematic diagram of a neural network according to the present invention;
FIG. 3 is a diagram illustrating the variation of neural network error with step size in an embodiment of the present invention;
FIG. 4 is a diagram illustrating a creep rate prediction result according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in 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, and it is obvious that the described embodiments are some, but not all, 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 any inventive step, are within the scope of the present invention.
Example (b): see fig. 1-4.
As shown in FIG. 1, the neural network vermicular cast iron vermicular rate prediction method based on Tensorflow comprises the following steps:
s1, recording the change of the temperature of the molten iron of the eutectic vermicular cast iron along with time through a computer integrated temperature data acquisition card, drawing a cooling curve, and selecting a proper characteristic value from the cooling curve for determining an input neuron and an output neuron of a neural network for detecting the vermicular cast iron vermicularity rate;
s2, in deep learning, the commonly used activation functions mainly comprise: a sigmoid function, a tanh function and a ReLU function, wherein in order to ensure smooth network training and ensure model quality, a proper neuron activation function needs to be determined for a neural network;
s3, the number of neurons in the hidden layer is an important parameter influencing the performance of the neural network, generally speaking, the prediction performance is reduced when the number of neurons is too small, the stability of the model is influenced, otherwise, overfitting is caused to training data when the number of neurons is too large, the time required by training is increased, and the practicability of the model is influenced, so that the neural network is based on an empirical formula
Figure BDA0003696213570000041
Calculating the number of the neurons of the hidden layer of the neural network;
s4, establishing a preliminary neural network prediction model according to the neural network input neuron, the neural network output neuron, the neural activation function and the number of neurons of the hidden layer determined in the step;
and S5, preprocessing input and output sample data, wherein the processed sample data is used for training the preliminary neural network prediction model established in the step S4 to obtain a neural network prediction model for detecting the vermicular cast iron vermicular rate.
In particular, as shown in figure 2, selecting proper characteristic values from the cooling curve for determining input neurons and output neurons of the neural network for detecting the vermicular cast iron in the step S1, the eutectic supercooling temperature TEU is the temperature at the end of the graphite nucleation process, then the graphite nuclei start to grow, and the higher the TEU is, the better the spheroidization effect is; the eutectic regeneration temperature TER is the temperature of the cooling temperature rising again in the eutectic solidification period due to the eutectic latent heat, so that the higher the TER is, the more the solidification latent heat is, the faster the nucleation and growth speed is, the corresponding graphite area can be increased, and the creep rate and the like can be improved; the input layer therefore selects TEU, TER and TS as input neurons for predicting nodularity, which is the vermicular cast iron vermicular rate.
And S2, selecting a proper neural network neuron activation function, wherein the neuron activation function is a Relu function.
Said empirical formula is used in step S3
Figure BDA0003696213570000051
And calculating the number of the neurons of the hidden layer of the neural network, wherein m is the number of the neurons of the input layer, n is the number of the neurons of the output layer, and c is a constant and has a value range of 4-13. In this embodiment, the number of hidden layer neurons is selected to be 11 by operation.
In the step S5, the input and output sample data are preprocessed in a normalization mode, and all the sample data are converted into data between 0 and 1
Referring to FIGS. 3-4, a method for predicting the creep rate of a Tensorflow-based neural network vermicular cast iron according to the present invention is further illustrated by an example.
Description of the experiment: the raw materials mainly comprise a vermiculizer, an inoculant, scrap steel and a carburant. The components are mixed according to the site requirement of the vermicular iron brake drum, and the specific components are shown in the following table 1.
TABLE 1 vermicular cast iron base iron composition and final composition after control
Figure BDA0003696213570000052
The experimental process comprises the following steps: scrap steel is used as basic furnace charge, carburant and ferrosilicon are used for adjusting carbon content and silicon content, a 300Kg intermediate frequency induction furnace is adopted for smelting, and a vermiculizer R2A (the adding amount is 0.05-0.18 percent) and an inoculant YFY-8 (the adding amount is 0.53-0.62 percent) are adopted.
Both the creeping and the inoculation are carried out by a ladle bottom punching method; the tapping temperature of molten iron is about 1510-1550 ℃, and the casting temperature of a sample is 1380-1420 ℃. The thermal analysis solidification curve of the vermicular cast iron which is just subjected to vermicular treatment and inoculation is collected by adopting a thermal analysis sample cup, a microstructure analysis sample is cast at the same time, the vermicular rate of the sample cup solidification sample is measured, then the collected stable thermal analysis solidification curve, corresponding characteristic values TEU, TER, TS and vermicular rate are stored in a database, and the data of the experimental part is shown in table 2.
Table 2: part of characteristic temperature data collected by experiment
Figure BDA0003696213570000061
To fully train the neural network, 20 sets of data were randomly drawn from the 24 sets of cooling curve thermal analysis data collected as training samples, and the remaining 4 sets were used to test the predicted effect of the neural network. Firstly, establishing a preliminary neural network model according to the method of the invention, after determining each parameter of the neural network, simulating and training the model by applying a Tensorflow neural network tool box, repeatedly correcting the weight of the connection between neurons until the error reaches a required value, as shown in fig. 3, which shows the variation trend of the network error during the training process, it can be seen that, after training, the minimum error of the neural network model for predicting the vermicular cast iron vermicular rate reaches 0.006, which shows that the BP neural network can accurately fit the relationship between the characteristic temperature and the vermicular rate.
Further, the accuracy of the neural network prediction model is verified. The characteristic temperature of the validation sample is used as the input of the trained neural network, and the corresponding creep rate prediction result can be obtained as shown in the following table 3.
TABLE 3 comparison of prediction results of creep rates
Figure BDA0003696213570000071
Comparing the real value with the predicted value of the data in the table, as shown in fig. 4, the prediction result of the vermicular cast iron shown in the graph is very close to the actual value, and the relative error is not more than 3.64%, which shows that the application of the trained neural network algorithm to predict the vermicular cast iron has higher accuracy, and the prediction model can effectively represent the relationship between the characteristic temperature point and the vermicular cast iron.
In conclusion, compared with the traditional linear regression-constructed model, the method disclosed by the invention is feasible for rapidly predicting the creep rate based on the neural network-constructed algorithm model, and the neural network has self-learning capability and self-adaptive capability due to the characteristics of the neural network, so that when the production raw material or the production condition changes, only enough data samples need to be provided again to retrain the original model, and the network model can be updated automatically. Therefore, the novel thermal analyzer adopting the neural network has wider feasibility and generalizability, and particularly has greater guiding significance on the creeping quality in the production process.
The above description is only for the preferred embodiment of the present invention, and is not intended to limit the structure of the present invention in any way. Any simple modification, equivalent change and modification of the above embodiments according to the technical spirit of the present invention are within the technical scope of the present invention.

