CN1979496A - Copper-alloy pipe-material casting-milling technology parameter designing and optimizing method - Google Patents

Abstract

The invention discloses a method for designing and optimizing cast-rolling process parameters of copper alloy tubing, using database as design basic, using nerve network as design method of process parameters and indexes, and using genetic algorithm as process parameter optimizing means, integrating nerve network, genetic algorithm, finite element simulation, experiment design, CAD parameterized design and database technique into the process design and parameter optimization, designing and optimizing the cast-rolling process parameters of the copper alloy tubing. And the invention has high automation degree, and can be applied to machining deformation of copper alloy tubing, and make personnel short of special knowledge able to make accurate and standard machining process.

Description

Method of Design and Optimization of Copper Alloy Pipe Casting and Rolling Process Parameters

technical field

The present invention relates to copper alloy pipe processing technology, specifically a method of applying neural network, genetic algorithm, finite element simulation, test design, CAD parametric design and database technology to process parameter design and optimization to determine the copper alloy pipe material. Method for optimal process parameters of casting and rolling production.

technical background

The casting and rolling process, also known as horizontal continuous casting-planetary rolling tube feeding method, is a method for the production of precision copper tubes developed and invented by the Finnish OUTOKUMPU company in the mid-1980s. This process has significant advantages such as short process, high yield, low cost, and less investment in equipment. It is the current advanced ARC copper tube production technology. It cancels ingot heating, extrusion, etc., and the hollow billet is directly produced by the horizontal continuous casting unit. The billet is rolled by a three-roll planetary rolling mill. After rolling, the floating core is drawn, and then rolled into a coil online. It makes it possible to produce copper pipes with a single weight exceeding 500kg. The process includes three main processes: horizontal continuous casting, three-roller planetary rolling and drawing of floating core.

The casting and rolling production of copper alloy pipes is a typical multi-variety, multi-standard, multi-process processing technology. The pipes often change in size, shape and material, which makes the processing design of copper alloy pipes very variable, and the workload Large and inefficient, even engineers with specialized knowledge and rich experience are difficult to complete in a short period of time. Due to the large-scale and continuous production, once a process design error occurs, it is likely to cause a lot of waste of resources and seriously affect the production of the enterprise. Research and development of the design and optimization method of copper alloy pipe casting and rolling process parameters can guide pipe processing and production, improve product quality, speed up production and research and development cycles, reduce costs, and enhance enterprise competitiveness.

In the past, the design of process parameters was based on the engineer's own experience and repeated field tests to determine the production process parameters. This method not only has low intelligence and automation, but also has a single design method, which affects the normal production of the enterprise, and it is difficult to effectively design process parameters. , especially it is difficult to ensure that the obtained process parameters are optimal parameters. At present, some enterprises also use the form of reasoning machines to build knowledge bases by summarizing and sorting out empirical knowledge and formulas. According to the expression model of knowledge, the knowledge is mapped to a structure or program that can be recognized by the computer, so that it can reason about the design and optimization of process parameters in a logical way. However, metal forming is a very complex deformation process, which has both material nonlinearity and geometric nonlinearity, coupled with the influence of complex external constraints, resulting in a very complicated forming process. The interactive effects of various process parameters make it difficult to use traditional reasoning machines to find the optimal combination of process parameters. Moreover, because the acquisition of empirical knowledge is indirect, it is difficult for the inference engine to acquire knowledge, and the knowledge combination explosion will often be caused by the bad structure of the knowledge base.

Contents of the invention

In order to overcome the shortcomings of the existing methods, the purpose of the present invention is to propose a method for designing and optimizing the parameters of the copper alloy pipe casting and rolling process. By adopting the present invention, the intelligent process parameter design and optimization of the three main processes of horizontal continuous casting, three-roll planetary rolling and floating core head drawing in the casting and rolling process can be realized according to the finished product specification and the alloy material used, and according to the copper pipe production process , it has a high degree of automation and can be applied to various complex processing deformations, so that personnel who lack rich professional knowledge can also formulate accurate and standardized processing techniques.

The technical solution of the present invention is based on the database, the neural network is the design method of process parameters and process indicators, the genetic algorithm is the process parameter optimization means, and the neural network, genetic algorithm, finite element simulation, test design and CAD parametric design are comprehensively integrated. And database technology, and applied to the process design and parameter optimization, the process parameter design and optimization of the three process steps of copper alloy pipe casting and rolling, in order to obtain the optimal process parameters. details as follows:

1. Database design

Establish a database for horizontal continuous casting, three-roll planetary rolling and moving core drawing, and use the database as the basis for parameter design and optimization. The database stores the production data, standard data and temporary data of these three processes in the factory; the factory Production data includes: specification library, equipment library, operation library, raw material library, mold library, spare parts library, component library, process design library, and design result library; standard data include national standards for products of various specifications as well as European, American, Japanese and other countries’ standards. Standard documents; temporary data include initial design data and intermediate data generated by calculation.

2. Finite element simulation

The use of finite element simulation technology can solve various complex boundary and nonlinear problems. It can improve product quality, shorten product development cycle, reduce cost and increase productivity when used in the processing field. With the improvement of computer level, simulating forming process with commercial finite element software has become a powerful analysis method, and the accuracy can meet the application of engineering design, and can replace on-site process test. Designers input geometric parameters, process parameters and material parameters in the finite element pre-processing, and after calculation, they can get the required calculation results in the post-processing environment. Using finite element numerical simulation steps, the horizontal continuous casting temperature field, temperature gradient and cooling rate values in horizontal continuous casting are respectively obtained, and the crack initiation tendency value is derived; the rolling force calculation value and rolling forming defects in three-roll planetary rolling The simulation results; and the calculation value of the drawing force in the drawing of the moving core and the simulation results of the drawing forming defects. Roll forming defects include breakage, tearing and jamming. The defect of drawing forming is breakage.

