CN106909727B - Laser welding temperature field finite element simulation method based on BP neural network and genetic algorithm GA - Google Patents

Laser welding temperature field finite element simulation method based on BP neural network and genetic algorithm GA Download PDF

Info

Publication number
CN106909727B
CN106909727B CN201710090680.4A CN201710090680A CN106909727B CN 106909727 B CN106909727 B CN 106909727B CN 201710090680 A CN201710090680 A CN 201710090680A CN 106909727 B CN106909727 B CN 106909727B
Authority
CN
China
Prior art keywords
heat source
finite element
neural network
genetic algorithm
element simulation
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.)
Active
Application number
CN201710090680.4A
Other languages
Chinese (zh)
Other versions
CN106909727A (en
Inventor
宋燕利
华林
徐勤超
余成
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.)
Wuhan University of Technology WUT
Original Assignee
Wuhan University of Technology WUT
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 Wuhan University of Technology WUT filed Critical Wuhan University of Technology WUT
Priority to CN201710090680.4A priority Critical patent/CN106909727B/en
Publication of CN106909727A publication Critical patent/CN106909727A/en
Application granted granted Critical
Publication of CN106909727B publication Critical patent/CN106909727B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

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
    • G06F30/23—Design optimisation, verification or simulation using finite element methods [FEM] or finite difference methods [FDM]
    • 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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/12—Computing arrangements based on biological models using genetic models
    • G06N3/126—Evolutionary algorithms, e.g. genetic algorithms or genetic programming
    • 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
    • Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
    • Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
    • Y04S10/00—Systems supporting electrical power generation, transmission or distribution
    • Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications

Landscapes

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

Abstract

The invention discloses a laser welding temperature field finite element simulation method based on a BP neural network and a genetic algorithm GA, which comprises the following steps: s1, laser welding finite element simulation: selecting a surface body combined heat source model consisting of a Gaussian surface heat source and a cylindrical heat source, selecting a surface heat source heat energy distribution coefficient, a surface heat source acting radius and an effective heat power coefficient as design variables, designing an orthogonal test table, and calculating finite element simulation errors of the surface body combined heat source model under different parameters; s2, establishing, training and testing a BP neural network; and S3, solving optimization parameters by the genetic algorithm, determining the feasibility of the optimization result, and if a large difference exists, reconstructing the BP neural network and carrying out optimization solution of the genetic algorithm. The invention can improve the efficiency and the precision in the traditional welding temperature field simulation and is easy to obtain the optimal simulation solution.