Claims (5)

1. A neural network vermicular cast iron vermicular rate prediction method based on Tensorflow is characterized by comprising the following steps:
s1, recording the change of the temperature of the molten iron of the eutectic vermicular graphite cast iron along with time through a computer integrated temperature data acquisition card, drawing a cooling curve, and selecting a proper characteristic value from the cooling curve to be used for determining an input neuron and an output neuron of a neural network for detecting the vermicular graphite cast iron vermicularity rate;
s2, selecting a proper neural network neuron activation function to ensure smooth network training and ensure the quality of a model;
s3, according to an empirical formula
Figure FDA0003696213560000011
Calculating the number of the neurons of the hidden layer of the neural network;
s4, establishing a preliminary neural network prediction model according to the neural network input neuron, the neural network output neuron, the neural activation function and the number of neurons of the hidden layer determined in the step;
and S5, preprocessing input and output sample data, wherein the processed sample data is used for training the preliminary neural network prediction model established in the step S4 to obtain a neural network prediction model for detecting the vermicular cast iron vermicular rate.
2. The method for predicting vermicular cast iron creep rate of Tensorflow-based neural network according to claim 1, wherein the step S1 selects proper characteristic values from the cooling curve for determining input neurons and output neurons of the neural network for detecting vermicular cast iron creep rate, wherein the input neurons comprise a minimum temperature TEU for eutectic growth of vermicular cast iron, a eutectic ending temperature TS and a eutectic regeneration temperature TER, and the output neurons are vermicular cast iron creep rate.
3. The method for predicting the creep rate of Tensorflow-based neural network vermicular cast iron according to claim 1, wherein a proper neural network neuron activation function is selected in step S2, and the neuron activation function is a Relu function.
4. The method for predicting vermicular cast iron creep rate of neural network based on Tensorflow according to claim 1, wherein the empirical formula is used in step S3
Figure FDA0003696213560000012
And calculating the number of the neurons of the hidden layer of the neural network, wherein m is the number of the neurons of the input layer, n is the number of the neurons of the output layer, and c is a constant and has a value range of 4-13.
5. The neural network vermicular cast iron vermicular rate prediction method based on Tensorflow according to claim 1, wherein the preprocessing is performed on the input and output sample data in the step S5, the preprocessing is a normalization processing, and the sample data are all converted into data between 0 and 1.
CN202210680749.XA 2022-06-15 2022-06-15 Neural network vermicular graphite cast iron vermicular rate prediction method based on Tensorflow Pending CN115270397A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210680749.XA CN115270397A (en) 2022-06-15 2022-06-15 Neural network vermicular graphite cast iron vermicular rate prediction method based on Tensorflow