Since the finite element modeling requires corresponding professional knowledge and long calculation, in order to allow non-professionals to quickly design, the present invention arranges the finite element numerical simulation scheme with the method of uniform experimental design. Uniform experimental design can economically, scientifically and rationally arrange the number of finite element numerical simulations. Reasonable test program design can obtain comprehensive information reflecting the quantitative law between input and output with less time for numerical simulation and lower cost.

3. Neural Network

Multi-layer neural network has highly nonlinear fitting properties and wide adaptability to multiple input and multiple output problems. It is good at dealing with reasoning in associative memory, image thinking, etc., and has self-organization and self-learning capabilities. It can solve the bottleneck problem of knowledge acquisition. Neural network-based methods can be viewed as an implicit representation of rules.

For the existing process design scheme, the present invention learns by multi-layer artificial neural network, calculates and obtains the drawing matching mold design process parameter ( The total drawing passes and the copper pipe wall thickness and outer diameter value of each pass).

Neural network can also be combined with finite element numerical calculation to mine rich knowledge of finite element numerical calculation and find useful information rules. The threshold value and weight matrix obtained after neural network training are used as an implicit rule, which can map the relationship between process parameters and process indicators. The process indicators of copper alloy pipe casting and rolling include the crack initiation tendency value of horizontal continuous casting, the rolling force and rolling forming defect value of three-roll planetary rolling, the drawing force and drawing forming defect value of floating core drawing . According to the finite element numerical simulation calculation results, the present invention learns through a multi-layer artificial neural network, and uses the trained neural network to calculate the horizontal continuous casting process, the three-roll planetary rolling process and the corresponding process parameters under different conditions. The process index value in the drawing process of the moving core.

4. Genetic Algorithm

The genetic algorithm is different from the traditional optimization algorithm. It is a highly parallel, random and adaptive search algorithm developed by referring to the natural selection and evolution mechanism in the biological world. It is more likely to obtain the global optimal solution. It uses artificial evolution to search the target space randomly, regards the possible solutions in the problem domain as an individual or chromosome of the group, and encodes each individual into a symbolic string form, simulating Darwin's genetic selection and natural elimination In the process of biological evolution, repeated operations based on genetics (heredity, crossover and mutation) are carried out on the population, and each individual is evaluated according to the predetermined target fitness function, and the better ones are continuously obtained according to the evolution rules of survival of the fittest and survival of the fittest. At the same time, the optimal individual in the optimization group is searched by the global parallel search method to obtain the optimal solution that meets the requirements. The step of using genetic algorithm to search for optimal process parameters is especially suitable for dealing with complex and nonlinear problems that cannot be solved by previous search algorithms, and has been widely used in engineering fields.

The process parameters to be optimized in the process parameters include the drawing system parameters and cooling system parameters of horizontal continuous casting; the roll deflection angle, roll inclination angle, opening degree and trolley speed of three-roll planetary rolling; Outer die cone angle, core head cone angle, drawing speed, outer die sizing section length and core head sizing section length for each pass. The invention uses the genetic algorithm as an optimization method to search for optimal process parameters.

First, according to the corresponding optimization objectives, parameter encoding is performed to form an initial population; the neural network calculates the process index value of each individual, that is, the fitness value in the genetic algorithm, and then performs operator operations; the evolution of the population from generation to generation, Until the optimal solution is found, the best process parameters are determined, and the results are made into process cards and design documents. Among them: the operator operation includes three basic forms of selection, crossover and mutation.

5. Graphical result expression of process parameter design and optimization

Using the CAD parametric design method, the best process parameters obtained are combined with CAD software to carry out the CAD parametric design of the rolling roll shape CAD in the three-roller planetary rolling of the mold, and the core head die CAD parametric design of the floating core head drawing, and the design calculation , data processing and graphics rendering for comprehensive processing.

The present invention has the following advantages:

1. High degree of automation. The present invention integrates neural network, genetic algorithm, finite element simulation, test design, CAD parametric design and database, and can avoid adverse effects such as fracture, wrinkling and necking during the forming process of copper alloy pipe material casting, and is a realization An intelligent method for the design and optimization of copper alloy pipe casting and rolling process parameters.

2. The traditional process is mainly based on the experience of the designer in order to avoid the adverse effects of fracture, excessive forming force, unreasonable mold design and other adverse effects during the forming process of the copper alloy pipe casting and rolling process, and repeatedly modify some parameters of the forming process Or modifying the shape of the mold, the traditional process is costly and the product development cycle is long, which can no longer adapt to the fierce market competition worldwide and the development requirements of modern industry. Due to the high degree of automation, low cost and short product development cycle, the present invention can be applied to the complex casting and rolling forming process of copper alloy pipes, enabling personnel lacking rich professional knowledge to formulate accurate and standardized processing processes.

Description of drawings

Figure 1-1 is the finite element model of the horizontal continuous casting process.

Figure 1-2 is the finite element model of the three-roll planetary rolling process.

Figure 1-3 is the finite element model of the drawing process of the moving core.

Figure 2 is the neural network structure of the crack initiation tendency value in horizontal continuous casting.

Fig. 3 is the neural network structure for prediction of rolling force and rolling forming defects in three-roll planetary rolling.

Figure 4 is the neural network structure for the prediction of the drawing force and drawing defects of the moving core.

Fig. 5 is the artificial neural network structure for the drawing and matching die design of the swimming core.

Figure 6 shows the parameters of the cooling system and continuous casting casting system optimized by genetic algorithm.

Figure 7 is the genetic algorithm optimization of three-roller planetary rolling parameters.