Description

Laser welding temperature field finite element simulation method based on BP neural network and genetic algorithm GA
Technical Field
The invention relates to a laser welding finite element simulation method, in particular to a laser welding temperature field finite element simulation method based on a BP neural network and a genetic algorithm GA.
Background
Laser welding has been widely used in joining processes by virtue of its advantages of high energy density, small heat affected zone, small deformation, strong welding flexibility, etc. In the automobile industry, the use of laser tailor-welded blanks is an important means for realizing the light weight of automobiles, and the laser tailor-welded blanks not only can meet the requirements of each part of an automobile body structure on materials, thickness, strength, corrosion resistance and the like, but also can improve the assembly precision of the automobile body, improve the rigidity, reduce the number of parts and improve the integration degree of the automobile body. With the emphasis on welding quality and welding production efficiency, finite element simulation is widely applied to the whole process of repeated welding by virtue of the advantages of low cost, high efficiency and the like. The simulation of the welding temperature field is a premise and a basis for reasonably selecting a welding method and process parameters and subsequent welding metallurgy analysis and stress analysis, so that the simulation of the welding temperature field has important significance.
The heat source model is a key factor influencing the simulation accuracy of the laser welding temperature field, and the heat energy absorption rate of the material to laser and the distribution condition of the absorbed heat energy on the material are problems to be solved before simulation. Because the heat source parameters and the simulation result have uncertain relation, in the welding temperature field simulation, trial calculation needs to be repeated according to the welding test result to correct the heat source related parameters, so as to achieve approximate simulation of the welding temperature field. However, this method is extremely inefficient and it is difficult to obtain optimal heat source parameter values.
Disclosure of Invention
In order to solve the technical problems of low efficiency and difficulty in obtaining an optimal solution of heat source parameter values in the traditional simulation method, the invention provides a laser welding temperature field finite element simulation method based on a BP neural network and a genetic algorithm GA (genetic algorithm), which can improve the efficiency in the traditional welding temperature field simulation and is easy to obtain the optimal solution.
The technical scheme adopted by the invention for solving the technical problems is as follows:
a laser welding temperature field finite element simulation method based on a BP neural network and a genetic algorithm GA is characterized by comprising the following steps:
s1 laser welding finite element simulation
S101, selecting a heat source model: selecting a Gaussian surface heat source and a cylinderSelecting heat energy distribution coefficient f of surface heat source1Acting radius r of surface heat sourcecAnd an effective thermal power coefficient η as a design variable;
s102, calculating finite element simulation errors of the lower surface body combined heat source model with different parameters: designing an orthogonal test table, wherein the design variables selected in the step S101 are factors of the orthogonal test table, selecting a plurality of levels in the value range of each design variable, determining the number of groups to be tested and specific parameters of each test group, implementing an orthogonal test scheme, establishing a corresponding surface body combined heat source model according to the parameters of each test group, selecting all surface body combined heat source models to carry out finite element simulation on a laser welding temperature field, if the boundary line of a molten pool simulated by a finite element is symmetrical to the boundary line of the molten pool measured by an experiment about the central line of a welding seam, the simulation is consistent with the experiment, selecting q points on the central line of the welding seam, and calculating the finite element simulation error delta of each test grouperr，
Figure BDA0001228838690000021
xiIs the transverse distance, y, from a point i on the center line of the weld to the boundary line of the finite element simulated weld pooliThe transverse distance from a point i on the central line of the welding seam to the boundary line of the molten pool measured in an experiment;
s2, establishing a BP neural network model, and training and testing the BP neural network: randomly selecting part of finite element simulation error deltaerrTraining BP neural network as training sample until the mean square error between the predicted output value and the actual value is limited in the allowable error range, and using the rest finite element to simulate errorerrDetecting the prediction precision of the neural network as a test sample until the mean square error of the network prediction output value and the actual value of the test sample is limited within an allowable error range to obtain a BP neural network with certain prediction capability;
s3, solving optimization parameters by using a genetic algorithm: group individual heat energy distribution coefficient f from surface heat source1Acting radius r of surface heat sourcecAnd an effective thermal power coefficient η, selecting a finite element simulation error deltaerrAsAnd (3) the fitness of the individuals is obtained by adopting a genetic algorithm to obtain a group of individuals with optimal effect, namely a group of parameter combinations with optimal effect, laser welding finite element simulation is carried out by using a heat source model under the optimal parameter combinations, the simulation results are compared with the experiment results, the feasibility of the optimization results is determined, and if the difference is large, the construction of the BP neural network and the optimization solution of the genetic algorithm are carried out again.
According to the technical scheme, in step S101, the heat flow density distribution rule formula of the plasma cloud on the sample in the surface body heat source model at the node (x, y) obtained by the Gaussian surface heat source function is as follows
Figure BDA0001228838690000022
The heat flux density formula of the keyhole effect at the node (x, y, z) which can be obtained by the heat source function of the cylinder is
Figure BDA0001228838690000023
In the formula (f)1Heat energy distribution coefficient for surface heat source, f2A coefficient of thermal energy distribution for a bulk heat source, and f1+f2＝1，rcRadius of effective action for surface heat source, rvThe effective acting radius of the body heat source, h the acting depth of the heat source, Q the heat flow output quantity of the laser, η the effective heat power coefficient of the sample absorbing laser, and (x, y, z) the node coordinate with the center of the surface heat source as the origin.
According to the technical scheme, in the step S102, 1/2 symmetric models are taken for calculation, the grid density is increased in a welding seam area, the lower grid density is selected when the welding seam area is far away from the welding seam area, convection heat transfer and heat radiation can occur when temperature difference exists between materials and surrounding air in the welding process, the total heat transfer coefficient H is used for representing heat energy loss, and the calculation formula is as follows