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210680749.XA CN115270397A (en) 2022-06-15 2022-06-15 Neural network vermicular graphite cast iron vermicular rate prediction method based on Tensorflow

Publications (1)

Publication Number Publication Date
CN115270397A true CN115270397A (en) 2022-11-01

Family

ID=83761348

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210680749.XA Pending CN115270397A (en) 2022-06-15 2022-06-15 Neural network vermicular graphite cast iron vermicular rate prediction method based on Tensorflow

Country Status (1)

Country Link
CN (1) CN115270397A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116046837A (en) * 2023-01-28 2023-05-02 潍柴动力股份有限公司 Vermicular rate determination method, device and equipment for vermicular cast iron molten iron

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116046837A (en) * 2023-01-28 2023-05-02 潍柴动力股份有限公司 Vermicular rate determination method, device and equipment for vermicular cast iron molten iron

Similar Documents

Publication Publication Date Title
CN112036081B (en) Method for determining addition amount of silicon-manganese alloy in converter tapping based on yield prediction
CN110066895B (en) Stacking-based blast furnace molten iron quality interval prediction method
CN108132277B (en) Method for predicting vermicular rate of hypereutectic composition vermicular graphite cast iron
CN115270397A (en) Neural network vermicular graphite cast iron vermicular rate prediction method based on Tensorflow
WO2021129350A1 (en) Converter steelmaking smelting late-stage carbon content forecasting method
CN116469481B (en) LF refined molten steel composition forecasting method based on XGBoost algorithm
CN113192568A (en) Refining furnace desulfurization terminal point forecasting method and system
CN111915080A (en) Raw fuel cost optimal proportioning method based on molten iron quality constraint
CN109252009A (en) BOF Steelmaking Endpoint manganese content prediction technique based on regularization extreme learning machine
CN113343576A (en) Prediction method of calcium yield in calcium treatment process based on deep neural network
CN112359159A (en) Hot blast stove automatic burning method and system based on deep reinforcement learning
CN112011664A (en) Furnace-front control method for vermicular cast iron vermicular effect
CN117850503B (en) Detection control system for casting spheroidization temperature control
CN116911057A (en) Real-time prediction method for temperature of molten pool in converter steelmaking converting process
CN102968644A (en) Method for predicting smelting finishing point of argon-oxygen refined iron alloy
CN111126550B (en) Neural network molten steel temperature forecasting method based on Monte Carlo method
CN109840309B (en) Method for calculating molten iron desulfurizer consumption
CN111518980A (en) Correction method and system for converter end point carbon content prediction model
CN108998617B (en) Method for controlling nodularity of nodular cast iron
Gao et al. End-point prediction of basic oxygen furnace (BOF) steelmaking based on improved twin support vector regression
CN117250220B (en) Hot metal thermal analysis method and device
CN106021860A (en) A nodular cast iron base iron metallurgical state comprehensive evaluation method
CN113408186A (en) Method for determining influence factors of metallurgical coke quality prediction
CN117556711B (en) Blast furnace coal injection optimization method, system, terminal and medium based on neural network
CN117572914B (en) LF refined molten steel temperature control method and device based on interpretable machine learning

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