Fig. 8 is a genetic algorithm optimization of the parameters of the swimming core.

Fig. 9 is the operation process of the design and optimization method of the copper alloy pipe casting and rolling process parameters.

Detailed ways

Further illustrate the present invention below in conjunction with accompanying drawing.

Example

The present invention applies neural network, genetic algorithm, finite element simulation, test design, CAD parametric design, and database technology to process design and parameter optimization, and integrates them comprehensively to optimize the design of each process step of copper alloy pipe casting and rolling. , to get the optimal process parameters. Process parameter design includes: horizontal continuous casting billet drawing system, optimization design of cooling system; velocity field calculation and rolling parameter design of three-roll planetary rolling; Parameter optimization design.

details as follows:

1) Establish the database of horizontal continuous casting, three-roll planetary rolling and moving core drawing, and use the database as the basis for parameter design and optimization. The production data, standard data and temporary data of these three processes in the factory are stored in the database. Factory production data includes: specification library, equipment library, operation library, raw material library, mold library, spare parts library, component library, process design library and design result library; standard data include national standards for products of various specifications, Europe, America, Japan, etc. National standard documents; temporary data include initial design data and intermediate data generated by calculation.

2) Establish the finite element model of horizontal continuous casting, three-roll planetary rolling and moving core drawing.

The knowledge of finite element numerical calculation comes from the input variables of commercial finite element preprocessing and the corresponding numerical calculation results. Designers input geometric parameters, process parameters and material parameters in the finite element pre-processing, and after calculation, they can get the required calculation results in the post-processing environment. Using finite element numerical simulation steps, the horizontal continuous casting temperature field, temperature gradient and cooling rate values in horizontal continuous casting can be obtained respectively, and the crack initiation tendency value can be derived; the calculation value of rolling force and rolling forming in three-roll planetary rolling Defect simulation results; and the calculation value of the drawing force in the drawing of the moving core and the simulation results of drawing forming defects. Among them, the defects of rolling forming include breakage, tearing and jamming, and the defects of drawing forming are breaking.

In order to obtain comprehensive information that reflects the quantitative law between the finite element simulation input and the finite element simulation output while reducing the number of finite element simulations, time and the cost of simulation, the present invention arranges the finite element with the method of uniform test design Numerical simulation scheme. Uniform experimental design can economically, scientifically and rationally arrange the number of finite element numerical simulations.

① Horizontal continuous casting finite element simulation

The finite element simulation of horizontal continuous casting is arranged by the method of uniform design of experiments. Through the finite element simulation of horizontal continuous casting, the temperature field T, temperature gradient G and cooling rate value R corresponding to copper tube billets of different sizes, material parameters, casting system and cooling system can be obtained. The finite element model is shown in the figure 1-1 (where 1 is the cooling system, 2 is the copper alloy melt, 3 is the continuous casting copper tube, 4 is the graphite mandrel, and 5 is the graphite lining used in the crystallizer). In horizontal continuous casting, the temperature, temperature gradient and cooling rate are directly proportional to the probability of crack generation. Therefore, according to the temperature field, temperature gradient and cooling rate value obtained by finite element simulation, the value of each finite element element node is calculated by the following formula Crack initiation propensity value.

${\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; T</span>}<span\; class="notranslate">\; *</span>\mathrm{<span\; class="notranslate">\; G</span>}<span\; class="notranslate">\; *</span>\mathrm{<span\; class="notranslate">\; <figure-callout\; id="10"\; label="R"\; filenames="A20051004790300181.png"\; state="\{\{state\}\}">R</figure-callout></span>}}{{\mathrm{<span\; class="notranslate">\; 10</span>}}^{\mathrm{<span\; class="notranslate">\; 5</span>}}}$

In the formula, C _i is the crack initiation tendency value, T is the temperature, G is the temperature gradient, R is the cooling rate, and i is the unit node number.

Find the maximum crack initiation tendency value C _Max in all unit nodes, and calculate the average value of crack initiation tendency, which is expressed as follows:

$\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; C</span>}}<span\; class="notranslate">\; =</span>\frac{\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}{\mathrm{<span\; class="notranslate">\; N</span>}}$

是裂纹萌生倾向平均值，i表示单元节点号，N是有限元模型的节点总数。In the formula,

is the average value of crack initiation tendency, i represents the unit node number, and N is the total number of nodes in the finite element model.

Table 1 Influencing parameters of horizontal continuous casting:

Copper pipe billet size Material parameter X ₃ Drawing system cooling system Copper tube billet outer diameter size X ₁ Copper Drawing time X ₄ Casting temperature X ₉ Wall thickness of copper tube billet X ₂ Cupronickel B10 One stop time X ₅ Inlet water temperature X ₁₀

Cupronickel B30 Push X ₆ Water pressure X ₁₁  … Travel time X ₇  … Second stop time X ₈

In this way, the quantitative relationship between the influencing parameters of horizontal continuous casting and the tendency of crack initiation is established.

②Three-high planetary rolling finite element simulation

The finite element simulation of three-roll planetary rolling is arranged by the method of uniform design of experiments. Through the finite element simulation of three-roll planetary rolling, the quantitative relationship between copper tube rolling blanks of different sizes, material parameters, rolling parameters, rolling force values and rolling defect values can be obtained. The finite element model is shown in Figure 1-2 Shown (wherein, 6 is a roll, 7 is a copper tube, and 8 is a mandrel). Here, the rolling defect value is represented by 0 and 1, 1 represents no defect, and 0 represents defect.