Figure BDA0001228838690000038
Wherein T is the real-time temperature of the material surface, TvThe ambient temperature is defined as a ═ 2.2, b ═ 0.25, and c ═ 4.6 × 10-8。
According to the above technical solution, the establishment of the BP neural network model in step S2 includes the following steps: will design variable f1、rcAnd η as input layers, ΔerrFormula for number of neurons in hidden layer as output layer
Figure BDA0001228838690000031
Wherein n is the number of input layer neurons, m is the number of output layer neurons, p is a constant, and 1<p<10；
The function tansig is chosen as the transfer function for the hidden layer, the linear function purelin as the transfer function for the output layer,
Figure BDA0001228838690000032
Figure BDA0001228838690000033
purelin(x)＝x。
according to the above technical scheme, in step S2, in order to eliminate the magnitude difference between the dimensional data and avoid overflow of numerical values caused by too large or too small weight, the normalization processing formula of the training and testing of the BP neural network is
Figure BDA0001228838690000034
In the formula, xkRepresenting data requiring normalization, xminDenotes the minimum value, x, in the datamaxRepresenting the maximum value in the data.
According to the technical scheme, in the step S3, the genetic algorithm comprises selection operation, and the probability p of the jth individual being selectedjIs calculated by the formula
Figure BDA0001228838690000035
In the formula, FjIs the fitness value of the individual, and N is the number of population individuals.
According to the above technical scheme, in step S3, the genetic algorithm comprises a crossover operation, in order to enable any position on the chromosome to be cross-recombined, a non-uniform arithmetic crossover operator is used, and the expression is
Figure BDA0001228838690000036
In the formula, mksAnd mlsRespectively represent chromosome mkAnd chromosome mlM is [0, 1] at the s (1, 2, 3) th position]A random number in between.
According to the above technical solution, in step S3, the genetic algorithm includes mutation operation, randomly selecting an individual, randomly changing the value of a position on a chromosome, and performing mutation operation using the following mutation formula
Figure BDA0001228838690000037
In the formula, mmaxIs mksUpper bound value of mminIs mksG is the algebra of the current evolution, GmaxFor maximum evolutionary number, r is [0, 1]]A random number in between.
The invention has the following beneficial effects: optimizing heat source model parameters in laser welding finite element simulation through a BP neural network and a genetic algorithm GA, selecting a surface body heat source model to perform finite element simulation on a laser welding temperature field, determining a representative heat source parameter combination in the laser welding simulation by using an orthogonal test analysis method and taking a heat source effective power coefficient, a heat energy distribution coefficient and a heat source acting radius which are difficult to determine in the simulation and have large influence on the result as input quantities, performing laser welding temperature field simulation to obtain finite element simulation errors of different test groups, training the BP neural network by taking the errors of finite element simulation results as output quantities, finally obtaining the neural network with high prediction precision of the finite element simulation errors, forming a parameter optimization method combining the neural network and the genetic algorithm, and optimizing the parameters of the heat source model in the laser welding simulation by adopting the genetic algorithm, and the error approximate calculation model established by the BP neural network is utilized to reduce a large amount of calculation caused by repeated numerical simulation in the optimization process so as to obtain the optimal heat source model parameters, thereby realizing the efficient and accurate simulation of the laser welding temperature field.
Drawings
The invention will be further described with reference to the accompanying drawings and examples, in which:
FIG. 1 is a graph comparing finite element simulation and experimental baths in an embodiment of the present invention.
FIG. 2 is a graph comparing the predicted output and the actual value of the training samples according to the embodiment of the present invention.
FIG. 3 is a graph comparing the predicted output and the actual value of the network of the test sample according to the embodiment of the present invention.
Fig. 4 is a fitness curve diagram of an optimal individual in an embodiment of the invention.
FIG. 5 is a comparison of finite element simulation and experimental molten pool after parameter optimization in an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
In a preferred embodiment of the present invention, a laser welding temperature field finite element simulation method based on a BP neural network and a genetic algorithm GA comprises the following steps:
s1 laser welding finite element simulation
S101, selecting a heat source model: selecting a surface body combined heat source model consisting of a Gaussian surface heat source and a cylindrical heat source, and selecting a heat energy distribution coefficient f of the surface heat source1Acting radius r of surface heat sourcecAnd an effective thermal power coefficient η as a design variable;
s102, calculating finite element simulation of lower surface body combined heat source models with different parametersError: designing an orthogonal test table, wherein the design variables selected in the step S101 are factors of the orthogonal test table, selecting a plurality of levels in the value range of each design variable, determining the number of groups to be tested and specific parameters of each test group, implementing an orthogonal test scheme, establishing a corresponding surface body combined heat source model according to the parameters of each test group, selecting all surface body combined heat source models to carry out finite element simulation on a laser welding temperature field, if the boundary line of a molten pool simulated by a finite element is symmetrical to the boundary line of the molten pool measured by an experiment about the central line of a welding seam, the simulation is consistent with the experiment, selecting q points on the central line of the welding seam, and calculating the finite element simulation error delta of each test grouperr，
Figure BDA0001228838690000051
xiIs the transverse distance, y, from a point i on the center line of the weld to the boundary line of the finite element simulated weld pooliThe transverse distance from a point i on the central line of the welding seam to the boundary line of the molten pool measured in an experiment;
s2, establishing a BP neural network model, and training and testing the BP neural network: randomly selecting part of finite element simulation error deltaerrTraining BP neural network as training sample until the mean square error between the predicted output value and the actual value is limited in the allowable error range, and using the rest finite element to simulate errorerrDetecting the prediction precision of the neural network as a test sample until the mean square error of the network prediction output value and the actual value of the test sample is limited within an allowable error range to obtain a BP neural network with certain prediction capability;