Table 2 Influencing parameters of three-roll planetary rolling:

Specification and size of copper tube rolling billet Material parameter Y ₇ Rolling parameters Copper tube billet outer diameter size X ₁ Copper Roll deflection angle Y ₁ Wall thickness of copper tube billet X ₂ Cupronickel B10 Roller inclination angle Y ₂ Copper tube outer diameter after rolling D _Roll Cupronickel B30 Opening Y ₃ Copper tube wall thickness after rolling S _Roll  … Roll speed Y ₄  … Speed of cart Y ₅  … Friction coefficient Y ₆

③Swimming core drawing finite element simulation

The method of uniform design of experiments is used to arrange the finite element simulation of the drawing of the moving core head. Through the drawing simulation of the swimming core, the quantitative relationship between copper pipes of different sizes, material parameters, drawing parameters, drawing force P _Draw and drawing defect value Q _Draw can be obtained. The finite element model is shown in Figure 1-3 Shown (wherein 9 is an outer mold, 10 is a copper pipe, and 11 is a swimming core head). Here, the drawing defect value is represented by 0 and 1, 1 represents no defect, and 0 represents defect.

Table 3 Parameters affecting the drawing of the moving core head:

Copper pipe drawing specification and size Material parameter Z ₈ Drawing parameters Copper tube outer diameter before drawing D _Draw0 Copper Drawing speed Z ₄ Copper pipe wall thickness before drawing S _Draw0 Cupronickel B10 Core cone angle Z ₅ Copper tube outer diameter after drawing D _Draw1 Cupronickel B30 Outer die cone angle Z ₆ Copper pipe wall thickness after drawing S _Draw1  … Friction coefficient Z ₇  … The length of the sizing section of the outer mold Z ₉  … The length of the sizing section of the core head Z ₁₀

3) neural network

The simulated input quantity and the result output quantity obtained by finite element simulation with uniform experimental design arrangement are sorted and homogenized, and used as the training samples of the neural network. For the three processes of horizontal continuous casting, three-roll planetary rolling, and swimming core drawing, use multi-layer (this embodiment uses 3 layers) artificial neural networks to learn corresponding sample data, train and establish corresponding neural networks . Using the trained neural network, after testing, the crack initiation tendency value of the corresponding horizontal continuous casting process, the rolling force value and rolling defect value in the three-roll planetary rolling can be calculated under different conditions. Drawing force value and drawing defect value in moving core drawing.

①Neural network of crack initiation tendency value in horizontal continuous casting

The neural network structure of the crack initiation tendency value in horizontal continuous casting is as follows: the input layer has 11 nodes, and its parameters are the input quantities of the finite element simulation, which are the outer diameter of the slab X ₁ , the wall thickness of the slab X ₂ , and the material parameters X ₃ , casting time X ₄ , first stop time X ₅ , push X ₆ , push time X ₇ , second stop time X ₈ , casting temperature X ₉ , inlet water temperature X ₁₀ , water pressure X ₁₁ . The material parameters are 0 for copper, 1 for cupronickel B10, 2 for cupronickel B30, and so on for other copper alloy materials. There are two nodes in the output layer, whose parameters are the output of the finite element simulation results, which are the average value C of the crack initiation tendency and the maximum value C _Max of the crack initiation tendency. The neural network adopts the BP network based on the Levenberg-Marquardt optimization algorithm (LM algorithm), the hidden layer is a Sigmoid activation function, and the output layer is a Purelin activation function. The LM algorithm can greatly shorten the training time.

After normalizing the finite element simulation input and output arranged by uniform test design, a training sample file is generated and stored in the database. After training with the neural network, if the error is within the allowable range, the threshold value and weight matrix of the neural network will be stored in the database as a neural network matrix file. In this way, a mapping relationship model is established, which can map the relationship between various influencing parameters of horizontal continuous casting and the value of crack initiation tendency, which is an implicit relationship (see Figure 2).

If there is a newly added finite element simulation result, it can be normalized and integrated into the original training sample file. After retraining, the newly obtained neural network threshold and weight matrix can be stored as a neural network matrix file. in the database. The training sample file and the neural network matrix file can be updated at any time as the number of finite element simulations increases.

②Neural network for prediction of rolling force and rolling forming defects in three-roll planetary rolling

The neural network structure for predicting the rolling force and rolling forming defects of the three-roll planetary rolling is as follows: the input layer has 10 nodes, and its parameters are the input quantities of the finite element simulation of the three-roll planetary rolling, which are the roll deflection angle Y ₁ , Roll inclination angle Y ₂ , opening degree Y ₃ , roll speed Y ₄ , trolley speed Y ₅ , friction coefficient Y 6 , material parameters Y ₇ , elongation coefficient Y ₈ , initial diameter-to-wall ratio _{Y 9 and} reduction-to-wall ratio _Y10 . The material parameters are 0 for copper, 1 for cupronickel B10, 2 for cupronickel B30, and so on for other copper alloy materials. According to the principle of similarity, the elongation coefficient, the initial diameter-to-wall ratio and the reduced diameter-to-wall ratio can be used to represent the change in the size of the copper tube before and after rolling, and the three values are expressed as follows:

Elongation $Y_8=\frac{(x_1-x_2)*x_2}{({\mathrm{D.}}_{\mathrm{roll}}-S_{\mathrm{roll}})*S_{\mathrm{roll}}}$

initial diameter-to-wall ratio $Y_{}$${}_9=\frac{x_1}{x_2}$

Wall and diameter reduction ratio $Y_{}$${}_{10}=\frac{(x_2-S_{\mathrm{roll}})/x_2}{(x_1-{\mathrm{D.}}_{\mathrm{roll}})/x_1}$

In the formula, X ₁ , X ₂ are the outer diameter and wall thickness of the copper tube before rolling, that is, the outer diameter of the copper tube slab and the wall thickness of the copper tube slab; D _Roll and S _Roll are the copper tube after rolling, respectively. outside diameter and wall thickness.

There are two nodes in the output layer, whose parameters are the output of the finite element simulation results, which are the drawing force P _Roll and the drawing defect value Q _Roll . The neural network adopts the BP network based on the Levenberg-Marquardt optimization algorithm (LM algorithm), the hidden layer is a Sigmoid activation function, and the output layer is a Purelin activation function.

After normalizing the finite element simulation input and output arranged by uniform test design, a training sample file is generated and stored in the database. After training with the neural network, if the error is within the allowable range, the threshold value and weight matrix of the neural network will be stored in the database as a neural network matrix file. In this way, a mapping relationship model is established, which can map the relationship between the various influencing parameters of the drawing of the moving core head, the drawing force and the value of the drawing defect. This relationship is an implicit relationship (see Figure 3).

If there is a newly added finite element simulation result, it can be normalized and integrated into the original training sample file. After retraining, the newly obtained neural network threshold and weight matrix can be stored as a neural network matrix file. in the database. The training sample file and the neural network matrix file can be updated at any time as the number of finite element simulations increases.

③Neural network for prediction of drawing force and drawing defect of swimming core

The neural network structure for the prediction of the drawing force and the drawing defect of the swimming core is as follows: the input layer has 10 nodes, and its parameters are the input quantities of the finite element simulation of the swimming core drawing, which are the initial diameter-to-wall ratio Z ₁ , minus Diameter-to-wall ratio Z ₂ , elongation coefficient Z ₃ , drawing speed Z ₄ , core cone angle Z ₅ , external die cone angle Z ₆ , friction coefficient Z ₇ , material parameters Z ₈ , length of external die sizing section Z _9. Length Z ₁₀ of the sizing section of the core head. The material parameters are 0 for copper, 1 for cupronickel B10, 2 for cupronickel B30, and so on for other copper alloy materials. According to the principle of similarity, the elongation coefficient, the initial diameter-to-wall ratio and the reduced-diameter-to-wall ratio can be used to represent the change in the size of the copper tube before and after drawing, and the three values are expressed as follows:

Elongation $Z_1=\frac{({\mathrm{D.}}_{\mathrm{draw}0}-S_{\mathrm{draw}0})*S_{\mathrm{draw}0}}{({\mathrm{D.}}_{\mathrm{draw}1}-S_{\mathrm{draw}1})*S_{\mathrm{draw}1}}$

initial diameter-to-wall ratio $Z_2=\frac{{\mathrm{D.}}_{\mathrm{draw}0}}{S_{\mathrm{draw}0}}$

Wall and diameter reduction ratio $Z_{}$${}_3=\frac{(S_{\mathrm{draw}0}-S_{\mathrm{draw}1})/S_{\mathrm{draw}0}}{({\mathrm{D.}}_{\mathrm{draw}0}-{\mathrm{D.}}_{\mathrm{draw}1})/{\mathrm{D.}}_{\mathrm{draw}0}}$

In the formula, D _Draw0 and S _Draw0 are the outer diameter and wall thickness of the copper tube before drawing, respectively, and D _Draw1 and S _Draw1 are the outer diameter and wall thickness of the copper tube after drawing, respectively.

There are two nodes in the output layer, whose parameters are the output of the finite element simulation results, which are the drawing force P _Draw and the drawing defect value Q _Draw . The neural network adopts the BP network based on the Levenberg-Marquardt optimization algorithm (LM algorithm), the hidden layer is a Sigmoid activation function, and the output layer is a Purelin activation function.

After normalizing the finite element simulation input and output arranged by uniform test design, a training sample file is generated and stored in the database. After training with the neural network, if the error is within the allowable range, the threshold value and weight matrix of the neural network will be stored in the database as a neural network matrix file. In this way, a mapping relationship model is established, which can map the relationship between the various influencing parameters of the drawing of the moving core head, the drawing force and the value of the drawing defect. This relationship is an implicit relationship (see Figure 4).

If there is a newly added finite element simulation result, it can be normalized and integrated into the original training sample file. After retraining, the newly obtained neural network threshold and weight matrix can be stored as a neural network matrix file. in the database. The training sample file and the neural network matrix file can be updated at any time as the number of finite element simulations increases.

4) There are many design methods for matching molds in the drawing of moving cores, mainly including the method of decreasing equation, the method of decreasing coefficient, the method of metal hardening degree, the method of decreasing proportional number sequence and the method of Kd-Ks, etc. These methods have certain practical value, but they are not necessarily suitable for actual production. Die matching design needs to consider various external factors such as equipment conditions, lubrication conditions, material properties of tube blanks, etc., so these theoretical methods are not universal. Experienced engineers usually have their own set of effective mold matching design methods suitable for their own enterprise characteristics. In order to make full use of this experience, the method of artificial neural network is used to train and study the matching design schemes of various specifications of products collected from the field. In this way, the neural network can automatically carry out the mold matching design according to the engineer's design idea.

According to the existing drawing matching schemes of various specifications, select the normal and stable matching schemes as the effective matching schemes, and use them as the sample data of the neural network after sorting out. Taking a mold matching scheme of a standard size as an example, the finishing method is as follows:

The total number of drawing passes for a copper tube of a certain size is: n; the outer diameter and wall thickness of the finished copper tube are: D _n and S _n ; the outer diameter and wall thickness of the copper tube after three-roll planetary rolling The outer diameter and wall thickness of the copper tube before drawing are: D ₀ and S ₀ ; the outer diameter and wall thickness of the copper tube after the i-th drawing are: D _i and S _i . There are five sample input quantities: D _n , S _n , D ₀ , S ₀ and i/n, and two sample output quantities are D _i and S _i . In this way, n-1 samples can be formed by drawing and matching schemes for copper pipes of each specification and size. The samples formed by the mold matching scheme of this specification can be expressed in the following table:

sample number Sample input volume Sample output D ₀ D _n S ₀ S _n i/n D. _Si 1 D ₀ D _n S ₀ S _n 1/n D ₁ S ₁ 2 D ₀ D _n S ₀ D _n 2/n _D2 S ₂ 3 D ₀ D _n S ₀ S _n 3/n _D3 S ₃ ..... ..... ..... ..... ..... ..... ..... ..... i D ₀ D _n S ₀ S _n i/n D. _Si n-1 D ₀ D _n S ₀ S _n (n-1)/n D _n-1 S _n-1

According to this sorting method, various effective model matching schemes are combined to form a training sample file, which is stored in the database, and the training sample file is learned by using a multi-layer artificial neural network. The structure of the artificial neural network for the drawing and matching mold design of the swimming core can be shown in Figure 5.