s3, solving optimization parameters by using a genetic algorithm: group individual heat energy distribution coefficient f from surface heat source1Acting radius r of surface heat sourcecAnd an effective thermal power coefficient η, selecting a finite element simulation error deltaerrUsing a genetic algorithm to obtain a group of individuals with optimal effect as the fitness of the individuals to obtain a group of parameter combinations with optimal effect, performing laser welding finite element simulation by using a heat source model under the optimal parameter combinations, comparing simulation results with experimental results to determine the feasibility of the optimization results, and if the feasibility exists, determining the feasibility of the optimization resultsAnd constructing the BP neural network and carrying out genetic algorithm optimization solving again under the condition of larger difference.
In the preferred embodiment of the present invention, in step S101, the heat flow density distribution rule of the plasma cloud on the sample in the surface body heat source model at the node (x, y) obtained by the gaussian surface heat source function is expressed as
Figure BDA0001228838690000052
The heat flux density formula of the keyhole effect at the node (x, y, z) which can be obtained by the heat source function of the cylinder is
Figure BDA0001228838690000053
In the formula (f)1Heat energy distribution coefficient for surface heat source, f2A coefficient of thermal energy distribution for a bulk heat source, and f1+f2＝1，rcRadius of effective action for surface heat source, rvThe effective acting radius of the body heat source, h the acting depth of the heat source, Q the heat flow output quantity of the laser, η the effective heat power coefficient of the sample absorbing laser, and (x, y, z) the node coordinate with the center of the surface heat source as the origin.
In the preferred embodiment of the present invention, in step S102, 1/2 symmetric model is taken for calculation, the grid density is increased in the weld zone, and the lower grid density is selected when the weld zone is far away from the weld zone, during the welding process, the material and the ambient air have temperature difference to generate convective heat transfer and thermal radiation, and a total heat transfer coefficient H is used to represent the heat energy loss, and the calculation formula is as follows
Figure BDA0001228838690000054
Wherein T is the real-time temperature of the material surface, TvThe ambient temperature is defined as a ═ 2.2, b ═ 0.25, and c ═ 4.6 × 10-8。
In a preferred embodiment of the present invention, the establishment of the BP neural network model in step S2 includes the following steps: will design variable f1、rcAnd η as input layers, ΔerrFormula for number of neurons in hidden layer as output layer
Figure BDA0001228838690000061
Wherein n is the number of input layer neurons, m is the number of output layer neurons, p is a constant, and 1<p<10；
The function tansig is chosen as the transfer function for the hidden layer, the linear function purelin as the transfer function for the output layer,
Figure BDA0001228838690000062
Figure BDA0001228838690000063
purelin(x)＝x。
in the preferred embodiment of the present invention, in step S2, in order to eliminate the difference in magnitude between the dimensional data and avoid overflow of the numerical value caused by too large or too small weight, the normalization processing formula of the training and testing of the BP neural network is as follows
Figure BDA0001228838690000064
In the formula, xkRepresenting data requiring normalization, xminDenotes the minimum value, x, in the datamaxRepresenting the maximum value in the data.
According to the technical scheme, in the step S3, the genetic algorithm comprises selection operation, and the probability p of the jth individual being selectedjIs calculated by the formula
Figure BDA0001228838690000065
In the formula, FjIs the fitness value of the individual, and N is the number of population individuals.
In a preferred embodiment of the present invention, the genetic algorithm comprises a crossover operation in step S3, in order to allow any position on the chromosome to be cross recombined, a non-uniform arithmetic crossover operator is used, expressed as
Figure BDA0001228838690000066
In the formula, mksAnd mlsRespectively represent chromosome mkAnd chromosome mlM is [0, 1] at the s (1, 2, 3) th position]A random number in between.
In the preferred embodiment of the present invention, the genetic algorithm comprises a mutation operation, randomly selecting an individual, and randomly changing the value of a position on the chromosome, and performing the mutation operation using the following mutation formula in step S3
Figure BDA0001228838690000067
In the formula, mmaxIs mksUpper bound value of mminIs mksG is the algebra of the current evolution, GmaxFor maximum evolutionary number, r is [0, 1]]A random number in between.
The invention is illustrated below. In this embodiment, based on BP and GA algorithms, finite element simulation is performed on laser welding by using ANSYS. In the embodiment, a DP600 high-strength steel laser tailor-welded blank is taken as a research object, different heat source parameter combinations are designed through orthogonal tests by utilizing the fitting capability of an artificial neural network between uncertainty relations to form a training sample of the neural network, the mapping relation between the heat source parameters and finite element simulation errors is established, a genetic algorithm with parallel random search and global optimization capability is used for solving the parameter optimization problem, and the accurate simulation of a laser welding temperature field is realized. Which comprises the following steps:
s1 laser welding finite element simulation
S101, selecting a heat source model: the adopted test material is DP600 high-strength steel plate with 50mm multiplied by 25mm multiplied by 1mm, the welding equipment is JKD5120 type continuous laser welding machine, the main welding process parameters are shown in table 1,
TABLE 1
Figure BDA0001228838690000071
Selecting a surface body combined heat source model consisting of a Gaussian surface heat source and a cylindrical heat source, wherein the heat flow density distribution rule formula of plasma cloud on a sample in the surface body heat source model at a node (x, y) obtained by a Gaussian surface heat source function is as follows
Figure BDA0001228838690000072
The heat flux density formula of the keyhole effect at the node (x, y, z) which can be obtained by the heat source function of the cylinder is
Figure BDA0001228838690000073
Wherein the effective acting radius r of the body heat sourcevThe diameter of the laser spot, the depth h of the heat source action as the thickness of the sample plate, and the heat energy distribution coefficient f of the surface heat source1Empirically determining the radius r of the surface heat source to be between 0.4 and 0.6cDetermining the width of a welding seam obtained by a welding test to be between 1.2mm and 1.6mm, and determining the effective thermal power coefficient η of the sample to laser absorption to be between 0.65 and 0.75;