After the trained artificial neural network is tested, if it meets the test requirements, the threshold value and weight matrix of the neural network will be stored in the database as a neural network matrix file. In this way, a mapping relationship model is established, which can map the relationship between the size specification of the copper pipe and the drawing and matching die.

Before using the neural network to design the drawing matching mold of the swimming core, the value of the total drawing pass n must be calculated first. Its calculation method is usually determined based on empirical formulas.

来确定总道次数。平均道次延伸系数根据不同的E_n值来确定的，E_n值为材料厚度指数According to the outer diameter D ₀ and wall thickness S ₀ of the copper tube before drawing, the outer diameter D _n and wall thickness S _n of the finished copper tube, and the average pass elongation coefficient

to determine the total number of runs. The average pass elongation coefficient is determined according to different E _n values, and the E _n value is the material thickness index

${\mathrm{<span\; class="notranslate">\; E.</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; =</span>\frac{{\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}}{{\mathrm{<span\; class="notranslate">\; D.</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}}<span\; class="notranslate">\; \&times;</span>\mathrm{<span\; class="notranslate">\; 10</span>}$

When the performance of the drawing machine is good, the lubrication is good, and the condition of the mold is normal, E _n can be appropriately larger, that is, E _n+1 ; otherwise, it should be smaller, that is, E _n-1 .

The formula for calculating the total elongation coefficient λ _∑ of drawing is as follows:

${\mathrm{<span\; class="notranslate">\; \&lambda;</span>}}_{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}<span\; class="notranslate">\; =</span>\frac{{\mathrm{<span\; class="notranslate">\; D.</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}}{{\mathrm{<span\; class="notranslate">\; D.</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}}<span\; class="notranslate">\; \&Center\; Dot;</span>\frac{{\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}}{{\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}}$

The formula for calculating the total number of drawing passes n is:

$\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; ln</span>}{\mathrm{<span\; class="notranslate">\; \&lambda;</span>}}_{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; ln</span>}\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; \&lambda;</span>}}}$

The dimensions and specifications of the product to be molded, that is, the outer diameter D _n and wall thickness S _n of the finished copper tube, and the outer diameter and wall thickness of the copper tube after three-roll planetary rolling, that is, the outer diameter D of the copper tube before drawing ₀ and wall thickness S ₀ , and the value of i/n, i refers to the size of the copper tube after drawing the i-th pass (1≤i<n) to be predicted at present, and n is the total drawing pass calculated according to the empirical formula secondary value. These values are used as the neural network prediction input. After the neural network matrix file is transferred, the neural network is used to carry out mold matching design, and the dimensions D _i and S _i of the copper tube after the i-th drawing can be obtained.

5) Genetic algorithm

Use genetic algorithm to search for the optimized parameters of continuous casting cooling system and continuous casting casting system in horizontal continuous casting, rolling parameters in three-roll planetary rolling and drawing parameters of floating core drawing, and perform parameter coding. Constitute the initial population; calculate the fitness value of each individual according to the neural network, and then perform operator operations; the population evolves from generation to generation until the optimal solution is found, the optimal process design parameters are determined, and the results are made into process cards and design files. Among them: the operator operation includes three basic forms of selection, crossover and mutation.

① Genetic algorithm optimizes the parameters of cooling system and continuous casting casting system in horizontal continuous casting

In horizontal continuous casting, casting time, first stop time, push stroke, push time, second stop time, casting temperature, inlet water temperature, and water pressure are the key factors affecting the value of crack initiation tendency in horizontal continuous casting. The minimum and maximum value of the horizontal continuous casting crack initiation tendency value are lower than the safety value as the optimization target, and the outer diameter of the slab, the wall thickness of the slab, and the material parameters are used as fixed parameters, and the genetic algorithm is used to find the tensile strength under this condition. The optimal values of billet time, first stop time, push, push time, second stop time, casting temperature, inlet water temperature and water pressure are shown in Figure 6.

The evaluation function of the genetic algorithm is also called the fitness function, which converts the optimization goal to be solved into a fitness function. The fitness value here can be calculated by the horizontal continuous casting crack initiation tendency value prediction neural network to calculate the fitness value of each individual. Because the optimization goal is the problem that the average minimum and maximum value of the crack initiation tendency value are lower than the safe value, the construction formula is:

The fitness value is: $\mathrm{fit}\left(x_i\right)=\frac{C-C_{\max }\left(x_i\right)}{|C-C_{\max }\left(x_i\right)|}\&times;\frac{1}{\stackrel{\&OverBar;}{C}\left(x_i\right)}$

Among them, C _max (X _i ),

In the formula, X _i represents each input quantity in the neural network for predicting the crack initiation tendency value in horizontal continuous casting; C is the safety value of the crack initiation tendency value, which is a constant coefficient and can be selected and determined according to the actual situation; C _max (X _i ) , C(X _i ) represents the maximum value and the average value of the crack initiation tendency obtained by the neural network.