s102, establishing a finite element model: 1/2 symmetric model is taken for calculation, the grid density is increased in the welding seam area, the lower grid density is selected far away from the welding seam area, convection heat transfer and heat radiation can occur when the temperature difference exists between the material and the surrounding air in the welding process, the heat energy loss is expressed by a total heat transfer coefficient, and the calculation formula is that
Figure BDA0001228838690000074
S103, calculating finite element simulation errors of the heat source models with different parameters: the heat energy distribution coefficient f of the surface heat source obtained in the foregoing1Acting radius r of surface heat sourcecAnd the effective thermal power coefficient η, each taking 3 values according to L18(33) Orthogonal table design 18 finite element simulations, where the factors and levels are shown in table 2,
TABLE 2
Figure BDA0001228838690000081
Laser welding finite element simulation is carried out on each group of orthogonal tests by using software ANSYS, the welding process time is 1s, the cooling time is 550s, the welding process uses 250 load steps, and the cooling process uses 110 load steps, wherein one group of tests is taken as an example (a surface heat source heat energy distribution coefficient is 0.6, a surface heat source acting radius is 1.6mm, and an effective heat power coefficient is 0.65), a calculation method of finite element simulation errors is explained, a comparison graph of a 1500 ℃ isothermal zone (namely a molten pool) simulated by the finite element simulation of the group of tests and an actually measured molten pool is shown in FIG. 1, if a boundary line of the molten pool simulated by the finite element and a boundary line of the molten pool measured by the tests are symmetrical about a weld centerline L, the simulation is matched with the test, 21 points are uniformly taken from a point to B point (namely q is 21), and are represented by i (i is 1, 2, 3, …, 21, a point is 1, and B point is 21) for quantitative evaluation of, and sequentially calculating the transverse distance x from each point to the boundary line of the finite element simulation molten pooli(i is 1, 2, 3, …, 21) and the lateral distance y from each point to the experimentally measured bath boundary linei(i 1, 2, 3, …, 21) by
Figure BDA0001228838690000082
Representing simulation errors, carrying out finite element simulation and error calculation on each group of heat source parameters to obtain samples for training and testing the BP neural network, as shown in Table 3,
TABLE 3
Figure BDA0001228838690000083
Figure BDA0001228838690000091
S2 establishment, training and testing of BP neural network
S201, establishing a BP neural network model
(1) Determining a network structure: empirical formula for number of neurons in hidden layer
Figure BDA0001228838690000092
n is 3, m is 1, the network precision and the calculation time are comprehensively considered, a 3-layer BP neural network structure is selected, and the number of neurons in the hidden layer is 6;
(2) selecting a transfer function: selecting a function tansig with an output range of [ -1,1] as a transfer function of the hidden layer, and selecting a linear function purelin as a transfer function of the output layer;
s202, training and testing of the BP neural network: randomly selecting 14 groups of data of an orthogonal experiment as training samples for training a neural network, using the remaining 4 groups of data as test samples for detecting the prediction accuracy of the neural network, wherein the training and test samples of the neural network are shown in Table 3;
after the BP neural network is trained by training samples, the network prediction output and actual value of the training samples are shown in figure 2, the mean square error MSE value of the prediction output and the actual value is 0.0793, the neural network is overfitting, the fact that the neural network has prediction capability is not proved by only comparing the MSE of the training samples, the prediction error of the network on the test samples needs to be tested, the network prediction output and actual value of the test samples are shown in figure 3, and the mean square error MSE value is 0.0081;
s3 solving parameter optimization problem by genetic algorithm
S301, population initialization: the individual codes in the population use real number codes, and each individual has a heat energy distribution coefficient f of a surface heat source1Acting radius r of surface heat sourcecAnd an effective thermal power coefficient η, i.e., a chromosome length of 3, corresponding to the chromosome length of { f }1，rcη denotes;
s302, fitness function: error Δ of finite element simulationerrAs fitness of the individual;
s303, obtaining a group of individuals with optimal effects according to the selection operation, the cross operation and the variation operation optimization solution of the genetic algorithm, and obtaining a group of parameter combinations with optimal effects;
selecting operation: probability p of being selected for jth individualjIs calculated by the formula
Figure BDA0001228838690000101
In the formula FiIs the fitness value of the individual, and N is the number of population individuals;
and (3) cross operation: in order to make any position on the chromosome can be cross-recombined, a non-uniform arithmetic cross operator is used, and the expression is
Figure BDA0001228838690000102
In the formula, mksAnd mlsRespectively represent chromosome mkAnd chromosome mlM is [0, 1] at the s (1, 2, 3) th position]A random number in between;
mutation operation: randomly selecting an individual, randomly changing the value of a certain position on the chromosome, and performing mutation operation by using the following mutation formula
Figure BDA0001228838690000103
In the formula, mmaxIs mksUpper bound value of mminIs mksG is the algebra of the current evolution, GmaxFor maximum evolutionary number, r is [0, 1]]A random number in between;
s304, optimizing results and analyzing: when the heat source parameters are optimized by using a genetic algorithm, the population size is 20, the evolution times is 100, the cross probability is 0.4, the variation probability is 0.2, and the heat energy distribution coefficient f of a surface heat source1Acting radius r of surface heat sourcecAnd the effective thermal power coefficient η are respectively in the value ranges of [0.4 and 0.6%]、[1.2，1.6]And [0.65, 0.75 ]]After 100 evolutions, the fitness of the optimal individualThe values (i.e., errors in finite element modeling herein) tend to stabilize as shown in fig. 4, resulting in the optimal heat source parameters as shown in table 4.
TABLE 4
Heat energy distribution coefficient f of surface heat source1 Radius of action of surface heat source rc Effective thermal power coefficient η
0.5677 1.3443 0.7017
And (3) performing laser welding finite element simulation by using the optimized heat source parameters, wherein the simulation error obtained by calculating the simulation result and the experiment result shown in figure 5 is 0.2462, and the result shows that the optimized heat source parameters can greatly improve the precision of the laser welding finite element simulation.
It will be understood that modifications and variations can be made by persons skilled in the art in light of the above teachings and all such modifications and variations are intended to be included within the scope of the invention as defined in the appended claims.

Claims (8)

1. A laser welding temperature field finite element simulation method based on a BP neural network and a genetic algorithm GA is characterized by comprising the following steps:
s1 laser welding finite element simulation
S101, selecting a heat source model: selecting a surface body combined heat source model consisting of a Gaussian surface heat source and a cylindrical heat source, and selecting a heat energy distribution coefficient f of the surface heat source1Acting radius r of surface heat sourcecAnd an effective thermal power coefficient η as a design variable;
s102, calculating finite element simulation errors of the lower surface body combined heat source model with different parameters: designing an orthogonal test table, wherein the design variables selected in the step S101 are factors of the orthogonal test table, selecting a plurality of levels in the value range of each design variable, determining the number of groups to be tested and specific parameters of each test group, implementing an orthogonal test scheme, establishing a corresponding surface body combined heat source model according to the parameters of each test group, selecting all surface body combined heat source models to carry out finite element simulation on a laser welding temperature field, if the boundary line of a molten pool simulated by a finite element is symmetrical to the boundary line of the molten pool measured by an experiment about the central line of a welding seam, the simulation is consistent with the experiment, selecting q points on the central line of the welding seam, and calculating the finite element simulation error delta of each test grouperr，
Figure FDA0002168149760000011
xiIs the transverse distance, y, from a point i on the center line of the weld to the boundary line of the finite element simulated weld pooliThe transverse distance from a point i on the central line of the welding seam to the boundary line of the molten pool measured in an experiment;
s2, establishing a BP neural network model, and training and testing the BP neural network: randomly selecting part of finite element simulation error deltaerrTraining BP neural network as training sample until the mean square error between the predicted output value and the actual value is limited in the allowable error range, and using the rest finite element to simulate errorerrDetecting the prediction precision of the neural network as a test sample until the mean square error of the network prediction output value and the actual value of the test sample is limited within an allowable error range to obtain a BP neural network with certain prediction capability;
s3, solving optimization parameters by using a genetic algorithm: group individual heat energy distribution coefficient f from surface heat source1Acting radius r of surface heat sourcecAnd an effective thermal power coefficient η, selecting a finite element simulation error deltaerrAs the fitness of the individual, a group of individuals with optimal effect is obtained by adopting a genetic algorithm, and a group of parameters with optimal effect is obtainedAnd combining, namely performing laser welding finite element simulation by using the heat source model under the optimal parameter combination, comparing a simulation result with an experimental result, determining the feasibility of an optimization result, and re-constructing the BP neural network and performing genetic algorithm optimization solution if a large difference exists.
2. The laser welding temperature field finite element simulation method based on the BP neural network and the genetic algorithm GA as claimed in claim 1, wherein in step S101, the distribution rule of plasma cloud on the upper surface of the sample in the surface body heat source model is expressed by Gaussian surface heat source function
Figure FDA0002168149760000021
The keyhole effect is expressed by the cylinder heat source function as
Figure FDA0002168149760000022
In the formula (f)1Heat energy distribution coefficient for surface heat source, f2A coefficient of thermal energy distribution for a bulk heat source, and f1+f2＝1，rcRadius of effective action for surface heat source, rvThe effective acting radius of the body heat source, h the acting depth of the heat source, Q the heat flow output quantity of the laser, η the effective heat power coefficient of the sample absorbing laser, and (x, y) the node coordinate with the center of the surface heat source as the origin.
3. The finite element simulation method of the laser welding temperature field based on the BP neural network and the genetic algorithm GA as claimed in claim 1, wherein in step S102, 1/2 symmetric models are taken for calculation, the grid density is increased in the weld zone, the lower grid density is selected when the weld zone is far away from the weld zone, during the welding process, the convection heat transfer and the heat radiation can occur when the temperature difference exists between the material and the ambient air, the total heat transfer coefficient H is used for representing the heat energy loss, and the calculation formula is that
Figure FDA0002168149760000023
Wherein T is the real-time temperature of the material surface, TvThe ambient temperature is defined as a ═ 2.2, b ═ 0.25, and c ═ 4.6 × 10-8。
4. The BP neural network and genetic algorithm GA based laser welding temperature field finite element simulation method according to claim 1, wherein the establishment of BP neural network model in step S2 comprises the following steps: will design variable f1、rcAnd η as input layers, ΔerrFormula for number of neurons in hidden layer as output layer
Figure FDA0002168149760000024
Wherein n is the number of input layer neurons, m is the number of output layer neurons, p is a constant, and 1<p<10；
The function tansig is chosen as the transfer function for the hidden layer, the linear function purelin as the transfer function for the output layer,
Figure FDA0002168149760000025
Figure FDA0002168149760000026
purelin(x)＝x。
5. the method of claim 1, wherein in step S2, in order to eliminate the difference in magnitude between the dimensional data and avoid overflow of numerical values due to too large or too small weight, the BP neural network is trained and tested using a normalization process formula
Figure FDA0002168149760000027
In the formula, xkRepresenting data requiring normalization, xminDenotes the minimum value, x, in the datamaxRepresenting the maximum value in the data.
6. The BP neural network and genetic algorithm GA-based laser welding temperature field finite element simulation method of claim 1, wherein in step S3, the genetic algorithm comprises a selection operation, and the probability p that the jth individual is selectedjIs calculated by the formula
Figure FDA0002168149760000031
In the formula, FjIs the fitness value of the individual, and N is the number of population individuals.
7. The method of claim 1, wherein the genetic algorithm comprises a crossover operation in step S3, and in order to make any position on the chromosome cross-regroupable, a non-uniform arithmetic crossover operator is used, and the expression is
Figure FDA0002168149760000032
In the formula, mksAnd mlsRespectively represent chromosome mkAnd chromosome mlM is [0, 1] at the s (1, 2, 3) th position]A random number in between.
8. The method of claim 1, wherein in step S3, the genetic algorithm comprises a mutation operation, randomly selecting an individual, randomly changing a value of a position on a chromosome, and performing the mutation operation using the following mutation formula
Figure FDA0002168149760000033
In the formula, mmaxIs mksUpper bound value of mminIs mksG is the algebra of the current evolution, GmaxFor maximum evolutionary number, r is [0, 1]]A random number in between.
CN201710090680.4A 2017-02-20 2017-02-20 Laser welding temperature field finite element simulation method based on BP neural network and genetic algorithm GA Active CN106909727B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201710090680.4A CN106909727B (en) 2017-02-20 2017-02-20 Laser welding temperature field finite element simulation method based on BP neural network and genetic algorithm GA