Genetic algorithms can use floating-point encoding and integer encoding. The population size of the genetic algorithm is generally 20 to 100. Generally speaking, choosing a larger number of initial populations can process more solutions at the same time, so it is easy to find the global optimal solution. The disadvantage is that each iteration time is increased. The hybridization rate of the genetic algorithm is generally set at 0.4~0.9. The higher the frequency of the hybridization operation, the faster it can converge to the most promising optimal solution area, but too high frequency may also lead to premature convergence. The mutation rate of the genetic algorithm generally ranges from 0.001 to 0.1. The larger the population size and chromosome length, the smaller the mutation rate. The maximum evolutionary generation of the genetic algorithm, as a simulation termination condition, depends on the specific situation and according to multiple trial runs, generally in the range of 100 to 500 generations.

In this embodiment, the genetic algorithm used for horizontal continuous casting optimization adopts floating-point coding, the value of the initial population is 100, the value of the hybridization rate is 0.85, the value of the mutation rate is 0.08, and the value of the maximum evolution algebra is 200.

②Genetic algorithm to optimize three-high planetary rolling parameters

In the three-roll planetary rolling, the roll deflection angle, roll inclination angle, opening degree, trolley speed, these die parameters and process parameters are the key factors affecting the rolling force and rolling forming defects. Taking the minimum rolling force and no rolling defect as the optimization goal, the material parameters, friction coefficient, roll speed, elongation coefficient, initial diameter-to-wall ratio, and wall-to-diameter ratio are used as fixed parameters, and the genetic algorithm is used to find the condition The optimal values of roll deflection angle, roll inclination angle, opening degree and trolley speed are shown in Figure 7.

The evaluation function of the genetic algorithm is also called the fitness function, which converts the optimization target to be solved into a fitness function. The fitness value here can be calculated by the rolling force of the three-roll planetary rolling and the neural network for forming defect prediction. fitness value. Because the optimization goal is the minimum rolling force and no forming defects, the construction formula is:

The fitness value is: $\mathrm{fit}\left(Y_i\right)=\frac{Q_{\mathrm{roll}}\left(Y_i\right)}{P_{\mathrm{roll}}\left(Y_i\right)}$

Among them, P _Roll (Y _i ), S _Roll (Y _i )=ANN _{three-roll planetary rolling rolling force and forming defect prediction neural network} (Y _i )

In the formula, Y _i represents the rolling force of the three-roll planetary rolling and each input value in the forming defect prediction neural network; P _Roll (Y _i ), S _Roll (Y _i ) represent the rolling force value obtained by the neural network and defect predictions.

In this embodiment, the genetic algorithm encoding used for parameter optimization of the three-roll planetary rolling adopts floating-point encoding, the value of the initial population is 80, the value of the hybridization rate is 0.75, the value of the mutation rate is 0.08, and the value of the maximum evolution algebra is 100.

③Genetic algorithm optimizes the parameters of the swimming core head

In the drawing of the moving core of copper alloy, the cone angle of the outer mold, the cone angle of the core, the drawing speed, the length of the sizing section of the outer mold and the length of the sizing section of the core, these mold parameters and process parameters affect the drawing force , The key factor of drawing forming defects. Taking the minimum drawing force and no forming defect as the optimization goal, the material parameters, friction coefficient, drawing speed, elongation coefficient, initial diameter-to-wall ratio and wall-to-diameter ratio are used as fixed parameters, and the genetic algorithm is used to find the The optimal values of the cone angle of the outer die, the taper angle of the core head, the length of the sizing section of the outer die and the length of the sizing section of the core head are shown in Figure 8.

The evaluation function of the genetic algorithm is also called the fitness function, which converts the optimization target to be solved into a fitness function. The fitness value of this embodiment can be calculated by the drawing force of the swimming core head and the forming defect prediction neural network. The fitness value of an individual. Because the optimization goal is the minimum value of the pulling force and no forming defects, the construction formula is:

$\mathrm{<span\; class="notranslate">\; fit</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; Z</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{{\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; draw</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; Z</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span>}{{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; draw</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; Z</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span>}$

P _Draw (Z _i ), S _Draw (Z _i )＝The drawing force and forming defect prediction neural network (Z _i ) _{of ANN swimming core drawing}

In the formula, Z _i represents the drawing force of the moving core and the input quantities in the forming defect prediction neural network; P _Draw (Z _i ), S _Draw (Z _i ) represent the drawing force obtained by the neural network values and defect predictions.

In this embodiment, the genetic algorithm encoding used for the optimization of the parameters of the swimming core adopts floating-point encoding, the value of the initial population is 50, the value of the hybridization rate is 0.8, the value of the mutation rate is 0.1, and the value of the maximum evolution algebra is 100.

6) Adopt the CAD parametric design method, combine the best process parameters obtained with CAD software to carry out the CAD parametric design of the rolling roll shape CAD in the three-roller planetary rolling of the mold, and the core head mold CAD of the floating core head drawing. Comprehensive processing of design calculation, data processing and graphic drawing.

The operation process of the copper alloy pipe casting and rolling process parameter design and optimization method established by the present invention is shown in Figure 9, specifically:

(1) According to the system prompt, input the product material type and light pipe size specification to be designed;

(2) Input the size of the horizontal continuous casting copper pipe;

(3) Horizontal continuous casting design: The neural network calculates the crack initiation tendency value of the horizontal continuous casting temperature field, and optimizes the cooling system and continuous casting casting system parameters of the horizontal continuous casting according to the genetic algorithm. If the final optimized parameters are obtained, then Execute step (5), otherwise return to step (3);

(4) Input the size of the three-roll planetary rolling copper tube;

(5) Three-roller planetary rolling design: use neural network to calculate rolling force and rolling forming defect value; optimize rolling parameters according to genetic algorithm; if you get the final optimized parameters, then perform step (6), otherwise return Step (5);

(6) CAD parametric design of roll shape CAD for three-roll planetary rolling;

(7) Design of drawing and matching mold for swimming core head: use neural network to design drawing and matching mold;

(8) Use the neural network to calculate the drawing force and drawing forming defect value of each pass; and use the genetic algorithm to optimize and design the drawing parameters of the swimming core;

(9) CAD parametric design of the drawing of the moving core; if the final optimized parameters are obtained, the program is ended, otherwise, return to step (8).