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201710090680.4A CN106909727B (en) 2017-02-20 2017-02-20 Laser welding temperature field finite element simulation method based on BP neural network and genetic algorithm GA

Publications (2)

Publication Number Publication Date
CN106909727A CN106909727A (en) 2017-06-30
CN106909727B true CN106909727B (en) 2020-05-05

Family

ID=59208400

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201710090680.4A Active CN106909727B (en) 2017-02-20 2017-02-20 Laser welding temperature field finite element simulation method based on BP neural network and genetic algorithm GA

Country Status (1)

Country Link
CN (1) CN106909727B (en)

Families Citing this family (18)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107582030A (en) * 2017-08-25 2018-01-16 天津大学 Human body surface skin temperature field measurement method based on BP neural network
CN107855687B (en) * 2017-10-17 2019-07-26 华中科技大学鄂州工业技术研究院 A kind of increasing material manufacturing fusion penetration on-line checking and control method and system
CN109242169B (en) * 2018-08-28 2021-11-02 贵州电网有限责任公司 Short-term load prediction method for large power customer
CN109508488B (en) * 2018-11-07 2022-08-02 西北工业大学 Shot peening forming process parameter prediction method based on genetic algorithm optimization BP neural network
CN109829200A (en) * 2018-12-29 2019-05-31 上海市特种设备监督检验技术研究院 A kind of laser welding temperature field heat source model
CN109871602A (en) * 2019-01-30 2019-06-11 西安工程大学 A kind of critical heat flux density prediction technique returned based on Gaussian process
CN110186570B (en) * 2019-05-16 2021-01-15 西安理工大学 Additive manufacturing laser 3D printing temperature gradient detection method
CN110490866B (en) * 2019-08-22 2020-05-15 四川大学 Metal additive forming size real-time prediction method based on depth feature fusion
CN111046609A (en) * 2019-12-24 2020-04-21 上海智殷自动化科技有限公司 BP neural network based welding process parameter optimization method
CN111461398A (en) * 2020-02-27 2020-07-28 季华实验室 Welding process parameter optimization method and device and readable storage medium
CN112560345B (en) * 2020-12-16 2022-12-06 中国电建集团河北省电力勘测设计研究院有限公司 Design method of underground electric power space ventilation system
CN112632721A (en) * 2020-12-21 2021-04-09 天津大学 MAG welding joint performance prediction method and system based on combined model
CN112906136B (en) * 2021-03-16 2022-07-22 上海交通大学 Method and system for predicting laser thermoforming deformation of hull plate
CN113516368A (en) * 2021-06-11 2021-10-19 北京大学 Method, device, equipment and medium for predicting uncertainty risk of city and community
CN113369753A (en) * 2021-07-01 2021-09-10 中铁磁浮科技（成都）有限公司 Welding parameter determination method based on finite element analysis and welding method
CN115161445B (en) * 2022-06-30 2024-02-27 武汉大学 Method for optimizing medium-frequency induction heating local postweld heat treatment parameters of 9% Cr hot-strength steel pipeline
CN115563549B (en) * 2022-10-27 2023-10-20 武汉理工大学 Welding defect cause diagnosis method and system and electronic equipment
CN115563736A (en) * 2022-10-28 2023-01-03 江南大学 Turbine blade arc additive real-time temperature field prediction method

Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105022900A (en) * 2015-08-19 2015-11-04 电子科技大学 Structure optimization method for heavy numerical control vertical lathe static pressure rotary table based on thermosetting coupling analysis

Patent Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105022900A (en) * 2015-08-19 2015-11-04 电子科技大学 Structure optimization method for heavy numerical control vertical lathe static pressure rotary table based on thermosetting coupling analysis

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
Numerical and experimental study of molten pool formation duringcontinuous laser welding of AZ91 magnesium alloy;Kamel Abderrazak et al.;《ELSEVIER》;20080714;全文 *
智能优化算法及其在焊接优化设计领域的应用;李广军等;《电焊机》;20110630;第41卷(第6期);全文 *
激光拼焊板焊缝建模与冲压成形基础研究;宋燕利;《中国博士学位论文全文数据库（电子期刊）工程科技Ⅰ辑 》;20121130;全文 *
膜式水冷壁多焊道焊接变形预测与焊接电源群控研究;赵章风;《中国博士学位论文全文数据库（电子期刊）工程科技Ⅱ辑 》;20120630;全文 *

Also Published As

Publication number Publication date
CN106909727A (en) 2017-06-30

Similar Documents

Publication Publication Date Title
CN106909727B (en) Laser welding temperature field finite element simulation method based on BP neural network and genetic algorithm GA
Fei et al. Springback prediction for incremental sheet forming based on FEM-PSONN technology
CN1292737A (en) Formation method and device for curved plates
CN1979496A (en) Copper-alloy pipe-material casting-milling technology parameter designing and optimizing method
Teimouri et al. Multi-responses optimization of ultrasonic machining process
Chen et al. A parallel strategy for predicting the quality of welded joints in automotive bodies based on machine learning
CN111651929A (en) Multi-objective optimization method based on fusion of Dynaform and intelligent algorithm
Jiang et al. The optimization of the variable binder force in U-shaped forming with uncertain friction coefficient
Singh Sivam et al. An investigation of hybrid models FEA coupled with AHP-ELECTRE, RSM-GA, and ANN-GA into the process parameter optimization of high-quality deep-drawn cylindrical copper cups
CN116882585A (en) Genetic algorithm and neural network coupled aluminum alloy hub low-pressure casting process optimization method
Wang et al. A study on prediction of process parameters of shot peen forming using artificial neural network optimized by genetic algorithm
CN106682773A (en) Human-machine welding time prediction method based on simulation annealing algorithm
Dhas et al. Weld residual stress prediction using artificial neural network and Fuzzy logic modeling
Omidvar et al. Selection of laser bending process parameters for maximal deformation angle through neural network and teaching–learning-based optimization algorithm
Ciancio et al. Design of a high performance predictive tool for forging operation
Zhan et al. The feasibility of intelligent welding procedure qualification system for Q345R SMAW
Vasudevan et al. Genetic algorithm for optimisation of A-TIG welding process for modified 9Cr–1Mo steel
Yan et al. Optimization of stamping process parameters based on orthogonal test and intelligent algorithm
Pholdee et al. Process optimization of a non-circular drawing sequence based on multi-surrogate assisted meta-heuristic algorithms
Park et al. Optimization of roll forming process with evolutionary algorithm for green product
Gao et al. A 6-sigma robust optimization method for stamping forming of automobile covering parts based on residual error and radial basis interpolation
Shandilya et al. Artificial neural network modeling and optimization using genetic algorithm of machining process
CN115130239A (en) Mechanical property prediction method for metal additive manufacturing based on multi-scale modeling
Miao et al. Surrogate model development using simulation data to predict weld residual stress: A case study based on the NeT-TG1 benchmark
Ma et al. Forecasting of the fatigue life of metal weld joints based on combined genetic neural network

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