In summary, the method for the design and optimization of copper alloy pipe casting and rolling process parameters established by the present invention applies neural network, genetic algorithm, finite element simulation, test design, CAD parametric design and database technology to process design and parameter In the process of optimization, on the one hand, the past experience is fully utilized, and on the other hand, intelligent technology and finite element technology are combined to achieve efficient process parameter optimization design. This overcomes the disadvantages of single design method in the past.

With the accumulation of experience, the improvement of finite element simulation technology and the strengthening of understanding of the various processes of copper tube processing, the effectiveness of the system will be further enhanced.

The invention combines neural network, finite element simulation, genetic algorithm, CAD technology and database technology to make up for the shortcomings of traditional design methods in the past. The finite element software is used to simulate the forming process, which can accurately reflect the actual production and processing process of the pipe, and guide and predict the development, development and processing of the product. Using the highly nonlinear fitting properties of the neural network to learn and train the finite element simulation input parameters, corresponding simulation results, and empirical data, the threshold and weight matrix obtained after training are used as implicit rules, which can map the process The relationship between parameters and process indicators. It overcomes the bottleneck problem of obtaining knowledge and the explosion of reasoning combinations in the past, and is easy to update. The global optimization feature of genetic algorithm is used to search for the optimal process parameters to achieve the desired process index, which solves the problem that the previous optimization methods are easy to fall into local extremum. The established copper pipe production database integrates production technical data and long-term accumulated technical experience to guide process design. According to the design results, combined with CAD software for parametric design, automatic generation of mold drawings. This system is used in the process design of copper alloy tube horizontal continuous casting, three-roller planetary rolling and moving core drawing production process, solves various practical problems in the process of copper tube processing, and formulates accurate and standardized processing technology.

Claims (5)

1. A method for the design and optimization of copper alloy pipe casting and rolling process parameters, characterized in that: the design basis is based on database, the neural network is the design method of process parameters and process indicators, the genetic algorithm is the process parameter optimization means, and the integrated neural network is integrated. Network, genetic algorithm, finite element simulation, experimental design, CAD parametric design and database technology are used in process design and parameter optimization to design and optimize the process parameters of copper alloy pipe casting and rolling; specifically include: 1) database Establish a database for horizontal continuous casting, three-roll planetary rolling and moving core drawing as the basis for parameter design and optimization. The database stores production data, standard data and temporary data; 2) Finite element simulation Using finite element numerical simulation steps, the horizontal continuous casting temperature field, temperature gradient and cooling rate values in horizontal continuous casting are respectively obtained, and the crack initiation tendency value is derived; the rolling force calculation value and rolling forming defects in three-roll planetary rolling Simulation results; and the calculation value of the drawing force in the drawing of the moving core and the simulation results of drawing forming defects; 3) neural network Through the multi-layer artificial neural network to learn the calculation results of finite element numerical simulation, and use the trained neural network to calculate the corresponding horizontal continuous casting process and three-roll planetary rolling of the process parameters of copper alloy pipe casting and rolling under different conditions The process index value in the process and the drawing process of the floating core; Through the training and learning of drawing and matching schemes of various specifications collected from the site, the trained neural network is used to obtain the matching mold design scheme of the required specifications; 4) Genetic algorithm According to the optimization objectives of continuous casting cooling system and continuous casting casting system parameters in horizontal continuous casting, rolling parameters in three-roll planetary rolling and drawing parameters in floating core head drawing, parameter encoding is carried out to form an initialization population; The process index value of each individual is calculated by the neural network, that is, the fitness value in the genetic algorithm, and then the operation operator is operated; the population evolves from generation to generation until the optimal solution is found, and the optimal process parameters are determined. The results are made into process cards and design files; 5) Graphic expression Using the CAD parametric design method, the best process parameters obtained are combined with CAD software to carry out the CAD parametric design of the rolling roll shape CAD in the three-roller planetary rolling of the mold, and the core head die CAD parametric design of the floating core head drawing, and the design calculation , data processing and graphics rendering for comprehensive processing. 2. The method for designing and optimizing process parameters of copper alloy pipe casting and rolling according to claim 1, characterized in that: said finite element numerical simulation adopts the method of uniform experimental design. 3. The method for designing and optimizing process parameters of copper alloy pipe casting and rolling according to claim 1, characterized in that: if there is a newly added finite element simulation result, normalize it, integrate it into the original training sample file, and re- After training, the newly obtained neural network threshold and weight matrix are stored in the database as a neural network matrix file. 4. The method for designing and optimizing process parameters of copper alloy pipe casting and rolling according to claim 1, characterized in that: the process parameters to be optimized include: casting system parameters and cooling system parameters of horizontal continuous casting; three-roll planetary rolling The roll deflection angle, roll inclination angle, opening degree and cart speed; the outer die cone angle, core head cone angle, drawing speed, outer die sizing section length and core head Calibration section length. 5. according to the method for the design and optimization of the copper alloy pipe casting and rolling process parameters of claim 1, it is characterized in that: the neural network is a BP network based on the Levenberg-Marquardt optimization algorithm, and the hidden layer is a Sigmoid activation function, and the output The layer uses Purelin activation function.

Images